Intelligent LED spectrum regulation and control system based on multi-dimensional data fusion

Abstract

The invention relates to the technical field of intelligent spectrum regulation and control, and discloses an intelligent spectrum regulation and control system based on multi-modal data fusion and dynamic optimization, which comprises a multi-modal data acquisition module, a real-time data processing module, a dynamic weight distribution algorithm, a deep reinforcement learning regulation and control algorithm and a distributed collaborative optimization module. According to the system, real-time fusion and dynamic priority adjustment of multi-dimensional data are realized through acquisition of multi-modal data (environment spectrum, temperature and humidity, carbon dioxide concentration and user behavior) and dynamic weight distribution based on LSTM (Long Short Term Memory); dynamic optimization of a spectrum regulation and control strategy is realized by combining a deep reinforcement learning model and utilizing a weighted reward function of a spectrum output error and energy consumption, and efficient light supplement and low energy consumption requirements are balanced; a distributed optimization algorithm with edge computing and cloud collaboration is adopted, computing resources are dynamically allocated, task processing efficiency is greatly improved, and system power consumption is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent spectrum control, and specifically to an intelligent LED spectrum control system based on multi-dimensional data fusion, which is used to achieve high-precision spectrum control in scenarios such as agricultural supplementary lighting, medical phototherapy, and intelligent lighting. Background Art

[0002] In current technology, spectral control systems usually rely on a single sensor to collect environmental data (such as temperature, humidity or spectral intensity) for static fill light control. For example, after obtaining environmental spectral data through a light sensor, a preset control rule is used to control the spectral output. However, this method has the following shortcomings: 1. The dynamic coupling relationship between the ambient spectrum and temperature and humidity, carbon dioxide concentration and user behavior has not been fully considered, resulting in a lack of accuracy and flexibility in the spectral output, which cannot adapt to complex and changing scene requirements.

[0003] 2. The fixed-weight control mechanism is difficult to adjust data priority according to real-time environmental changes, which can easily cause delayed control effects or waste of resources.

[0004] 3. The system usually adopts a single-node computing mode, which has response delays for tasks with high real-time requirements (such as agricultural lighting), and has high energy consumption during long-term operation.

[0005] To overcome the above problems, the existing technology also attempts to adopt dynamic weight adjustment methods or optimization solutions based on distributed computing. However, these methods, such as dynamic weight adjustment, usually only target a single data source or a simple model, and fail to achieve the fusion of multimodal data and real-time priority adjustment, resulting in inaccurate weight calculations between multi-source data; distributed computing solutions usually migrate all tasks to the cloud for processing, increasing communication delays and failing to meet the high real-time requirements of spectral regulation; there is a lack of systematic algorithm design for multi-objective optimization (such as spectral regulation accuracy and energy consumption control), resulting in difficulty in balancing resource allocation and energy consumption control.

[0006] Therefore, how to achieve efficient fusion of multimodal data, combine reinforcement learning with distributed collaborative optimization technology, dynamically adjust the spectral output weight, and balance system power consumption and control accuracy has become the technical problem to be solved by the present invention. Summary of the invention

[0007] The present invention provides an intelligent LED spectrum control system based on multi-dimensional data fusion, a system and a computer-readable storage medium, the main purpose of which is to solve the problems of insufficient real-time performance, low resource allocation efficiency and insufficient multi-modal data fusion.

[0008] To achieve the above object, the present invention provides an intelligent LED spectrum control system based on multi-dimensional data fusion, the system comprising: The multi-dimensional data acquisition module is used to collect environmental spectrum data, temperature and humidity data, carbon dioxide concentration data and user behavior data in real time, among which: environmental spectrum data is collected by high-precision spectrum sensors; temperature and humidity data are collected by digital sensors; carbon dioxide concentration data is measured by infrared gas sensors; user behavior data is analyzed by cameras combined with image recognition algorithms; The data fusion processing module is connected to the multidimensional data acquisition module and is used to perform standardized preprocessing and dynamic fusion on the collected multidimensional data, including: using a fuzzy logic-based standardization algorithm to perform dimensionless processing on the data of different physical quantities; using a multimodal adaptive weight allocation algorithm, according to the time series variation characteristics of environmental spectrum data, temperature and humidity data, carbon dioxide concentration data and user behavior data, the weight coefficient of the spectral output parameter is calculated, and the weight coefficient is calculated by the following formula:

[0009] in: For spectrum control The weight of the class data; For the collection of Real-time value of class data; It is a dynamic adjustment factor obtained based on historical data fitting, which is obtained by combining the time series prediction model LSTM with historical data training; is the dimension of the collected data; Indicates Spectral output coefficient at each time point; For the The value of the class data; The spectrum control algorithm module is connected to the data fusion processing module and is used to control the LED spectrum output in real time according to the data fusion result, including: using a spectrum optimization model based on deep reinforcement learning, the model input is the weight coefficient and the spectrum output target value after data fusion; in the model training process, the following objective function is used for dynamic optimization:

[0010] in: is the optimization goal of the policy network; Based on the current network parameters Spectral control strategy; For time The spectrum output effect score at the moment; is the reward discount factor, which controls the weight of the impact of future rewards on the current strategy; The LED spectrum output module is connected to the spectrum control algorithm module and is used to dynamically adjust the LED spectrum output according to the optimization result, including: realizing spectrum output through a multi-channel adjustable spectrum LED light source; configuring a fast-response current driver to dynamically adjust the current value of each band spectrum channel according to the target parameters output by the spectrum control algorithm module to achieve precise control of the spectrum.

[0011] The system collaborative control module is connected to the above modules and is used to coordinate the operation of the multi-dimensional data acquisition module, data fusion processing module, spectrum control algorithm module and LED spectrum output module, including: configuring an event-driven mechanism for dynamically triggering data acquisition and processing; dynamically allocating computing resources through a distributed collaborative optimization algorithm to reduce system power consumption and improve real-time performance.

[0012] Preferably, the fuzzy logic normalization algorithm in the data fusion processing module includes the following steps: normalizing the data of different physical quantities to a range between 0 and 1 respectively; based on the dynamic range of the data, using triangular membership functions and dynamically adjusted threshold intervals, constructing fuzzy sets to reduce the scale differences between physical quantities.

[0013] Preferably, in the multi-modal adaptive weight allocation algorithm, the dynamic adjustment factor The calculation includes: collecting historical data in the target application scenario; performing time series fitting analysis based on the LSTM model to predict the long-term weight trend of each type of data on the spectral output; using the gradient descent method to dynamically update the weight adjustment factor to optimize the real-time and robustness of the weight calculation.

[0014] Preferably, the training process of deep reinforcement learning in the spectral control algorithm module includes: generating an initial sample library of spectral control strategies based on multidimensional data collected offline; using a convolutional neural network to extract features of multidimensional data; constructing a reinforcement learning environment and defining the reward function as a weighted comprehensive score of the accuracy score of the spectral output and the energy consumption; and using distributed parallel computing to accelerate the strategy optimization of deep reinforcement learning.

[0015] Preferably, the multi-channel adjustable spectrum LED light source in the LED spectrum output module includes: LED light sources of four spectral bands: red, green, blue and white; the light source of each band is independently powered by a constant current driver; and the PWM (pulse width modulation) technology is used to adjust the luminous intensity and duration of the light source of each band in real time to achieve dynamic synthesis of the target spectrum.

[0016] Preferably, the event-driven mechanism in the system collaborative control module includes: detecting the real-time sampling frequency of the multidimensional data acquisition module; triggering the data fusion processing module to recalculate the weight coefficient when the spectral output deviation exceeds a preset threshold; and managing data transmission and control instructions between multiple modules through asynchronous queues to avoid communication blockage between modules.

[0017] Preferably, the calculation formula of the weight coefficient further combines the confidence parameter of the data , the confidence parameter is determined by the following formula:

[0018] in: is the sensitivity adjustment factor, determined by experimental fitting; For real-time collection The value of the class data; is the historical mean of this data category.

[0019] Preferably, the reward discount factor in the deep reinforcement learning model is The value of is adjusted dynamically according to the following formula:

[0020] in: is the adjustment factor, which depends on the sensitivity requirement of spectral regulation; For time Spectral output score at the moment; It is the target scoring threshold, which is used to measure whether the spectral regulation achieves the expected effect.

[0021] Compared with the problems described in the background technology, the beneficial effects of the present invention are: 1. Through the collection and preprocessing of multimodal data (environmental spectrum, temperature and humidity, carbon dioxide concentration, user behavior), combined with the LSTM-based dynamic weight allocation algorithm, real-time fusion of data input and dynamic priority adjustment are achieved. Compared with the traditional single or fixed weight spectral control method, this solution can quickly adapt the different influence weights of various data on spectral output in a dynamic and complex environment to form a real-time control strategy. Through the adaptive adjustment of spectral intensity, the system can maintain the accuracy of spectral output error below 5% in different application scenarios (such as agricultural supplementary lighting and medical phototherapy), while meeting the real-time response requirements under changing environmental conditions.

[0022] 2. By building a deep reinforcement learning model based on the Actor-Critic architecture and combining the weighted reward function of spectral output error and energy consumption, dynamic optimization of the spectral control strategy is achieved. By combining online reasoning with periodic training, the system can achieve a dynamic trade-off between efficient fill light and energy consumption control, avoiding the limitations of traditional static algorithms in different scenarios. In addition, the multi-objective optimization mechanism in the reward function ensures the performance stability and resource utilization efficiency of the system in long-term operation by balancing spectral accuracy and energy consumption. For example, in medical phototherapy scenarios with high precision requirements, the system can automatically adjust the spectral output intensity through a reinforcement learning model while reducing energy consumption by 15%.

[0023] 3. Through the resource allocation strategy of edge computing and cloud servers, this technical solution solves the problem of computing conflicts between real-time tasks and periodic tasks. The random forest regression model is used to predict the future task load distribution, and the greedy strategy is combined to realize dynamic sorting of task priorities, which greatly improves the efficiency of computing resource allocation. The actual operation results show that when the computing load of the edge node is close to the upper limit, the automatic migration of low-priority tasks shortens the response time of real-time tasks by 40% and reduces the overall power consumption of the system by 15%-20%. In addition, by regularly analyzing the operation data to optimize the task allocation ratio, the long-term adaptability and resource utilization efficiency of the system are further enhanced.

[0024] 4. A complete closed-loop process from data collection, processing, regulation to feedback has been realized, and a dynamic optimization logic combining real-time data with long-term data has been constructed. Through the combination of distributed collaborative optimization and deep reinforcement learning algorithms, the system realizes dynamic interaction of data input, model training, and resource allocation, solving the defects of isolated data processing and single optimization of regulation in existing technologies. The closed-loop logic in the technical path enhances the adaptability of multimodal data in complex environments, enabling the system to cope with multi-scenario switching needs, such as agriculture, medical care, industry and other different fields, and achieve an overall improvement in system performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is the architecture diagram of the intelligent spectrum control system of the present invention.

[0026] Figure 2 This is a flow chart of the intelligent spectrum control data processing of the present invention.

[0027] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0028] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0029] The embodiment of the present application provides an intelligent LED spectrum control system based on multi-dimensional data fusion. The system includes: a multi-dimensional data acquisition module for real-time acquisition of environmental spectrum data, temperature and humidity data, carbon dioxide concentration data and user behavior data, wherein: environmental spectrum data is acquired by a high-precision spectrum sensor; temperature and humidity data is acquired by a digital sensor; carbon dioxide concentration data is measured by an infrared gas sensor; user behavior data is analyzed by a camera combined with an image recognition algorithm; The data fusion processing module is connected to the multidimensional data acquisition module and is used to perform standardized preprocessing and dynamic fusion on the collected multidimensional data, including: using a fuzzy logic-based standardization algorithm to perform dimensionless processing on the data of different physical quantities; using a multimodal adaptive weight allocation algorithm, according to the time series variation characteristics of environmental spectrum data, temperature and humidity data, carbon dioxide concentration data and user behavior data, the weight coefficient of the spectral output parameter is calculated, and the weight coefficient is calculated by the following formula:

[0030] in: For spectrum control The weight of the class data; For the collection of Real-time value of class data; It is a dynamic adjustment factor obtained based on historical data fitting, which is obtained by combining the time series prediction model LSTM with historical data training; is the dimension of the collected data; Indicates Spectral output coefficient at each time point; For the The value of the class data; The spectrum control algorithm module is connected to the data fusion processing module and is used to control the LED spectrum output in real time according to the data fusion result, including: using a spectrum optimization model based on deep reinforcement learning, the model input is the weight coefficient and the spectrum output target value after data fusion; in the model training process, the following objective function is used for dynamic optimization:

[0031] in: is the optimization goal of the policy network; Based on the current network parameters Spectral control strategy; For time The spectral output effect score at the moment; is the reward discount factor, which controls the weight of the impact of future rewards on the current strategy; The LED spectrum output module is connected to the spectrum control algorithm module and is used to dynamically adjust the LED spectrum output according to the optimization result, including: realizing spectrum output through a multi-channel adjustable spectrum LED light source; configuring a fast-response current driver to dynamically adjust the current value of each band spectrum channel according to the target parameters output by the spectrum control algorithm module to achieve precise control of the spectrum.

[0032] The system collaborative control module is connected to the above modules and is used to coordinate the operation of the multi-dimensional data acquisition module, data fusion processing module, spectrum control algorithm module and LED spectrum output module, including: configuring an event-driven mechanism for dynamically triggering data acquisition and processing; dynamically allocating computing resources through a distributed collaborative optimization algorithm to reduce system power consumption and improve real-time performance.

[0033] Preferably, the fuzzy logic normalization algorithm in the data fusion processing module includes the following steps: normalizing the data of different physical quantities to a range between 0 and 1 respectively; based on the dynamic range of the data, using triangular membership functions and dynamically adjusted threshold intervals, constructing fuzzy sets to reduce the scale differences between physical quantities.

[0034] Preferably, in the multi-modal adaptive weight allocation algorithm, the dynamic adjustment factor The calculation includes: collecting historical data in the target application scenario; performing time series fitting analysis based on the LSTM model to predict the long-term weight trend of each type of data on the spectral output; using the gradient descent method to dynamically update the weight adjustment factor to optimize the real-time and robustness of the weight calculation.

[0035] Preferably, the training process of deep reinforcement learning in the spectral control algorithm module includes: generating an initial sample library of spectral control strategies based on multidimensional data collected offline; using a convolutional neural network to extract features of multidimensional data; constructing a reinforcement learning environment and defining the reward function as a weighted comprehensive score of the accuracy score of the spectral output and the energy consumption; and using distributed parallel computing to accelerate the strategy optimization of deep reinforcement learning.

[0036] Preferably, the multi-channel adjustable spectrum LED light source in the LED spectrum output module includes: LED light sources of four spectral bands: red, green, blue and white; the light source of each band is independently powered by a constant current driver; and the PWM (pulse width modulation) technology is used to adjust the luminous intensity and duration of the light source of each band in real time to achieve dynamic synthesis of the target spectrum.

[0037] Preferably, the event-driven mechanism in the system collaborative control module includes: detecting the real-time sampling frequency of the multidimensional data acquisition module; triggering the data fusion processing module to recalculate the weight coefficient when the spectral output deviation exceeds a preset threshold; and managing data transmission and control instructions between multiple modules through asynchronous queues to avoid communication blockage between modules.

[0038] Preferably, the calculation formula of the weight coefficient further combines the confidence parameter of the data , the confidence parameter is determined by the following formula:

[0039] in: is the sensitivity adjustment factor, determined by experimental fitting; For real-time collection The value of the class data; is the historical mean of this data category.

[0040] Preferably, the reward discount factor in the deep reinforcement learning model is The value of is adjusted dynamically according to the following formula:

[0041] in: is the adjustment factor, which depends on the sensitivity requirement of spectral regulation; For time Spectral output score at the moment; It is the target scoring threshold, which is used to measure whether the spectral regulation achieves the expected effect.

[0042] Preferably, the user behavior data analysis step of the multi-dimensional data acquisition module includes: acquiring user images in real time through a camera; extracting features from the image using a pre-trained convolutional neural network; judging the user's current behavior category based on a support vector machine model, and matching it with the spectral output demand library to generate spectral control target parameters.

[0043] Preferably, the distributed collaborative optimization algorithm in the system collaborative control module includes: dynamically allocating task computing resources of multiple modules; during the operation of the spectral regulation algorithm module, giving priority to allocating edge computing nodes to process real-time data, and uploading complex computing tasks to the cloud for completion; recording the operating status of each module through system logs, and optimizing resource allocation strategies in combination with long-term operating data.

[0044] Embodiment 1: The multimodal data acquisition module of this embodiment includes the following devices and data sources: environmental spectral data acquisition unit: equipped with a high-precision spectral sensor (wavelength range of 400-700nm, resolution of 1nm), which collects spectral curve data of the current light environment with a period of 1 second; temperature and humidity data acquisition unit: using a digital temperature and humidity sensor (sampling accuracy of ±0.1°C and ±1%RH), it collects current environmental temperature and humidity data once per second; carbon dioxide concentration data acquisition unit: equipped with an infrared gas sensor with a monitoring range of 400-2000ppm and a resolution of ±5ppm, which collects changes in CO2 concentration in the environment in real time; user behavior data acquisition unit: obtains real-time user behavior video through a camera, and classifies user behavior into "work", "rest" or "sleep" in combination with an image classification model based on a ResNet-50 convolutional neural network.

[0045] After data collection is completed, all data enter the preprocessing process to achieve standardization and normalization: Spectral data normalization: light intensity data collected by the spectral sensor Perform maximum value normalization as follows:

[0046] in, is the collected raw light intensity data, is the maximum light intensity value during the acquisition period.

[0047] Temperature, humidity and CO2 data standardization: using z-score standardization algorithm:

[0048] in: The original data collected; is the historical average value of the data; is the historical standard deviation of the data.

[0049] The confidence of the user behavior classification result is directly taken as the input parameter by the classification probability output by the ResNet-50 network. For example, the confidence of classification as “work” is 0.85.

[0050] After data preprocessing is completed, the dynamic weight allocation algorithm is used to calculate the influence weight of each data in spectral regulation. Its core steps are as follows: input data: standardized spectral data, temperature and humidity data, CO2 concentration data and user behavior confidence value; dynamic weight factor Generation: Use LSTM (Long Short-Term Memory Network) to perform time series analysis on historical data and predict the importance weight of current data:

[0051] in: For the Class data at historical time points The value of is the prediction window length (taking historical data within 30 minutes as input).

[0052] Dynamically adjusted weights Calculated using the following formula:

[0053] in: For the Real-time weight of class data; After real-time standardization Class data.

[0054] Dynamic weight adjustment mechanism: If the change in a certain type of data (such as spectrum or CO2 concentration) in real-time data exceeds the 10% threshold, the weight adjustment factor Real-time updates will be performed using the gradient descent method to ensure the sensitivity of the spectral control response.

[0055] The calculated weight It is used to fuse all collected data and generate the input target value of the spectral control algorithm. The data fusion process is as follows: The spectral target value is calculated according to the following formula:

[0056] in: is the target value of the spectral output; For the Input features of class data.

[0057] When the weight of spectral data is higher (such as spectral accuracy is prioritized in agricultural scenarios), the system prioritizes meeting spectral control needs; when the user behavior confidence is higher (such as user status is prioritized in medical phototherapy scenarios), the system dynamically increases the impact weight of user behavior data.

[0058] Assume that the system is applied to agricultural lighting scenarios: the spectral sensor detects that the current ambient light is insufficient (the light intensity in the 400-500nm band is less than 30%); the temperature and humidity sensor detects that the humidity is too high (85%RH); the CO2 concentration is 1000ppm; and the user behavior is classified as "rest".

[0059] Through dynamic weight assignment, the weight of spectral data The value is assigned to 0.6, the weight of temperature and humidity data The weight of CO2 data is 0.2. is 0.15, the weight of user behavior data The target value of the spectrum generated after fusion is 0.05. for:

[0060] The system adjusts the LED spectral output according to the target value to achieve dynamic fill light and environmental adaptation.

[0061] Embodiment 2: This embodiment is based on a deep reinforcement learning algorithm and realizes intelligent spectrum control by dynamically controlling the multi-channel LED spectrum output. The hardware system includes the following components: data acquisition module: acquires multi-modal data such as environmental spectrum, temperature and humidity, and carbon dioxide concentration; computing processing unit: equipped with embedded chips (such as NVIDIA Jetson AGX Orin), supporting real-time reasoning and optimization calculations of deep learning models; spectrum output module: equipped with multi-channel adjustable spectrum LED light source, each channel is independently configured with a constant current driver, and supports PWM modulation; communication module: supports data transmission between edge nodes and the cloud for model training and task allocation optimization.

[0062] The definition of state, action and reward function is as follows: ): The state space includes multimodal data of the environment, such as spectral intensity, temperature and humidity, carbon dioxide concentration, and user behavior classification results. The specific definitions are as follows:

[0063] Among them, all variables are normalized real-time data.

[0064] Action Space( ): The action space is defined as the PWM duty cycle adjustment value of each spectral channel LED:

[0065] Each PWM duty cycle The value range is [0, 1], which represents the luminous intensity of the LED in the corresponding band.

[0066] Reward function ( ): The reward function is determined by the comprehensive score of spectral output error and energy consumption, as follows:

[0067] in: is the actual spectrum output value; is the target spectral output value (calculated by the weight allocation algorithm in Example 1); The system power consumption (measured in real time by the current sensor of the spectrum output module); It is the factor affecting power consumption and adjusts the trade-off between spectral accuracy and energy consumption.

[0068] Also, in the training process of the reinforcement learning model, such as data set construction: under laboratory conditions, collect paired samples of different environmental variables (such as spectral intensity, temperature and humidity, CO2 concentration, etc.) and corresponding spectral control target values ​​to generate offline training data sets. The amount of data is no less than 10,000 groups, covering typical scenarios such as agricultural supplementary lighting, medical phototherapy and intelligent lighting.

[0069] The network structure design can reinforce the learning model based on the Actor-Critic architecture, which consists of the following two parts: Actor network: input state space , output action space The probability distribution of is: in Parameters of the Actor network. Critic network: input state and actions , output the value estimate of the current strategy : in are the parameters of the Critic network.

[0070] Reward function optimization: Use the Proximal Policy Optimization (PPO) algorithm to train the model and update the parameters of the Actor and Critic networks. The objective function is: in: is the reward discount factor (value range [0, 1], used to balance short-term and long-term benefits); For time The instant reward at the moment. The optimization process of PPO uses the gradient descent method to limit the step size of the strategy update to ensure convergence stability.

[0071] Training configuration: learning rate: 0.0003; Batch size: 64; number of training iterations: 2000 rounds; use a distributed training framework (such as PyTorch Distributed Data Parallel) to improve training efficiency.

[0072] In terms of online reasoning and deployment of the model, its real-time reasoning process is: multi-modal data is collected every second and standardized, and then input into the reinforcement learning model; the Actor network responds to the input state Generate Spectrum Output Action probability distribution; select the action with the maximum probability as the execution result, and adjust the PWM duty cycle of the spectrum output module in real time.

[0073] In addition, in terms of collaborative deployment between the edge and the cloud: model inference is deployed on edge nodes to achieve real-time response; the cloud regularly updates model parameters and sends them to edge devices via OTA (Over-the-Air) to ensure long-term adaptability of the model.

[0074] Assuming that the system is applied to medical phototherapy scenarios, the following are the actual application steps: the ambient spectrum detects that the blue light (450nm) intensity is insufficient, the temperature and humidity are maintained at 25°C and 40%RH, and the user behavior is classified as "sleep"; according to the dynamic weight allocation algorithm in Example 1, the target spectrum output is generated , it is necessary to enhance the blue light band output; the reinforcement learning model outputs the control strategy in real time, increases the PWM duty cycle of the blue light channel to 0.85, and reduces the output of other bands; after the spectrum control is completed, the actual spectrum output With target value The error is less than 5%, meeting the accuracy requirements of medical phototherapy.

[0075] Embodiment 3: Distributed collaborative optimization algorithms are used to efficiently allocate computing resources between edge computing nodes and cloud servers, reduce overall system power consumption and improve real-time performance. This system consists of the following hardware: edge computing nodes: equipped with embedded computing modules (such as NVIDIA Jetson Xavier NX), supporting real-time data processing and spectral regulation, and providing real-time computing support for spectral output modules and multimodal data acquisition modules; cloud servers: equipped with high-performance GPU computing units (such as NVIDIAA100) and distributed databases for long-term data storage, model training and strategy optimization; communication modules: edge nodes and cloud servers are connected via 5G communication modules or Wi-Fi 6, supporting high-bandwidth and low-latency data transmission.

[0076] Dynamic decision rules for task allocation: The system classifies real-time tasks (such as multimodal data fusion, spectral control reasoning) and periodic tasks (such as model training, long-term strategy optimization), such as real-time tasks: priority execution on edge nodes to ensure low-latency response; periodic tasks: migration to cloud servers for execution to reduce edge node computing load; the length of edge node task queues For dynamic decision-making task migration: in It is the task queue length threshold, which is adjusted dynamically according to the historical operation data.

[0077] Long-term resource allocation optimization strategy: The system records the operating data of edge nodes and cloud servers (such as task completion time, resource occupancy, power consumption, etc.) in a 10-minute cycle, and optimizes the resource allocation strategy through the following steps: Data collection: record task running time , calculate the load and power consumption .

[0078] Data modeling: Based on the random forest regression model, the task load distribution in the next hour is predicted and optimization suggestions are generated.

[0079] Strategy update: Genetic algorithm is used to optimize the edge-cloud task allocation ratio, and the objective function is to minimize the system's task completion time and power consumption: in and It is the weight factor of time and power consumption, which is set dynamically based on the application scenario.

[0080] In terms of real-time task scheduling, the scheduling trigger mechanism: the system sets the trigger conditions for task scheduling, such as when the CPU / GPU utilization of the edge node exceeds 80% or the length of the task queue exceeds 5 items, the task migration is triggered; Task migration process: Task classification: Check the task priority and give priority to retaining tasks with high real-time requirements (such as spectral control calculations) on the edge node; Migration mechanism: Send low-priority tasks (such as data storage and mid-term analysis) to the cloud server for processing through the communication module; Feedback mechanism: After the cloud processing is completed, the results are transmitted to the edge node in real time through the 5G communication module for the execution of subsequent control instructions; Task scheduling algorithm: The scheduling algorithm adopts a task allocation logic based on a greedy strategy: Sort by priority from high to low and allocate computing resources one by one.

[0081] In terms of system energy consumption optimization, such as edge node power consumption control: edge nodes automatically switch to low power consumption mode when the task is idle (such as reducing the processor main frequency to 50%), and use dynamic voltage and frequency scaling technology (DVFS) to dynamically adjust the processor voltage and frequency according to the task load; cloud server resource utilization is improved: cloud servers use batch task processing technology (such as data batch merging processing) to reduce task allocation frequency and reduce invalid computing energy consumption.

[0082] Assuming that the system is applied to agricultural lighting supplement scenarios, the following is the actual application process of the system's distributed collaborative optimization, such as scenario input: the edge node collects multimodal data (spectral intensity, temperature and humidity, CO2 concentration, etc.) every second, and completes real-time spectral control reasoning; data storage tasks and model retraining tasks are migrated to the cloud server.

[0083] Real-time task allocation: When the CPU utilization of the edge node reaches 85%, the task scheduling module triggers the migration of low-priority tasks; data storage tasks are transmitted to the cloud server for processing through the 5G communication module.

[0084] It can also perform long-term optimization. For example, the system analyzes operating data (such as task completion time and power consumption) every 10 minutes and predicts the task load distribution in the next hour. The optimized strategy shows that in the current scenario, the cloud should bear 70% of the computing load and the edge nodes should bear 30%.

[0085] After distributed collaborative optimization, the task delay was reduced from 100ms to 60ms, and the overall system power consumption was reduced by 15%, meeting the dual requirements of agricultural lighting for real-time spectral regulation and energy consumption.

[0086] Embodiment 4: In modern agricultural greenhouse cultivation, plant growth is highly dependent on light intensity and spectral quality. Traditional lighting systems mostly use a fixed ratio of red light and blue light for lighting, ignoring the differentiated spectral requirements of different crops, different growth stages and environmental conditions, resulting in unsatisfactory lighting effects, high energy consumption and serious waste of resources.

[0087] For example, a smart agricultural greenhouse planting base grows a variety of crops, including tomatoes, cucumbers and strawberries. In order to improve the yield and quality of crops, users need to supplement specific spectral bands during the day and suppress excessive respiration of plants at night, while reducing the energy consumption of supplementary lighting. The system needs to adapt the spectral output in real time and dynamically adjust the spectral composition according to environmental data (light intensity, temperature and humidity, CO2 concentration) and the current growth stage of the crop to achieve the dual goals of precise supplementary lighting and energy saving.

[0088] The intelligent spectral control system can be deployed inside the greenhouse, including the following modules, such as the multimodal data acquisition module: the environmental sensors installed in the greenhouse collect light intensity (350-750nm), temperature (range: 15-35°C), humidity (range: 30-80%) and carbon dioxide concentration (range: 400-2000ppm) data in real time; the camera module collects real-time images of crop leaf color and leaf morphology for analyzing the health of crops; Spectral control module: equipped with 12-channel adjustable LED light source, including red light (660nm), blue light (450nm), green light (520nm), far red light (730nm) and other multi-bands, supporting fine adjustment of PWM duty cycle; computing and control module: the system uses edge computing nodes (such as NVIDIA Jetson Nano) to realize data fusion and spectral control reasoning, and cloud servers (supporting GPU acceleration) are used for long-term strategy optimization and model update.

[0089] The multimodal data acquisition module collects environmental data at a frequency of 1 second, and after normalization, it is input into the LSTM-based dynamic weight allocation algorithm. The algorithm calculates the real-time priority of different spectral bands based on the dynamic changes of ambient light intensity, carbon dioxide concentration, temperature and humidity. Combined with the spectral requirements of the crop growth stage (such as the demand for blue light is higher than red light in the tomato seedling stage) and the user-set fill light strategy, the target spectral output value StargetS_{\text{target}}Starget is generated. The spectral control module dynamically adjusts the LED duty cycle of each band according to the calculation results. For example, when the ambient light intensity is insufficient and the carbon dioxide concentration is low, the system automatically increases the output of the red and blue light bands and reduces the energy consumption of the green light band.

[0090] At the same time, in order to improve real-time performance and reduce energy consumption, the system adopts a distributed optimization mechanism of edge-cloud collaboration, such as edge nodes: realize real-time data fusion and spectral control reasoning, and give priority to real-time tasks that are sensitive to delays; cloud servers: analyze the environmental data and crop growth status of the past 24 hours every day, and optimize future spectral control strategies; use random forest regression models to predict future loads, generate task allocation ratios, and ensure load balance on edge nodes; task migration and feedback: when the task queue of an edge node exceeds the load threshold, the system migrates low-priority tasks (such as historical data analysis) to the cloud for execution; after the task is completed, the cloud feeds back the optimization results and updates the inference model of the edge device.

[0091] Taking tomato planting as an example, the system's lighting performance in the seedling stage (the red light / blue light demand ratio is about 1:2) and the flowering stage (the red light / blue light demand ratio is about 2:1) is as follows: Spectral accuracy: The system ensures that the target spectrum output error is less than 3% through multimodal data fusion and dynamic regulation, significantly improving the accuracy of lighting; Energy consumption control: Through distributed collaborative optimization, edge nodes process delay-sensitive tasks in real time, and the cloud completes low-priority computing tasks, reducing overall power consumption by 15%-20%; Crop yield and quality improvement: After the system has been running for 3 months, the yield of a single tomato plant has increased by 15%, the fruit color has become brighter, and the quality has been significantly improved.

[0092] This system achieves high precision and high responsiveness in spectral regulation through the combination of dynamic weight allocation algorithm and reinforcement learning model; the distributed optimization mechanism balances real-time and resource efficiency, overcoming the energy consumption and delay problems of single-node computing mode; actual application in agricultural lighting scenarios has proved that the system has significant technical advantages in energy saving, increased production and improved quality.

[0093] Embodiment 5: In the treatment of skin diseases, phototherapy is widely used to treat diseases such as psoriasis, eczema and melasma, and its efficacy depends on the accuracy of spectral output and the stability of intensity. However, traditional phototherapy equipment mostly uses fixed spectral output or simple manual adjustment mode, which makes it difficult to dynamically adjust the spectral output according to the patient's skin type, lesion location and treatment response, which can easily lead to over-irradiation or insufficient efficacy, and even increase the risk of side effects.

[0094] System implementation methods, such as hardware and data configuration, use multimodal data acquisition modules: the ambient light sensor installed on the phototherapy device collects indoor light intensity (range: 100-1000 lux), humidity (range: 30-60%) and the temperature of the patient's skin surface (range: 32-37°C) in real time. The camera module obtains real-time image data of the patient's skin lesions (resolution 1080p) to evaluate the size, color and treatment response of the skin lesions; Spectral control module: The phototherapy device is equipped with a narrow-band UVB light source (311nm), a red light source (630nm) and a blue light source (450nm), with a light intensity output range of 0-500mW / cm², supporting real-time dynamic adjustment; Calculation and control module: The system uses an embedded processor (such as NXP i.MX8) to process data fusion and real-time spectral control, and connects to a cloud server (supporting AI model updates) to optimize long-term treatment plans.

[0095] In the dynamic regulation of spectral output, patient data fusion can be performed: the system collects real-time data of the patient's skin at a frequency of 5 seconds, including temperature, lesion area change (unit: cm²) and skin color RGB value, combined with ambient light intensity and humidity information, and calculates the output priority of each spectral band through a dynamic weight allocation algorithm based on LSTM; spectral regulation strategy: through the target output value The system dynamically calculates the photosensitivity of the patient's lesions and the current treatment response to generate a spectral control strategy. For example, when the lesion area is reduced to 50% of the initial value and the skin temperature rises by more than 0.5°C, the system automatically reduces the UVB light intensity output and increases the proportion of red light to promote healing; Real-time feedback and adjustment: The system collects lesion images through a camera, analyzes the treatment effect, and adjusts the spectral output plan in real time when it finds increased pigmentation or subsided inflammatory response.

[0096] In order to verify the superiority of the technical solution of the present invention, the comparative experiment is designed as follows: Experimental subjects: 100 psoriasis patients were randomly selected and divided into two groups, each with 50 people. The experimental group was treated with the intelligent spectrum control system of the present invention, and the control group was treated with traditional phototherapy equipment (fixed spectrum output, UVB ratio 80%, red light ratio 20%).

[0097] Treatment parameters: The experimental group dynamically adjusted the spectrum ratio, UVB range was 50%-80%, and red light range was 20%-50%. The specific parameters were automatically adjusted according to the patient's real-time data. The control group fixed the spectrum ratio, and UVB and red light output did not change with the treatment response; Treatment cycle: Both groups were treated for 8 weeks, 3 times a week, and each treatment lasted 10 minutes.

[0098] The lesion area of ​​the experimental group was reduced by an average of 72%, while that of the control group was 58%. The experimental group showed obvious improvement in lesions at the 4th week, while the improvement trend of the control group was not apparent until the 6th week. The incidence of photosensitivity reactions (such as erythema and burning sensation) in the experimental group was 8%, significantly lower than 20% in the control group. This shows that the dynamic control strategy of the present invention effectively reduces the side effects caused by excessive irradiation. The recovery time of the experimental group after each treatment was shortened by an average of 15%, and the operation interruption caused by the adjustment of equipment parameters during the treatment process was significantly reduced; and the experimental group achieved precise control of real-time spectral output through multimodal data fusion and dynamic weight allocation algorithm. The spectral error was always kept within 3%, which significantly improved the phototherapy effect compared with the error of the fixed output of the control group (about 15%); and the distributed optimization mechanism realized the coordinated optimization of edge computing and cloud computing, avoiding unnecessary excessive spectral output, and reducing the overall energy consumption of the equipment by about 18%. In terms of patient experience optimization, the dynamic control strategy combined with the real-time feedback of patients not only improved the treatment effect, but also effectively controlled the side effects, which was conducive to providing patients with a safer and more comfortable treatment experience.

[0099] Embodiment 6: This embodiment adds a dynamic adjustment mechanism in the data fusion processing module, which specifically includes the following steps, namely, the improvement of data preprocessing: performing multi-stage normalization and standardization processing on the collected multimodal data, and the specific method is as follows: Spectral data normalization: collect the original spectral data Normalize according to the following formula: in, is the maximum light intensity value in the current acquisition cycle. The function of this formula is to eliminate the scale difference of spectral data under different intensity conditions and improve the stability of subsequent processing.

[0100] Standardization of temperature, humidity and carbon dioxide concentration data: z-score standardization is used, and the specific formula is:

[0101] in, To collect real-time data, and are the historical mean and standard deviation of this type of data respectively. This method can dynamically adapt to the data distribution characteristics under different environmental conditions.

[0102] Also, for weight coefficient calculation, the parameter definition and implementation path in the formula have been improved, such as the dynamic adjustment factor Generated by the following steps: Historical data analysis: Based on the LSTM (Long Short-Term Memory Network) model, the historical change trend of each type of data is fitted into a time series to generate a preliminary weight factor prediction value; Real-time adjustment mechanism: When the real-time data is updated, the gradient descent method is used to adjust the Perform dynamic optimization to ensure that it can quickly respond to data changes. The optimization formula is as follows:

[0103] in, is the learning rate, The loss function for weighted optimization is defined as the weighted sum of squares of the spectral output errors.

[0104] Weight coefficient The final calculation formula is as follows:

[0105] in, is the confidence parameter of the data, and its calculation formula is:

[0106] parameter The sensitivity adjustment factor representing the confidence level is determined by the experimental fit; is the historical mean. The design of this formula can balance the importance of real-time data and the weight influence of historical data; and in the optimization of the spectral control algorithm module, such as the improvement of the objective function, in the deep reinforcement learning model, the optimization objective function of the spectral control strategy is adjusted, and the energy consumption balance factor is added to optimize the balance between the accuracy of the spectral output and the energy consumption. The optimized objective function is:

[0107] in, is the energy consumption weight adjustment factor, For the moment The system power consumption is measured by the real-time current sensor. , which can adapt to the energy consumption and accuracy requirements of different scenarios.

[0108] In addition, online training is combined with offline training. For example, the reinforcement learning model uses a combination of online reasoning and periodic offline training to improve the adaptability of the model. Offline training is performed through pre-sampled multidimensional data sets, with a data volume of no less than 20,000 groups, covering agricultural lighting, medical phototherapy and intelligent lighting scenarios. Online reasoning is based on real-time input status and dynamically outputs spectral control actions.

[0109] The experimental verification process is as follows: data source: collecting multimodal data in agricultural greenhouses, including light intensity (350-750nm), temperature and humidity (range: 10-50℃, 30%-90%RH) and carbon dioxide concentration (range: 400-2000ppm); verification indicators: evaluating the technical effect of this embodiment by comparing the accuracy of spectral output and system energy consumption before and after dynamic regulation; experimental results: spectral output error is reduced from 5% to 2.5%, and accuracy is improved by 50%; the overall power consumption of the system is reduced by 18%, and the real-time response time is shortened by 30%.

[0110] In terms of hardware configuration, such as data acquisition module: high-precision spectral sensor, digital temperature and humidity sensor, and infrared gas sensor; computing unit: embedded AI chip (NVIDIA Jetson Xavier NX); spectral output module: 12-channel adjustable spectrum LED light source, supporting PWM modulation; and in terms of software, PyTorch can be used to implement reinforcement learning models, and deployed through edge computing nodes, combined with the cloud to regularly optimize model parameters.

[0111] Embodiment 7: Based on the original technology, this embodiment designs a dynamic feedback control mechanism based on real-time fusion of multi-dimensional data to improve the accuracy and response efficiency of the spectrum control system. This solution is suitable for smart agricultural supplementary lighting, medical phototherapy and smart lighting scenarios.

[0112] Step 1: Multimodal data collection and standardization. For example, in terms of the data collection module composition, it includes: Environmental spectral data: A high-precision spectral sensor with a wavelength range of 400-700nm and a resolution of 1nm is used to collect spectral intensity curves with a period of 1 second. Temperature and humidity data: A digital temperature and humidity sensor is used with an accuracy of ±0.1°C and ±1%RH, respectively, and a collection period of 1 second. CO2 concentration data: An infrared gas sensor is used with a monitoring range of 400-2000ppm and a resolution of ±5ppm. User behavior data: Real-time video is collected using a camera and classified (work, rest, sleep) using a ResNet-50 convolutional neural network. The classification confidence is directly represented by the network output probability.

[0113] In terms of data standardization processing, the spectral data normalization formula is: in, For the original collected light intensity data, The maximum light intensity value during the acquisition period; Temperature, humidity and CO2 data standardization: using the z-score standardization algorithm: in, is the real-time value collected. is the historical average, is the historical standard deviation.

[0114] Step 2: Dynamic weight allocation optimization, that is, to achieve dynamic priority adjustment of multimodal data, a dynamic feedback mechanism is designed to clarify the logic and formula source of weight allocation, for example: Generation of dynamic adjustment factors: Dynamic adjustment factors Based on LSTM model generation, its core process is as follows: Input data: collected real-time data sequence ; Time series prediction: fit historical data through LSTM and predict weight adjustment trend; Optimization formula: Perform real-time optimization and update using gradient descent: in, is the learning rate, To optimize the objective function, it is defined as the weighted sum of squares of the spectral output errors.

[0115] Weight coefficient calculation formula:

[0116] in: : No. Real-time weight of class data; : Standardized real-time data value; : Dynamic adjustment factor; : Data confidence parameter, determined by the following formula: parameter is the sensitivity adjustment factor, is the historical average.

[0117] Step 3: Optimize the spectrum control strategy, that is, to improve the accuracy of spectrum control and energy consumption balance, the reward function and objective formula of the deep reinforcement learning model can be redesigned. For example, the objective function optimization:

[0118] in: : Spectral output accuracy score; : Real-time energy consumption, measured by current sensor; : Energy consumption weight adjustment factor, used to balance spectral accuracy and power consumption.

[0119] And for the dynamic adjustment of the reward function: according to the needs of the actual application scenario (agriculture or medical), For example, in agricultural scenarios: , give priority to reducing energy consumption; medical scenarios: reduce , giving priority to ensuring spectral accuracy; Step 4: Dynamically allocate computing resources. To solve the resource allocation problem of the system under high load conditions and further reduce power consumption, the system designed a distributed computing resource dynamic allocation mechanism based on real-time triggering and long-term optimization. The mechanism includes four core modules: task migration trigger conditions, synchronization mechanism, load prediction, and long-term optimization.

[0120] In the triggering conditions of task migration, such as the real-time monitoring indicators of edge nodes, CPU utilization (%): used to monitor the real-time computing pressure of edge computing nodes; GPU utilization (%): reflects the resource usage of computing-intensive tasks such as spectral control reasoning; task queue length (number of tasks): indicates the backlog of tasks currently to be processed by the edge node.

[0121] The triggering rule is: the system sets a dynamic threshold, and when the computing resources of the edge node are close to saturation, the migration of low-priority tasks is triggered:

[0122] in: : Edge node task queue length; :Dynamically adjusted task queue threshold, updated in real time according to historical load data and current computing pressure. Classification of high-priority and low-priority tasks: High-priority tasks: such as spectral control reasoning and multimodal data fusion, need to be completed on edge nodes first to ensure real-time performance; low-priority tasks: such as historical data storage and long-term model training, can be delayed and migrated to cloud servers.

[0123] In terms of synchronization mechanism and real-time guarantee, in order to ensure the real-time and synchronization of the system during task migration, the following synchronization mechanisms are designed, such as asynchronous queue management: the system uses distributed asynchronous queues to manage the task transmission of edge nodes and cloud servers respectively. The "first in, first out" rule is used to avoid communication blockage between modules and ensure the correct data processing order.

[0124] In terms of data consistency maintenance, such as during the migration process, task data is encrypted and verified by timestamp to prevent data duplication or loss during transmission; after receiving the results of cloud processing, the edge node re-verifies the task ID to ensure that the results correspond to the tasks one by one; at the same time, its real-time requirements are as follows: task migration delay: the system requires that the delay of task migration be controlled within 100ms to ensure the real-time requirements of high-priority tasks; dynamic adjustment of network bandwidth: when the network delay exceeds the set threshold (such as 200ms), the system automatically reduces the migration frequency and gives priority to retaining local computing; and in terms of load prediction mechanism, in order to optimize the long-term resource allocation efficiency, the system predicts future task loads based on the random forest model and dynamically adjusts the task allocation strategy. The input features of the random forest model, such as the model input features, cover the following three types of data: Real-time task load indicators: edge node CPU / GPU utilization; current task queue length; average current task completion time (unit: ms); historical operation data indicators: edge node task migration frequency; average cloud task queue length; system energy consumption (unit: watt); external environment indicators: current network bandwidth (unit: Mbps); network delay (unit: ms); the weight distribution rule is: the system sets different weights according to the importance of the features, that is, the real-time task load indicator weight: 50%, giving priority to reflecting the current load; the historical operation data indicator weight: 30%, used for long-term trend analysis; the external environment indicator weight: 20%, evaluating the network conditions for migration.

[0125] In terms of model output: predict the task load distribution within the next hour; dynamically adjust the task division ratio between edge nodes and the cloud.

[0126] And in the long-term operation, the system optimizes the task allocation strategy through genetic algorithm, with the goal of weighted minimization of task completion time and power consumption:

[0127] in: : No. The time to complete the task; : No. The power consumption of each task; , : The weight factor of time and power consumption is set dynamically.

[0128] The system can also dynamically allocate computing resources, collect and analyze data through the following closed-loop logic: real-time collection of operating data of edge nodes and cloud servers; task triggering and allocation: task migration or local processing based on triggering rules; load prediction and optimization: combining random forest prediction models with genetic algorithms to optimize long-term task allocation strategies; feedback and adjustment: the system regularly adjusts trigger conditions and allocation ratios based on operating data to ensure long-term stability.

[0129] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. 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 solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. An intelligent LED spectrum control system based on multi-dimensional data fusion, characterized in that: The system includes: The multi-dimensional data acquisition module is used to collect environmental spectrum data, temperature and humidity data, carbon dioxide concentration data and user behavior data in real time, among which: environmental spectrum data is collected by high-precision spectrum sensors; temperature and humidity data are collected by digital sensors; carbon dioxide concentration data is measured by infrared gas sensors; user behavior data is analyzed by cameras combined with image recognition algorithms; The data fusion processing module is connected to the multidimensional data acquisition module and is used to perform standardized preprocessing and dynamic fusion on the collected multidimensional data, including: using a fuzzy logic-based standardization algorithm to perform dimensionless processing on the data of different physical quantities; using a multimodal adaptive weight allocation algorithm, according to the time series variation characteristics of environmental spectrum data, temperature and humidity data, carbon dioxide concentration data and user behavior data, the weight coefficient of the spectral output parameter is calculated, and the weight coefficient is calculated by the following formula: in: For spectrum control The weight of the class data; For the collection of Real-time value of class data; It is a dynamic adjustment factor obtained based on historical data fitting, which is obtained by combining the time series prediction model LSTM with historical data training; is the dimension of the collected data; Indicates Spectral output coefficient at each time point; For the The value of the class data; The spectrum control algorithm module is connected to the data fusion processing module and is used to control the LED spectrum output in real time according to the data fusion result, including: using a spectrum optimization model based on deep reinforcement learning, the model input is the weight coefficient and the spectrum output target value after data fusion; in the model training process, the following objective function is used for dynamic optimization: in: is the optimization goal of the policy network; Based on the current network parameters Spectral control strategy; For time The spectral output effect score at the moment; is the reward discount factor, which controls the weight of the impact of future rewards on the current strategy; An LED spectrum output module is connected to the spectrum control algorithm module and is used to dynamically adjust the LED spectrum output according to the optimization result, including: realizing spectrum output through a multi-channel adjustable spectrum LED light source; configuring a fast-response current driver to dynamically adjust the current value of each band spectrum channel according to the target parameter output by the spectrum control algorithm module; The system collaborative control module is connected to the above modules and is used to coordinate the operation of the multi-dimensional data acquisition module, the data fusion processing module, the spectrum control algorithm module and the LED spectrum output module, including: configuring an event-driven mechanism for dynamically triggering data acquisition and processing; and dynamically allocating computing resources through a distributed collaborative optimization algorithm.

2. The intelligent LED spectrum control system based on multi-dimensional data fusion according to claim 1, characterized in that: The fuzzy logic normalization algorithm in the data fusion processing module includes the following steps: normalizing the data of different physical quantities to a range between 0 and 1 respectively; based on the dynamic range of data changes, using triangular membership functions and dynamically adjusted threshold intervals to construct fuzzy sets to reduce the scale differences between physical quantities.

3. The intelligent LED spectrum control system based on multi-dimensional data fusion according to claim 1, characterized in that: In the multi-modal adaptive weight allocation algorithm, the dynamic adjustment factor The calculation includes: collecting historical data in the target application scenario; performing time series fitting analysis based on the LSTM model to predict the long-term weight trend of each type of data on the spectral output; and using the gradient descent method to dynamically update the weight adjustment factor to optimize the real-time and robustness of the weight calculation.

4. The intelligent LED spectrum control system based on multi-dimensional data fusion according to claim 1, characterized in that: The training process of deep reinforcement learning in the spectral control algorithm module includes: generating an initial sample library of spectral control strategies based on multidimensional data collected offline; using a convolutional neural network to extract features of multidimensional data; constructing a reinforcement learning environment and defining a reward function as a weighted comprehensive score of the accuracy score of the spectral output and energy consumption; and using distributed parallel computing to accelerate the strategy optimization of deep reinforcement learning.

5. The intelligent LED spectrum control system based on multi-dimensional data fusion according to claim 1, characterized in that: The multi-channel adjustable spectrum LED light source in the LED spectrum output module includes: LED light sources of four spectral bands: red, green, blue and white; the light source of each band is independently powered by a constant current driver; and the PWM pulse width modulation technology is used to adjust the luminous intensity and duration of the light source of each band in real time.

6. The intelligent LED spectrum control system based on multi-dimensional data fusion according to claim 1, characterized in that: The event-driven mechanism in the system collaborative control module includes: detecting the real-time sampling frequency of the multi-dimensional data acquisition module; triggering the data fusion processing module to recalculate the weight coefficient when the spectral output deviation exceeds a preset threshold; and managing data transmission and control instructions between multiple modules through asynchronous queues.

7. The intelligent LED spectrum control system based on multi-dimensional data fusion according to claim 1, characterized in that: The calculation formula of the weight coefficient further combines the confidence parameter of the data , the confidence parameter is determined by the following formula: in: is the sensitivity adjustment factor, determined by experimental fitting; For real-time collection The value of the class data; is the historical mean of this data category.

8. The intelligent LED spectrum control system based on multi-dimensional data fusion according to claim 1, characterized in that: Reward discount factor in the deep reinforcement learning model The value of is adjusted dynamically according to the following formula: in: is the adjustment factor, which depends on the sensitivity requirement of spectral regulation; For time Spectral output score at the moment; Score threshold for the target.

Images