A modular multi-channel LED single-lamp controller and its precise dimming and independent detection method

Abstract

This invention discloses a modular multi-channel LED single-lamp controller and its precise dimming and independent detection method. The controller adopts a drawer-type hot-swappable architecture and can be expanded to multiple independent constant current outputs. The intelligent main control board is based on a dual-core heterogeneous design of an ARM main core and an AI NPU coprocessor, realizing deep integration of real-time control and edge intelligence. The method includes: adopting a three-level dimming strategy of coarse adjustment, fine adjustment, and AI optimization, combined with closed-loop feedback from a light sensor to achieve high-precision brightness control; using a single high-precision ADC, analog switch matrix, and programmable gain amplifier for time-division multiplexing sampling to complete independent detection of multiple channels of voltage, current, power, and energy; using a neural network fault prediction model to output the health level and remaining lifespan of each channel, performing predictive maintenance or fault isolation; supporting adaptive disaster recovery and load balancing to achieve seamless switching of fault channels; and based on a cloud-edge collaborative digital twin platform, completing remote monitoring, lifespan prediction, and intelligent operation and maintenance of lighting assets.

Description

Technical Field

[0001] This invention relates to the field of intelligent lighting control technology, and more specifically, to a modular multi-channel LED single-lamp controller and its precise dimming and independent detection method. Background Technology

[0002] Urban street lighting and smart light poles are important components of smart city construction. Traditional LED single-lamp controllers mostly adopt an integrated design, which mainly suffers from the following technical bottlenecks: Firstly, regarding multi-channel control, most existing controllers only support single-channel or a few-channel outputs, and there is a lack of effective electrical isolation between channels. When a short circuit or overload fault occurs in one channel, the fault will spread to other channels through the shared power supply or signal bus, causing the entire controller to fail. In addition, it is difficult for lamps of different voltage levels (such as low-voltage landscape lights and high-voltage street lights) to operate compatiblely under the same controller, limiting the integration requirements of "multi-pole integration" for smart light poles. Secondly, regarding independent detection, existing technologies usually only detect the total current and total voltage of the entire unit, and cannot distinguish the independent power consumption of each lamp. Although some solutions attempt to configure independent sampling circuits for each channel, the hardware cost increases linearly with the number of channels, and crosstalk is easily generated between multiple analog signals, making it difficult to balance cost and accuracy. Furthermore, regarding dimming accuracy, traditional PWM open-loop dimming methods rely on factory calibration. After long-term operation, due to factors such as LED light decay and component drift, the actual brightness can deviate from the set value by 15% to 25%, especially in the low brightness range where flicker and nonlinearity problems are particularly prominent. Finally, in terms of operation and maintenance management, existing controllers mostly adopt a reactive alarm mode, which cannot provide early warning of potential faults. Maintenance personnel can only respond passively, resulting in low maintenance efficiency and high operating costs. At the same time, each controller operates independently and lacks digital twin capabilities for collaboration with the cloud, making it impossible to achieve full lifecycle asset management and predictive maintenance.

[0003] Therefore, there is an urgent need for a modular multi-channel LED single-lamp controller and its control method that is highly integrated, highly precise, highly intelligent, and has edge computing capabilities. Summary of the Invention

[0004] In order to overcome the problems of poor multi-channel isolation, high cost of independent testing, low dimming accuracy, and passive operation and maintenance of existing LED single lamp controllers, this invention discloses a modular multi-channel LED single lamp controller and its precise dimming and independent testing method, which can effectively solve the above-mentioned technical problems.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A method for precise dimming and independent detection in a modular multi-channel LED single-lamp controller is disclosed. The controller includes an intelligent main control board, at least one modular multi-channel constant current driver board, and a multi-dimensional detection matrix module. The intelligent main control board comprises a dual-core heterogeneous architecture of an ARM main core and an AI NPU coprocessor. The multi-dimensional detection matrix module includes a single high-precision analog-to-digital converter (ADC), an analog switch matrix, and a programmable gain amplifier (PGA). The method includes: The intelligent main control board receives dimming commands or preset strategies, and generates PWM control signals according to the target brightness value using a hierarchical multi-mode dimming strategy. The hierarchical multi-mode dimming strategy includes coarse adjustment mode, fine adjustment mode and AI optimized dimming mode. The intelligent main control board performs time-division multiplexing sampling on each constant current drive channel through the multi-dimensional detection matrix module to obtain voltage, current, power and energy data for each channel. The time-division multiplexing sampling includes: controlling the analog switch matrix to sequentially select each sampling node through the GPIO of the intelligent main control board, and dynamically configuring the gain of the PGA, so that the single high-precision ADC can collect the voltage and current signals of each channel in a time-division manner. The intelligent main control board inputs the collected voltage, current and temperature data of each channel into the neural network fault prediction model built into the AI ​​NPU coprocessor, and outputs the health level and remaining service life prediction value of each drive channel. The intelligent main control board performs fault response actions according to the health level. The fault response actions include: generating a predictive maintenance message and uploading it to the cloud to trigger pre-dispatch of work orders when the health level is a warning or critical level; and sending a PWM forced zero command to the constant current drive board of the corresponding channel and cutting off the power supply branch of that channel when the health level is a fault level.

[0006] Furthermore, the hierarchical multimode dimming strategy includes: When the target brightness change exceeds a preset threshold or the difference between the current brightness and the target brightness is greater than a first difference threshold, the intelligent main control board switches to the coarse adjustment mode and uses the first step of long-range adjustment of the PWM duty cycle to quickly approach the target brightness. When the current brightness enters the preset deviation range of the target brightness, the intelligent main control board switches to the fine-tuning mode, adjusts the PWM duty cycle with the second step size range, and collects the actual light output of the LED in real time as feedback through the built-in light sensor to form a current-luminous flux dual closed-loop control until the deviation between the actual light output and the target brightness is less than the second difference threshold. During preset low-load periods, the intelligent main control board switches to the AI-optimized dimming mode, inputting real-time traffic flow, pedestrian flow, ambient illuminance, and weather condition data at intersections into the reinforcement learning model. The reinforcement learning model then outputs the target brightness scheme for each path to minimize total energy consumption.

[0007] Furthermore, the time-division multiplexing sampling also includes the following steps: Step A: The intelligent main control board traverses the currently connected N drive channels with a sampling period T_s, and performs the following for the i-th channel: turns off the analog switches of all other channels, and turns on the current sampling node and voltage sampling node of the i-th channel; Step B: Perform blind sampling on the i-th channel to obtain a signal strength estimate, and dynamically set the gain factor of the PGA based on the signal strength estimate. ; Step C: In gain Voltage values ​​collected and current value and according to and To reconstruct the true physical quantity, where m is the number of bits in the ADC. For the full-scale voltage range, For the full-scale current, This is the ADC reference voltage. This is the ADC reference current; Step D: Calculate instantaneous power The cumulative electrical energy of the i-th channel is obtained by integrating the results. Step E: Repeat steps A to D until all N channels have been sampled.

[0008] Furthermore, the neural network fault prediction model is an LSTM time-series prediction network or a backpropagation neural network, and its input features include: the historical effective value sequence of the current channel voltage, the historical effective value sequence of the current channel, the historical temperature sequence, the historical value sequence of the PWM duty cycle, and the historical value sequence of the power factor. The neural network fault prediction model outputs the health score H∈[0,1] of the channel and the remaining useful life prediction value RUL; The intelligent main control board classifies health levels based on health scores: H≥0.9 indicates healthy, 0.7≤H<0.9 indicates warning, 0.4≤H<0.7 indicates critical, and H<0.4 indicates fault. When the health status reaches the warning level, the intelligent main control board packages the predictive maintenance message into a data packet containing the device ID, channel number, health score, RUL, and fault feature vector, and uploads it to the cloud event arbitration center through the dual-mode communication module.

[0009] Furthermore, the method also includes an adaptive disaster recovery and load balancing process: The intelligent main control board monitors the health score and operating current of each drive channel in real time. When the health of a primary channel enters a critical level or the predicted remaining service life is lower than the preset service life threshold, the intelligent main control board selects the backup channel with the highest health from the pre-configured list of backup redundant channels. The intelligent main control board sends a pre-conduction command to the constant current drive board where the backup channel is located through the redundant control bus, so that the output voltage of the backup channel is established to the same level as the main channel. After the pre-conduction is completed, the intelligent main control board synchronously executes the following at the beginning of the next PWM cycle: seamlessly switching the PWM control signal of the main channel to the backup channel, sending a PWM forced zero command to the main channel and cutting off its power supply branch; during the switching process, phase interleaving modulation technology is used to ensure that the switching phase of the backup channel is synchronized with the original main channel. After the switch is completed, the intelligent main control board will upload all historical operating data packets of the main channel to the cloud and generate a fault root cause analysis report.

[0010] Furthermore, the task partitioning of the dual-core heterogeneous architecture includes: The ARM main core runs a real-time operating system and undertakes hard real-time tasks and soft real-time tasks. The hard real-time tasks include: PWM signal refresh, microsecond-level response of fault protection logic, and channel switching control of the analog switch matrix. The soft real-time tasks include: communication protocol stack processing, data log recording, and status indicator control. The AI ​​NPU coprocessor is dedicated to running lightweight neural network inference models, including real-time forward inference computation of the neural network fault prediction model, as well as offline training and online inference of the reinforcement learning dimming model. The ARM main core and the AI ​​NPU coprocessor exchange data through shared memory or message queues. The AI ​​NPU coprocessor notifies the ARM main core of the calculated health score and RUL via an interrupt.

[0011] Furthermore, the modular multi-channel constant current driver board adopts a drawer-type hot-swappable architecture. Each driver board integrates 4 or 8 independent constant current output channels. The basic configuration supports 4 channels, and can be expanded to a maximum of 64 outputs by inserting multiple driver boards. Each constant current drive channel is electrically isolated from the intelligent main control board through an optocoupler and a PWM isolation transformer. Each channel's power input terminal is equipped with an independent overcurrent protection circuit and transient voltage suppressor. The overcurrent protection circuit consists of a sampling resistor, an operational amplifier comparator, and a MOSFET self-locking turn-off circuit. The intelligent main control board communicates with each constant current driver board via I2C bus or SPI bus, identifies the insertion status of the driver board, and dynamically configures the output current range and voltage range of each channel to achieve unified access for LED loads of different voltage levels, including low voltage 12V-48V DC and high voltage 110V-220V AC.

[0012] Furthermore, the method also includes a cloud-edge collaborative digital twin operation and maintenance process: The intelligent main control board constructs a local twin parameter set for each LED load. The local twin parameter set includes: IV characteristic curve parameters of the drive channel, temperature-impedance characteristic curve parameters, PWM duty cycle-output current mapping table, and LED light decay model parameters based on the Arrhenius accelerated aging model. The intelligent main control board synchronizes the incremental data of the local twin parameter set to the cloud digital twin platform through the dual-mode communication module according to the strategy cycle. The dual-mode communication module includes PLC power line carrier communication and Zigbee / LoRa wireless communication dual redundant channels, which serve as hot backups for each other. The cloud-based digital twin platform receives the twin parameters of all access controllers, generates regional and global digital twin maps of the lighting system, and calculates the remaining luminous flux maintenance life of each LED load based on the light decay model parameters. When the remaining life is lower than a preset threshold, a maintenance work order is automatically generated and pushed to the operation and maintenance personnel.

[0013] Furthermore, a modular multi-channel LED single-lamp controller includes: The intelligent main control board includes a dual-core heterogeneous architecture of an ARM main core and an AI NPU coprocessor. The ARM main core is used to generate multiple PWM control signals and execute real-time control and fault protection logic. The AI ​​NPU coprocessor is used to run a neural network fault prediction model and a reinforcement learning dimming model. At least one modular multi-channel constant current driver board, each driver board integrates multiple independent constant current output channels, each channel has independent optocoupler and transformer dual electrical isolation, independent overcurrent protection circuit and adaptive voltage matching circuit; the driver board adopts a drawer-type hot-swappable structure and communicates with the intelligent main control board through a bus backplane; The multidimensional detection matrix module includes a high-precision Sigma-Delta ADC, an analog switch matrix composed of multiple low on-resistance analog switches, and a programmable gain amplifier (PGA). The input terminals of the analog switch matrix are connected to the current sampling nodes and voltage sampling nodes of each constant current drive channel, respectively, and the output terminal is connected to the input terminal of the high-precision Sigma-Delta ADC via the PGA. The channel selection control terminal of the analog switch matrix and the gain configuration terminal of the PGA are both connected to the GPIO port of the intelligent main control board. The dual-mode communication module includes a PLC power line carrier communication unit and a Zigbee / LoRa wireless communication unit. The dual-mode communication module is connected to the intelligent main control board and is used to realize dual redundant communication with the cloud platform and the edge gateway. An environmental sensing expansion interface is used to connect external light sensors, temperature and humidity sensors, and air quality sensors. The output of the environmental sensing expansion interface is connected to the ADC input of the intelligent main control board.

[0014] Furthermore, the ARM core of the intelligent main control board is an ARM Cortex-M7 with a main frequency of 480MHz. It has 16 independent high-resolution PWM generators built in, each with a 16-bit resolution and a PWM frequency range of 100Hz-20kHz. It also supports hardware-level PWM interleaved modulation (PIM) function, which makes the rising edges of each PWM signal interleaved on the time axis. The intelligent main control board also integrates a light sensor interface, a power management chip PMIC, and a hardware security encryption chip ATSHA204A. The modular multi-channel LED single lamp controller also includes a power supply unit, which includes an AC-DC preamp module. The input of the AC-DC preamp module is connected to AC mains power, and the output provides multiple isolated DC voltages to the intelligent main control board, each constant current drive board and the multi-dimensional detection matrix module.

[0015] Compared with existing technologies, the advantages of this invention are as follows: In terms of hardware architecture, this invention achieves true independence for each output through modular hot-swappable design and dual electrical isolation between channels. A fault in any channel can be precisely isolated without affecting the normal operation of other channels. Simultaneously, the controller can be flexibly expanded to adapt to lighting scenarios of different scales and is compatible with lighting equipment of various voltage levels. This provides a reliable foundation platform for the unified access and centralized management of multiple devices on smart light poles, improving the overall reliability and deployment flexibility of the system. Regarding detection technology, a time-division multiplexing sampling architecture using a single high-precision analog-to-digital converter combined with an analog switch matrix and a programmable gain amplifier eliminates the high cost and signal crosstalk problems caused by independent sampling circuits for each channel in traditional solutions. The dynamic gain adjustment mechanism enables the sampling system to accurately capture weak current signals and withstand large current surges, achieving high-precision independent measurement over a wide dynamic range while reducing hardware costs. The voltage, current, power, power factor, and accumulated energy of each output can be obtained independently. In terms of dimming quality and energy saving, a dual closed-loop control strategy linking coarse and fine adjustment is adopted. Combined with real-time feedback from the light sensor on the actual light output, a dual closed-loop calibration of current and luminous flux is formed, achieving smooth, flicker-free linear dimming across the entire brightness range and maintaining excellent brightness consistency even after long-term operation. The AI-optimized dimming mode dynamically generates the optimal brightness scheme based on real-time perception of traffic flow, pedestrian flow, ambient illuminance, and weather conditions, reducing power consumption while ensuring lighting safety and comfort. In terms of intelligent operation and maintenance, the built-in neural network fault prediction model can comprehensively analyze multi-dimensional data such as historical voltage, current, and temperature to detect the deterioration trend of equipment health status in advance and proactively generate maintenance work orders, transforming passive post-event repairs into proactive pre-event maintenance, shortening fault response and repair time. The adaptive disaster recovery mechanism automatically and seamlessly switches the load to the backup redundant channel when the channel is close to failure, ensuring that the core lighting function is not interrupted. The cloud-edge collaborative digital twin platform further aggregates the operating parameters of each controller, constructs a regional lighting digital map, dynamically predicts the remaining lifespan of lamps based on the light decay model, and automatically triggers maintenance plans. This realizes the upgrade of the operation and maintenance mode from experience-driven to data-driven, and from passive response to scientific decision-making, and comprehensively reduces the full life cycle operation cost of the lighting system. Attached Figure Description

[0016] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other embodiments can be derived from the provided drawings without creative effort.

[0017] Figure 1A schematic diagram of a modular multi-channel LED single-lamp controller provided in an embodiment of the present invention; Figure 2 The flowchart illustrates a method for precise dimming and independent detection of a modular multi-channel LED single-lamp controller, as provided in an embodiment of the present invention. Detailed Implementation

[0018] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] It is understood that the specific embodiments described herein are merely illustrative of the embodiments of this application and are not intended to limit the embodiments of this application. Furthermore, it should be noted that, for ease of description, the accompanying drawings only show the parts related to the embodiments of this application, not all structures. Those skilled in the art, after reading this specification, should be able to realize that any combination of technical features can constitute an optional implementation method, provided that the technical features do not contradict each other.

[0021] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship. In the description of this application, "multiple" means two or more, and "several" means one or more.

[0022] like Figure 1 As shown, Figure 1This is a schematic diagram of the system architecture of a modular multi-channel LED single-lamp controller provided in an embodiment of the present invention. The modular multi-channel LED single-lamp controller provided in this embodiment includes an intelligent main control board, at least one modular multi-channel constant current driver board, a multi-dimensional detection matrix module, a dual-mode communication module, and an environmental perception expansion interface. The intelligent main control board comprises a dual-core heterogeneous architecture with an ARM main core and an AI NPU coprocessor. The ARM main core is used to generate multiple PWM control signals and execute real-time control and fault protection logic, while the AI ​​NPU coprocessor is used to run neural network fault prediction models and reinforcement learning dimming models. In one specific embodiment, the ARM main core uses an ARM Cortex-M7 with a main frequency of 480MHz, and the AI ​​NPU coprocessor uses a K210 chip, which integrates a neural network processor and can efficiently run lightweight neural network inference models. The intelligent main control board and each modular multi-channel constant current driver board are connected via a bus backplane. The input terminals of the multi-dimensional detection matrix module are connected to the current sampling nodes and voltage sampling nodes of each constant current driver channel, respectively, and the output terminals of the multi-dimensional detection matrix module are connected to the ADC input terminals of the intelligent main control board. The dual-mode communication module connects to the intelligent main control board to achieve redundant communication with the cloud platform and edge gateway. The output of the environmental perception expansion interface connects to the ADC input of the intelligent main control board for connecting external light sensors, temperature and humidity sensors, and air quality sensors.

[0023] like Figure 2 As shown, Figure 2 This is a flowchart illustrating a modular multi-channel LED single-lamp controller's precise dimming and independent detection method, provided in an embodiment of the present invention. In this embodiment, the intelligent main control board receives dimming commands from a cloud platform or a locally preset strategy, and generates PWM control signals using a hierarchical multi-mode dimming strategy based on the target brightness value. The hierarchical multi-mode dimming strategy includes coarse adjustment mode, fine adjustment mode, and AI-optimized dimming mode. The intelligent main control board uses a multi-dimensional detection matrix module to perform time-division multiplexing sampling on each constant current drive channel to obtain voltage, current, power, and energy data for each channel. The specific process of time-division multiplexing sampling includes: using the GPIO control of the intelligent main control board to sequentially select each sampling node via an analog switch matrix, and dynamically configuring the gain factor of the programmable gain amplifier, enabling a single high-precision Sigma-Delta analog-to-digital converter to collect the voltage and current signals of each channel in a time-division manner. The intelligent main control board inputs the collected voltage, current, and temperature data of each channel into a neural network fault prediction model built into the AI ​​NPU coprocessor, outputting the health level and remaining lifespan prediction value for each drive channel. The intelligent main control board performs fault response actions based on the health level: when the health level is at the warning or critical level, it generates a predictive maintenance message and uploads it to the cloud to trigger the pre-dispatch of work orders; when the health level is at the fault level, it sends a PWM forced zero command to the constant current drive board of the corresponding channel and cuts off the power supply branch of that channel.

[0024] As described above, this embodiment of the invention achieves independent and precise dimming and independent detection of multiple LED lamps by setting up an intelligent main control board, a modular multi-channel constant current driver board, a multi-dimensional detection matrix module, and an AI NPU coprocessor. The intelligent main control board generates PWM control signals based on a hierarchical multi-mode dimming strategy. It obtains voltage, current, power, and energy data for each channel through time-division multiplexing sampling of the multi-dimensional detection matrix module. Using a built-in neural network fault prediction model, it outputs the health level and remaining lifespan prediction value for each driver channel and executes corresponding fault response actions based on the health level. This embodiment of the invention achieves high-precision independent detection and predictive maintenance of multiple LED lamps without increasing hardware costs, solving the technical problems of high cost, low dimming accuracy, and passive response in operation and maintenance of existing multi-channel controllers.

[0025] Based on the above embodiments, the hierarchical multimode dimming strategy provided by the present invention specifically includes the following process: When the target brightness change exceeds a preset threshold (e.g., the target brightness change rate is greater than 20% per second) or the difference between the current brightness and the target brightness is greater than a first difference threshold (e.g., the difference is greater than 15% of the current brightness), the intelligent main control board switches to coarse adjustment mode, adjusting the PWM duty cycle in a first-step long range to quickly approach the target brightness. The first-step long range is a step size interval larger than the step size used in fine adjustment mode; preferably, the first-step long range can be a duty cycle step size of 1% to 5%. In coarse adjustment mode, the intelligent main control board uses multiple large-step adjustments to quickly converge the actual brightness to near the target brightness.

[0026] When the current brightness falls within the preset deviation range of the target brightness (e.g., the deviation between the current brightness and the target brightness is less than 5% of the current brightness), the intelligent main control board switches to fine-tuning mode, adjusting the PWM duty cycle with a second step size range. The second step size range is a step range smaller than the step size used in coarse-tuning mode; preferably, the second step size range can be a duty cycle step size of 0.1% or smaller. In fine-tuning mode, the intelligent main control board uses a built-in light sensor to collect the actual light output of the LED in real time as feedback, forming a current-luminous flux dual closed-loop control. Specifically, the current closed-loop control uses the actual current value fed back by the built-in current sampling circuit as input, adjusting the PWM duty cycle to make the actual current approach the target current; the luminous flux closed-loop control uses the actual luminous flux collected by the light sensor as input, correcting the target current value, so that the deviation between the actual light output and the target brightness gradually decreases until the deviation between the actual light output and the target brightness is less than a second difference threshold (e.g., the deviation is less than 1% of the current brightness).

[0027] During preset low-load periods (e.g., midnight to 5:00 AM), the intelligent main control board switches to AI-optimized dimming mode. In this mode, the intelligent main control board inputs real-time traffic flow, pedestrian flow, ambient illuminance, and weather condition data from the intersection into the reinforcement learning model. The reinforcement learning model uses the target brightness of each road as its action space and minimizes the total energy consumption of the intersection as its reward objective, outputting the target brightness scheme for each road. In one specific embodiment, the reinforcement learning model employs a deep Q-network algorithm. The state space includes the current time, traffic density in each direction, number of waiting pedestrians, ambient illuminance value, and weather type encoding. The action space consists of discrete adjustment levels for the brightness of each road (e.g., 0%, 10%, 20%, ..., 100%, a total of 11 levels). After offline training, the model is deployed in the AI ​​NPU coprocessor and performs inference every 5 minutes during online operation, outputting the optimal brightness scheme to minimize total energy consumption.

[0028] Based on the above embodiments, the time-division multiplexing sampling provided in this embodiment of the invention specifically includes the following steps: Step A: The intelligent main control board iterates through the currently connected N drive channels at a sampling period. For the i-th channel, the intelligent main control board controls the analog switch matrix via GPIO to turn off the analog switches of all other channels, only turning on the current sampling node and voltage sampling node of the i-th channel. The sampling period is preferably 1 millisecond, and the value of N depends on the number of currently inserted modular multi-channel constant current drive boards and the number of channels on each drive board, supporting a maximum of 64 channels.

[0029] Step B: The intelligent main control board performs blind sampling on the i-th channel to obtain a signal strength estimate. The specific process of blind sampling is as follows: Voltage and current values ​​are quickly acquired once at the default gain factor (e.g., ×1), and the approximate signal strength range is determined based on the acquisition results. If the acquired voltage or current value is close to the full scale of the ADC, the signal strength is high, and the gain factor needs to be reduced; if the acquired voltage or current value is less than 10% of the ADC's full scale, the signal strength is low, and the gain factor needs to be increased. Based on the signal strength estimate, the intelligent main control board dynamically sets the gain factor of the programmable gain amplifier via GPIO. . Selected from a preset set of gain levels, including but not limited to eight levels: ×1, ×2, ×4, ×8, ×16, ×32, ×64, and ×128.

[0030] Step C: The intelligent main control board operates at the set gain factor. Voltage values ​​collected and current value The actual physical quantity is then reconstructed using the following formula: and

[0031] Where m is the number of bits in the ADC (in this embodiment, a 32-bit high-precision Sigma-Delta ADC with a resolution of ADS1263 is used). The voltage full-scale range (e.g., 0~36V). The current full-scale range (e.g., 0~5A). This is the ADC reference voltage (typically 2.5V or 4.096V). The ADC reference current (determined by both the sampling resistor value and the reference voltage; for example, when the sampling resistor is 10mΩ), = / 10mΩ = 250A (used to convert the current sampling voltage into the actual current value), the analog switch uses ADG708 or equivalent chip, and its on-resistance is ≤2.5Ω.

[0032] Step D: The intelligent main control board calculates the instantaneous power. The cumulative power of the i-th channel is obtained by integration. Specifically, the instantaneous power calculated in this sampling is multiplied by the sampling period to obtain the power increment within the sampling period, and then accumulated into the power accumulation register of the i-th channel.

[0033] Step E: The intelligent main control board repeats steps A to D until all N channels have been sampled, and then enters the next sampling cycle.

[0034] Based on the above embodiments, the present invention provides a neural network fault prediction model. The neural network fault prediction model is a Long Short-Term Memory (LSTM) time-series prediction network or a backpropagation neural network. In this embodiment, a LTM time-series prediction network is preferred because it can effectively capture the long-term dependencies of time-series data such as voltage, current, and temperature.

[0035] The input features of the Long Short-Term Memory (LSTM) time-series prediction network include: the historical effective voltage value sequence of the current channel (last 24 sampling points, one point per hour), the historical effective current value sequence (last 24 sampling points), the historical temperature sequence (last 24 sampling points, one point every 30 minutes), the historical PWM duty cycle value sequence (last 24 sampling points), and the historical power factor value sequence (last 24 sampling points). All input features are normalized and mapped to the [-1, 1] interval before being fed into the network.

[0036] The Long Short-Term Memory (LSTM) time-series prediction network outputs a health score H ∈ [0,1] for the channel, and a remaining lifetime prediction value RUL (in hours). A higher health score H indicates a healthier device, and the remaining lifetime prediction value RUL represents the predicted remaining usable time.

[0037] The intelligent main control board classifies health levels based on health scores: when H≥0.9, it is considered a healthy level; when 0.7≤H<0.9, it is considered a warning level; when 0.4≤H<0.7, it is considered a critical level; and when H<0.4, it is considered a fault level.

[0038] When the health status reaches the warning level, the intelligent main control board packages the predictive maintenance message into a data packet containing the device ID, channel number, health score, predicted remaining service life, and fault feature vector, and uploads it to the cloud event arbitration center via the dual-mode communication module. Among them, the fault feature vector is the feature output of the intermediate layer of the Long Short-Term Memory Time-Series Prediction Network, which is used to indicate the main factors that cause the health status to decline (such as excessive temperature, excessive current fluctuation, or abnormal power factor).

[0039] Based on the above embodiments, the adaptive disaster recovery and load balancing process provided by the embodiments of the present invention specifically includes the following steps: Step 501: The intelligent main control board monitors the health score and operating current of each drive channel in real time. When the health of a primary channel enters the critical level (i.e., 0.4≤H<0.7) or the predicted remaining service life is lower than the preset service life threshold (e.g., the predicted remaining service life is less than 720 hours), the disaster recovery preparation process is triggered.

[0040] Step 502: The intelligent main control board selects the backup channel with the highest health score from the pre-configured list of backup redundant channels. This list is pre-defined by maintenance personnel during system configuration and contains one or more backup channel identifiers corresponding to each primary channel. The intelligent main control board iterates through this list, queries the current health score of each backup channel, and selects the backup channel with the highest health score as the target switching channel.

[0041] Step 503: The intelligent main control board sends a pre-conduction command to the constant current driver board of the backup channel via the redundant control bus, causing the output voltage of the backup channel to reach the same level as the primary channel. The specific pre-conduction process is as follows: The intelligent main control board reads the current target current value of the primary channel and sends this target current value to the constant current driver board of the backup channel via the redundant control bus, triggering the constant current driver circuit of the backup channel to establish the corresponding output voltage, but without connecting the load. After the pre-conduction is completed, the backup channel is in hot standby mode.

[0042] Step 504: After pre-conduction is completed, the intelligent main control board synchronously performs a switching operation at the beginning of the next PWM cycle: seamlessly switching the PWM control signal of the primary channel to the backup channel, while simultaneously sending a PWM forced zero command to the primary channel and cutting off its power supply branch. During the switching process, the intelligent main control board uses phase interleaving modulation technology to ensure that the phase of the backup channel is synchronized with the original primary channel, avoiding current surges to the LED driver power supply due to phase abrupt changes. The specific implementation of phase synchronization is as follows: the intelligent main control board records the phase offset of the primary channel in the current PWM cycle, and sets the PWM generator phase register of the backup channel to this offset at the switching time.

[0043] Step 505: After the switch is completed, the intelligent main control board will package and upload all historical operating data of the main channel (including all voltage, current, temperature, power factor, health score sequence, etc. since the channel was put into use) to the cloud. The cloud will generate a fault root cause analysis report based on this data for maintenance personnel to make fault diagnosis and hardware repair decisions.

[0044] Based on the above embodiments, this invention further defines a specific task partitioning method for the dual-core heterogeneous architecture. The ARM main core of the intelligent main control board runs a real-time operating system (RT-Thread operating system is used in this embodiment), undertaking hard real-time tasks and soft real-time tasks. The hard real-time tasks include: periodic refreshing of PWM signals (refresh period ≤ 1 millisecond), microsecond-level response of fault protection logic (response time ≤ 10 microseconds), and channel switching control of the analog switch matrix (switching time ≤ 5 microseconds). The soft real-time tasks include: communication protocol stack processing (parse and encapsulate TCP / IP protocol stack), data log recording (recording sampled data and events to local flash memory), and control of status indicator lights.

[0045] The AI ​​NPU coprocessor is dedicated to running lightweight neural network inference models, including real-time forward inference computation for neural network fault prediction models (performing inference on all channels every 10 minutes) and offline training and online inference for reinforcement learning dimming models. Offline training of the reinforcement learning dimming model is completed in the cloud, and the trained model parameters are downloaded to the AI ​​NPU coprocessor via a dual-mode communication module. Online inference is performed every 5 minutes.

[0046] The ARM main core and the AI ​​NPU coprocessor exchange data via shared memory or message queues. Specifically, three buffers are defined in the shared memory area: an input buffer for the ARM main core to write raw data such as collected voltage, current, and temperature to the AI ​​NPU coprocessor; an output buffer for the AI ​​NPU coprocessor to write health scores and remaining lifespan predictions to the ARM main core; and a control buffer for transmitting control flags and interrupt status. After completing inference calculations, the AI ​​NPU coprocessor writes the health score and remaining lifespan predictions to the output buffer and notifies the ARM main core to read the data via a hardware interrupt.

[0047] Based on the above embodiments, this invention provides a modular multi-channel constant current driver board. The modular multi-channel constant current driver board adopts a drawer-type hot-swappable architecture, with each driver board integrating 4 or 8 independent constant current output channels. In the basic configuration, one driver board is inserted into the backplane of the intelligent main control board, supporting 4 outputs; by inserting multiple driver boards, it can be expanded to a maximum of 64 outputs. Each driver board communicates with the intelligent main control board through a standardized bus backplane connector, supporting hot-swapping, meaning that if a single driver board fails, it can be replaced online without power interruption.

[0048] Each constant current drive channel is electrically isolated from the intelligent main control board through an optocoupler and a PWM isolation transformer, with an isolation voltage of no less than 3000Vrms. Specifically, the PWM signal output by the intelligent main control board first undergoes a first-stage isolation through an optocoupler isolator (model 6N137), and then undergoes a second-stage isolation through a pulse transformer before being sent to the control terminal of the constant current drive chip.

[0049] Each channel's power input is equipped with an independent overcurrent protection circuit and transient voltage suppressor. The overcurrent protection circuit consists of a sampling resistor, an operational amplifier comparator, and a metal-oxide-semiconductor field-effect transistor (MOSFET) latch-up turn-off circuit. Its working principle is as follows: the sampling resistor is connected in series in the power supply circuit, converting the current signal into a voltage signal; the operational amplifier comparator compares this voltage signal with a preset overcurrent threshold voltage; when the current exceeds the threshold, the operational amplifier comparator outputs a high level, driving the MOSFET gate to turn off, and simultaneously sending a fault interrupt signal to the intelligent main control board. The overcurrent protection circuit's operating time does not exceed 5 microseconds. The transient voltage suppressor is connected in parallel at the power input to absorb transient overvoltages such as lightning surges and electrostatic discharge.

[0050] The intelligent main control board communicates with each constant current driver board via an I2C bus or serial peripheral interface bus, identifies the insertion status of the driver board, and dynamically configures the output current and voltage ranges of each channel to achieve unified access for LED loads of different voltage levels. Accessible voltage levels include low-voltage 12V to 48V DC and high-voltage 110V to 220V AC. For low-voltage DC loads, the constant current driver board operates in buck constant current mode; for high-voltage AC loads, the constant current driver board internally incorporates an active power factor correction circuit and an isolated flyback converter to convert AC input to constant current output.

[0051] Based on the above embodiments, the cloud-edge collaborative digital twin operation and maintenance process provided by the embodiments of the present invention includes the following: The intelligent main control board constructs a local twin parameter set for each LED load. The local twin parameter set includes: IV characteristic curve parameters of the drive channel (output voltage-output current relationship at different duty cycles, stored in the form of polynomial fitting coefficients), temperature-impedance characteristic curve parameters (LED equivalent resistance values ​​at different temperatures), a PWM duty cycle-output current mapping table (a mapping table with 256 uniformly sampled points), and LED light decay model parameters based on the Arrhenius accelerated aging model. The Arrhenius accelerated aging model takes the following form: Where L is the current luminous flux, L0 is the initial luminous flux, α is the aging rate constant, and t is the running time. The activation energy is given by k, the Boltzmann constant, and T, the junction temperature. Locally stored parameters include the initial luminous flux L0, an estimate of the aging rate constant α, and the activation energy. Typical values.

[0052] The intelligent main control board synchronizes incremental data of the local twin parameter set to the cloud-based digital twin platform via a dual-mode communication module according to a policy cycle (e.g., once per hour). The dual-mode communication module includes a PLC power line carrier communication unit and a Zigbee / LoRa wireless communication unit, which serve as hot-standby redundant channels for each other. The main control board prioritizes power line carrier communication for data transmission. When a degradation in power line carrier communication quality is detected (bit error rate exceeding one in a thousand or three consecutive handshake failures), it automatically switches to the Zigbee or LoRa wireless communication channel.

[0053] The cloud-based digital twin platform receives twin parameters from all access controllers and generates regional and global digital twin maps of the lighting system. The digital twin map uses a geographic information system map as its base map, overlaying the location of each light pole, the real-time status of each LED light fixture, its health score, and its predicted remaining lifespan. The cloud platform calculates the remaining luminous flux maintenance lifespan of each LED load based on luminous decay model parameters. Specifically, it substitutes the current luminous flux L into the Arrhenius model to obtain the cumulative operating time, then sets L to the target maintenance luminous flux (e.g., 70% of the initial luminous flux) to calculate the cumulative operating time when the target maintenance luminous flux is reached. The remaining luminous flux maintenance lifespan is the difference between these two values. When the remaining lifespan falls below a preset threshold (e.g., 1000 hours), the cloud platform automatically generates a maintenance work order and pushes it to the mobile terminals of the corresponding regional maintenance personnel.

[0054] The following describes the operation of the modular multi-channel LED single-lamp controller according to an embodiment of the present invention, using a complete workflow example.

[0055] After the system powers on, the power management module converts AC mains power into multiple isolated DC voltages to power the intelligent main control board, various constant current driver boards, and the multi-dimensional detection matrix module. The intelligent main control board loads firmware, initializes various peripheral modules (including the PWM generator, ADC, GPIO, communication interface, etc.), and then loads the configuration parameters stored in local flash memory, including dimming strategy parameters, calibration tables, and network configuration of the dual-mode communication module. Subsequently, the intelligent main control board performs a self-test of the multi-dimensional detection matrix module, confirming normal hardware operation by reading the ID register of the analog switch matrix and the configuration register of the ADC. The status indicator light changes from flashing red to solid green, indicating that the system has entered standby mode.

[0056] When a dimming command is received, the intelligent main control board first analyzes the target brightness value in the command and selects the dimming mode based on the difference between the current brightness value and the target brightness value. If the difference is large (greater than 15% of the current brightness), it enters coarse adjustment mode, quickly adjusting the PWM duty cycle in 2% steps; when the brightness falls within ±5% of the target brightness, it switches to fine adjustment mode, finely adjusting the PWM duty cycle in 0.1% steps, while simultaneously reading the actual luminous flux fed back by the light sensor to form a closed-loop calibration until the deviation between the actual luminous flux and the target brightness is less than 1%.

[0057] During dimming, the intelligent main control board simultaneously performs time-division multiplexing sampling with a 1-millisecond cycle. The analog switch matrix is ​​controlled via GPIO to sequentially select channels 1 through N. For each channel selection, blind sampling is first performed to determine the signal strength, and the PGA gain is dynamically set (×1 or ×2 gain is used for streetlight channels with high normal operating current, and ×16 or ×32 gain is used for standby or low-brightness low-current channels). Then, voltage and current data are collected at the optimal gain, instantaneous power is calculated, and energy is accumulated. After sampling of all N channels, the data is stored in the corresponding dual buffer for each channel.

[0058] Every 10 minutes, the AI ​​NPU coprocessor is awakened to execute neural network fault prediction inference. It reads voltage, current, temperature, and PWM duty cycle data from the most recent 24 sampling cycles of each channel from the dual buffer, performs normalization, and inputs it into the Long Short-Term Memory (LSTM) timing prediction network. The network outputs a health score and remaining lifetime prediction value for each channel. For channels with a health score below 0.9, the ARM main core performs corresponding actions based on the score level: at the warning level (0.7 ≤ H < 0.9), a predictive maintenance message is packaged and uploaded to the cloud via the dual-mode communication module; at the critical level (0.4 ≤ H < 0.7), in addition to uploading the message, a disaster recovery preparation process is initiated, selecting the channel with the highest health score from the backup channel list for pre-conduction; at the fault level (H < 0.4), a PWM forced zeroing command is immediately sent to the channel, and the power supply branch is cut off, while the load is switched to the pre-conducted backup channel.

[0059] Once the cloud-based digital twin platform receives the twin parameters from each controller, it updates the digital map of the area lighting and calculates the remaining lifespan of each LED luminaire based on the Arrhenius light decay model. When the remaining lifespan of a luminaire is less than 1000 hours, the cloud automatically generates a maintenance work order, which includes the luminaire ID, location coordinates, predicted remaining lifespan, and suggested replacement time, and pushes it to the mobile terminal of the corresponding maintenance personnel.

[0060] Through the above-described architecture, method, and process, the embodiments of the present invention realize modular expansion of multi-channel LED lamps, high-precision independent detection, dual closed-loop precise dimming, edge intelligent predictive maintenance, and cloud-edge collaborative digital twin operation and maintenance, comprehensively improving the reliability, energy efficiency, and intelligent operation and maintenance level of smart lighting systems.

[0061] Note that the above are merely preferred embodiments and the technical principles applied in this invention. Those skilled in the art will understand that the embodiments of this invention are not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the protection scope of this invention. For example, the ARM Cortex-M7 main core and K210 AI NPU coprocessor used in the above embodiments are merely examples; in actual implementation, other processor models can be selected based on cost, power consumption, and computing power requirements. Similarly, the ADS1263 ADC and ADG708 analog switch are also merely examples; other chips with similar functions and performance indicators are equally applicable. Therefore, although the embodiments of this invention have been described in detail above, the embodiments of this invention are not limited to the above embodiments. More other equivalent embodiments may be included without departing from the concept of the embodiments of this invention, and the scope of the embodiments of this invention is determined by the scope of the appended claims.

[0062] The same or similar labels correspond to the same or similar parts; The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent. Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A modular multi-channel LED single-lamp controller with precise dimming and independent detection method, characterized in that, An application is made to a modular multi-channel LED single-lamp controller, the controller comprising an intelligent main control board, at least one modular multi-channel constant current driver board, and a multi-dimensional detection matrix module. The intelligent main control board includes a dual-core heterogeneous architecture of an ARM main core and an AI NPU coprocessor. The multi-dimensional detection matrix module includes a single high-precision analog-to-digital converter (ADC), an analog switching matrix, and a programmable gain amplifier (PGA). The method includes: The intelligent main control board receives dimming commands or preset strategies, and generates PWM control signals according to the target brightness value using a hierarchical multi-mode dimming strategy. The hierarchical multi-mode dimming strategy includes coarse adjustment mode, fine adjustment mode and AI optimized dimming mode. The intelligent main control board performs time-division multiplexing sampling on each constant current drive channel through the multi-dimensional detection matrix module to obtain voltage, current, power and energy data for each channel. The time-division multiplexing sampling includes: controlling the analog switch matrix to sequentially select each sampling node through the GPIO of the intelligent main control board, and dynamically configuring the gain of the PGA, so that the single high-precision ADC can collect the voltage and current signals of each channel in a time-division manner. The intelligent main control board inputs the collected voltage, current and temperature data of each channel into the neural network fault prediction model built into the AI ​​NPU coprocessor, and outputs the health level and remaining service life prediction value of each drive channel. The intelligent main control board performs fault response actions according to the health level. The fault response actions include: generating a predictive maintenance message and uploading it to the cloud to trigger pre-dispatch of work orders when the health level is a warning or critical level; and sending a PWM forced zero command to the constant current drive board of the corresponding channel and cutting off the power supply branch of that channel when the health level is a fault level.

2. The method according to claim 1, characterized in that, The hierarchical multimode dimming strategy includes: When the target brightness change exceeds a preset threshold or the difference between the current brightness and the target brightness is greater than a first difference threshold, the intelligent main control board switches to the coarse adjustment mode and uses the first step of long-range adjustment of the PWM duty cycle to quickly approach the target brightness. When the current brightness enters the preset deviation range of the target brightness, the intelligent main control board switches to the fine-tuning mode, adjusts the PWM duty cycle with the second step size range, and collects the actual light output of the LED in real time as feedback through the built-in light sensor to form a current-luminous flux dual closed-loop control until the deviation between the actual light output and the target brightness is less than the second difference threshold. During preset low-load periods, the intelligent main control board switches to the AI-optimized dimming mode, inputting real-time traffic flow, pedestrian flow, ambient illuminance, and weather condition data at intersections into the reinforcement learning model. The reinforcement learning model then outputs the target brightness scheme for each path to minimize total energy consumption.

3. The method according to claim 1, characterized in that, The time-division multiplexing sampling also includes the following steps: Step A: The intelligent main control board traverses the currently connected N drive channels with a sampling period T_s, and performs the following for the i-th channel: turns off the analog switches of all other channels, and turns on the current sampling node and voltage sampling node of the i-th channel; Step B: Perform blind sampling on the i-th channel to obtain a signal strength estimate, and dynamically set the gain factor of the PGA based on the signal strength estimate. ; Step C: In gain Voltage values ​​collected and current value and according to and To reconstruct the true physical quantity, where m is the number of bits in the ADC. For the full-scale voltage range, For the full-scale current, This is the ADC reference voltage. This is the ADC reference current; Step D: Calculate instantaneous power The cumulative electrical energy of the i-th channel is obtained by integrating the results. Step E: Repeat steps A to D until all N channels have been sampled.

4. The method according to claim 1, characterized in that, The neural network fault prediction model is an LSTM time-series prediction network or a backpropagation neural network, and its input features include: the historical effective value sequence of voltage, the historical effective value sequence of current, the historical temperature sequence, the historical value sequence of PWM duty cycle, and the historical value sequence of power factor of the current channel. The neural network fault prediction model outputs the health score H∈[0,1] of the channel and the remaining service life prediction value RUL; The intelligent main control board classifies health levels based on health scores: H≥0.9 indicates healthy, 0.7≤H<0.9 indicates warning, 0.4≤H<0.7 indicates critical, and H<0.4 indicates fault. When the health status reaches the warning level, the intelligent main control board packages the predictive maintenance message into a data packet containing the device ID, channel number, health score, RUL, and fault feature vector, and uploads it to the cloud event arbitration center through the dual-mode communication module.

5. The method according to claim 1, characterized in that, The method also includes an adaptive disaster recovery and load balancing process: The intelligent main control board monitors the health score and operating current of each drive channel in real time. When the health of a primary channel enters a critical level or the predicted remaining service life is lower than the preset service life threshold, the intelligent main control board selects the backup channel with the highest health from the pre-configured list of backup redundant channels. The intelligent main control board sends a pre-conduction command to the constant current drive board where the backup channel is located through the redundant control bus, so that the output voltage of the backup channel is established to the same level as the main channel. After the pre-conduction is completed, the intelligent main control board synchronously executes the following at the beginning of the next PWM cycle: seamlessly switching the PWM control signal of the main channel to the backup channel, and sending a PWM forced zero command to the main channel and cutting off its power supply branch. During the switching process, phase interleaving modulation technology is used to ensure that the switching phase of the backup channel remains synchronized with the original primary channel; After the switch is completed, the intelligent main control board will upload all historical operating data packets of the main channel to the cloud and generate a fault root cause analysis report.

6. The method according to claim 1, characterized in that, The task partitioning of the dual-core heterogeneous architecture includes: The ARM main core runs a real-time operating system and undertakes hard real-time tasks and soft real-time tasks. The hard real-time tasks include: PWM signal refresh, microsecond-level response of fault protection logic, and channel switching control of the analog switch matrix. The soft real-time tasks include: communication protocol stack processing, data log recording, and status indicator control. The AI ​​NPU coprocessor is dedicated to running lightweight neural network inference models, including real-time forward inference computation of the neural network fault prediction model, as well as offline training and online inference of the reinforcement learning dimming model. The ARM main core and the AI ​​NPU coprocessor exchange data through shared memory or message queues. The AI ​​NPU coprocessor notifies the ARM main core of the calculated health score and RUL via an interrupt.

7. The method according to claim 1, characterized in that, The modular multi-channel constant current driver board adopts a drawer-type hot-swappable architecture. Each driver board integrates 4 or 8 independent constant current output channels. The basic configuration supports 4 channels, and can be expanded to a maximum of 64 outputs by inserting multiple driver boards. Each constant current drive channel is electrically isolated from the intelligent main control board through an optocoupler and a PWM isolation transformer. Each channel's power input terminal is equipped with an independent overcurrent protection circuit and transient voltage suppressor. The overcurrent protection circuit consists of a sampling resistor, an operational amplifier comparator, and a MOSFET self-locking turn-off circuit. The intelligent main control board communicates with each constant current driver board via I2C bus or SPI bus, identifies the insertion status of the driver board, and dynamically configures the output current range and voltage range of each channel to achieve unified access for LED loads of different voltage levels, including low voltage 12V-48V DC and high voltage 110V-220V AC.

8. The method according to claim 1, characterized in that, The method also includes a cloud-edge collaborative digital twin operation and maintenance process: The intelligent main control board constructs a local twin parameter set for each LED load. The local twin parameter set includes: IV characteristic curve parameters of the drive channel, temperature-impedance characteristic curve parameters, PWM duty cycle-output current mapping table, and LED light decay model parameters based on the Arrhenius accelerated aging model. The intelligent main control board synchronizes the incremental data of the local twin parameter set to the cloud digital twin platform through the dual-mode communication module according to the strategy cycle. The dual-mode communication module includes PLC power line carrier communication and Zigbee / LoRa wireless communication dual redundant channels, which serve as hot backups for each other. The cloud-based digital twin platform receives the twin parameters of all access controllers, generates regional and global digital twin maps of the lighting system, and calculates the remaining luminous flux maintenance life of each LED load based on the light decay model parameters. When the remaining life is lower than a preset threshold, a maintenance work order is automatically generated and pushed to the operation and maintenance personnel.

9. A modular multi-channel LED single-lamp controller, characterized in that, include: The intelligent main control board includes a dual-core heterogeneous architecture of an ARM main core and an AI NPU coprocessor. The ARM main core is used to generate multiple PWM control signals and execute real-time control and fault protection logic. The AI ​​NPU coprocessor is used to run a neural network fault prediction model and a reinforcement learning dimming model. At least one modular multi-channel constant current driver board, each driver board integrates multiple independent constant current output channels, each channel has independent optocoupler and transformer dual electrical isolation, independent overcurrent protection circuit and adaptive voltage matching circuit; the driver board adopts a drawer-type hot-swappable structure and communicates with the intelligent main control board through a bus backplane; The multidimensional detection matrix module includes a high-precision Sigma-Delta ADC, an analog switch matrix composed of multiple low on-resistance analog switches, and a programmable gain amplifier (PGA). The input terminals of the analog switch matrix are connected to the current sampling nodes and voltage sampling nodes of each constant current drive channel, respectively, and the output terminal is connected to the input terminal of the high-precision Sigma-Delta ADC via the PGA. The channel selection control terminal of the analog switch matrix and the gain configuration terminal of the PGA are both connected to the GPIO port of the intelligent main control board. The dual-mode communication module includes a PLC power line carrier communication unit and a Zigbee / LoRa wireless communication unit. The dual-mode communication module is connected to the intelligent main control board and is used to realize dual redundant communication with the cloud platform and the edge gateway. An environmental sensing expansion interface is used to connect external light sensors, temperature and humidity sensors, and air quality sensors. The output of the environmental sensing expansion interface is connected to the ADC input of the intelligent main control board.

10. The controller according to claim 9, characterized in that, The intelligent main control board has an ARM Cortex-M7 main core with a main frequency of 480MHz. It has 16 independent high-resolution PWM generators built-in, each with a 16-bit resolution and a PWM frequency range of 100Hz-20kHz. It also supports hardware-level PWM interleaved modulation (PIM) function, which makes the rising edges of each PWM signal interleaved on the time axis. The intelligent main control board also integrates a light sensor interface, a power management chip PMIC, and a hardware security encryption chip ATSHA204A. The modular multi-channel LED single lamp controller also includes a power supply unit, which includes an AC-DC preamp module. The input of the AC-DC preamp module is connected to AC mains power, and the output provides multiple isolated DC voltages to the intelligent main control board, each constant current drive board and the multi-dimensional detection matrix module.

Images