An LED display screen intelligent monitoring, predictive maintenance and adaptive scene arrangement system

Abstract

The application discloses an LED display screen intelligent monitoring, predictive maintenance and self-adaptive scene arrangement system, and relates to the technical fields of intelligent operation and maintenance and Internet of Things, and comprises the following steps: a sensor layer comprises at least one ultra-low-cost intelligent sensor small plate embedded in the interior of an LED display screen box body; a communication module is configured with a 6-byte lightweight communication protocol; a central control host comprises a core board and an extension floor; an MQTT message bus adopts a four-level hierarchical namespace topic structure; an AI predictive maintenance engine comprises an AI inference engine and an AI training pipeline; and an AI self-adaptive scene arrangement module adopts a four-stage progressive cold start strategy.The application adopts the above-described LED display screen intelligent monitoring, predictive maintenance and self-adaptive scene arrangement system to realize full-stack intelligent management of the LED display screen at extremely low hardware cost, and fills the technical gap of low-cost state sensing and self-adaptive scene arrangement in the field of LED display.

Description

Technical Field

[0001] This invention relates to the fields of intelligent operation and maintenance and Internet of Things technology, and in particular to an intelligent monitoring, predictive maintenance and adaptive scene orchestration system for LED displays. Background Technology

[0002] LED displays are widely used in exhibition halls, conference rooms, outdoor advertising, stage performances, and other scenarios. However, existing LED display integrated management systems suffer from the following core technical deficiencies: (1) Lack of low-cost, multi-dimensional real-time status sensing methods. Traditional LED displays do not have the ability to collect environmental parameters such as temperature, illuminance, and voltage in real time. Maintenance relies on regular on-site inspections (usually 1-2 times per month), which cannot detect hidden faults such as power module aging, cooling fan blockage, and abnormal power supply voltage in a timely manner. According to publicly available data, about 73% of LED display faults have detectable abnormal temperature or voltage signals 1-3 days before they occur, but existing systems cannot capture them.

[0003] (2) High cost of sensing hardware and high deployment threshold. Existing third-party monitoring solutions for LED displays (NovaCare, Kalite LED Management Platform, etc.) rely on dedicated monitoring cards or protocol conversion modules and only support their own brand of receiver cards, which cannot be compatible with multi-brand equipment. The dual barriers of hardware cost and installation labor have resulted in the vast majority of existing LED displays being in a state of no sensing monitoring.

[0004] (3) The system is closed and has poor cross-manufacturer compatibility. Existing protocols are not open to each other: NovaStar uses the proprietary NovaBus protocol, Carlet uses the proprietary NovaStar compatible protocol (which is not actually fully compatible), and the protocols of small and medium-sized manufacturers such as Shenzhen Gray change frequently. Once a user adopts a certain brand of LED screen, he or she is bound to the manufacturer and cannot be managed uniformly across brands.

[0005] (4) Severe fragmentation of equipment control protocols. The communication protocols of the equipment in the exhibition hall are highly fragmented: traditional lighting equipment uses RS485 / DALI protocol, LED display screens use TCP / IP proprietary protocol, air conditioners use RS485 / Modbus protocol, electric curtains and electric doors use relay dry contact control, infrared / radar sensors output GPIO level, and power line carrier communication has a separate protocol. The existing central control solution requires individual configuration, programming and debugging of each device, resulting in high system integration complexity.

[0006] (5) Lack of intelligent proactive operation and maintenance capabilities. Existing technology remains in a passive response mode after a failure occurs. According to industry statistics, the average response time from the occurrence of a failure to a customer complaint is about 4-8 hours, which leads to shortened equipment life, waste of operation and maintenance human resources, and a continuous decline in customer satisfaction.

[0007] Existing LED display maintenance solutions each have their shortcomings: original monitoring cards such as NovaCare and Carlet can only monitor temperature and voltage, rely on fixed threshold alarms, and are only compatible with their own brands; the PT100 general-purpose temperature controller with PLC can only monitor temperature; although the ADAM industrial acquisition card can collect temperature, voltage, and current, neither of them has AI processing capabilities.

[0008] Therefore, the core technical problems to be solved by this invention are as follows: how to achieve real-time (second-level) monitoring of three dimensions of LED display screen temperature, illuminance, and power supply voltage at extremely low hardware cost, making large-scale deployment economically feasible; how to achieve unified data acquisition of different brands of receiver cards through standard hardware interfaces without relying on the proprietary protocols of LED receiver card manufacturers; how to achieve unified management of exhibition hall equipment by being compatible with multiple device control protocols such as RS485, RS232, relays, and TCP / IP through a single hardware platform; how to use high-frequency time-series sensor data to train a lightweight anomaly detection model to achieve early prediction of four types of faults: power module, LED beads, receiver card, and fan; how to implement a cold start strategy that can operate normally from day one without historical data, and gradually improve the level of intelligence as data accumulates over time; and how to achieve unified high-frequency data acquisition (every 3 seconds / message), MQTT message distribution, AI real-time inference, and reliable offline operation on edge devices. Summary of the Invention

[0009] The purpose of this invention is to provide an intelligent monitoring, predictive maintenance, and adaptive scene arrangement system for LED displays, thereby solving the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides an intelligent monitoring, predictive maintenance and adaptive scene orchestration system for LED displays, including a sensor layer, a central control host, a communication module, an MQTT message bus, an AI predictive maintenance engine, an AI adaptive scene orchestration module and a front-end interaction module; The sensor layer includes at least one ultra-low-cost smart sensor board embedded inside the LED display cabinet. The smart sensor board is equipped with an ARM Cortex-M0+ architecture MCU, an NTC temperature sensor, a photosensitive sensor, and a voltage detection circuit. The smart sensor board is connected to the LED receiver card debugging port through a UART interface to collect the temperature, illuminance, and power supply voltage data of the LED display in real time. The communication module is configured with a 6-byte lightweight communication protocol. It uploads sensor-collected data through the UART communication channel at a baud rate of 115200bps and a reporting cycle of once every 3 seconds, using a 6-byte fixed frame format. The central control unit consists of a core board and an expansion baseboard. The core board is equipped with an ARM Cortex-A7 architecture processor, runs the Buildroot lightweight Linux system, builds an MQTT message bus, and has a built-in AI inference engine. The expansion baseboard integrates a variety of physical interfaces. The MQTT message bus adopts the real-time data processing architecture of EMQX MQTT Broker and uses a four-level hierarchical namespace topic structure; The AI ​​predictive maintenance engine consists of an AI inference engine and an AI training pipeline. The AI ​​inference engine is deployed on the central control unit, equipped with TFLite or ONNX Runtime, and is used to perform anomaly detection and health score calculation based on data collected by the number of sensors. It also adopts a hierarchical automatic response mechanism, and directly issues a brightness reduction protection command when an extreme anomaly is detected. The AI ​​training pipeline is deployed in the cloud, using the Scikit-learn or LightGBM framework. It retrains the IsolationForest anomaly detection model weekly based on the accumulated sensor data, and sends the optimized model to the central control unit via the MQTT message bus to complete the update. The AI ​​adaptive scene orchestration module runs on the central control host and adopts a four-stage progressive cold start strategy, gradually iterating from rule-based scheduling to intelligent scene scheduling based on time-series prediction. The front-end interaction module is built using the React+Ant Design Pro framework and pushes data to the central control host in real time via WebSocket.

[0011] Therefore, the present invention employs the above-mentioned intelligent monitoring, predictive maintenance, and adaptive scene arrangement system for LED displays, which has the following beneficial effects: (1) Significantly reduced hardware costs: The sensor board adopts an ultra-low cost MCU and a minimalist peripheral circuit design, making large-scale mass deployment an economically feasible solution.

[0012] (2) True cross-vendor protocol compatibility: It can be directly adapted to 6 mainstream receiver cards such as Nova Nebula, Carlett, and Shenzhen Grayscale, without the need for a dedicated protocol converter.

[0013] (3) Unified dual-mode hardware ecosystem: Version A and Version B share the same firmware, protocol and parsing software, significantly improving R&D efficiency and significantly reducing operation and maintenance costs.

[0014] (4) Unified control of multi-protocol devices: The central control base plate integrates 12 physical interfaces, which can be compatible with multiple device protocols such as lighting, air conditioning, LED screen, and curtain.

[0015] (5) Edge AI real-time inference: TFLite Runtime local inference, end-to-end latency <200ms, local instant protection against extreme anomalies.

[0016] (6) MQTT real-time message bus: A single machine supports the access of 200 sensors, and the four-level hierarchical topic structure enables accurate data distribution.

[0017] (7) Multi-type fault prediction: The Isolation Forest model with 9-dimensional features is used to predict four types of faults in advance.

[0018] (8) Three-level automatic response protection: set up advance response of early warning, alarm and protection level. The protection level is executed directly locally without relying on the cloud.

[0019] (9) P0-P3 cold start progressive strategy: The rule engine can run normally on the first day, and the full AI matures on the 30th day, so that it can be used immediately upon deployment.

[0020] (10) User feedback closed loop: supports AI recommendation, user decision, manual operation learning and false alarm feedback into the negative sample pool, and the user always maintains the final control.

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of an embodiment of an intelligent monitoring, predictive maintenance, and adaptive scene arrangement system for LED displays according to the present invention. Detailed Implementation

[0023] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0025] Please see Figure 1This invention provides an intelligent monitoring, predictive maintenance, and adaptive scene orchestration system for LED displays. It includes an ultra-low-cost dual-mode intelligent sensor layer that uses a 6-byte fixed frame format protocol to upload temperature, illumination, and voltage data every 3 seconds via UART, and is compatible with 6 mainstream LED receiver card brands. A modular central control host adopts a core board + expansion baseboard architecture, integrating multiple device control interfaces such as RS485×4, RS232, relay×4, DI×4, and Gigabit Ethernet×2, running an MQTT message bus and a TFLite edge AI inference engine. A 9-dimensional feature-based fault prediction model based on Isolation Forest provides early warnings for four types of faults: power module, LED beads, receiver card, and fan. A P0-P3 progressive cold start strategy ensures the system's rule engine is usable on day 1 and the full AI mode is mature by day 30.

[0026] In this embodiment, the system includes at least one ultra-low-cost intelligent sensor board, which is embedded inside the LED display cabinet and connected to the LED receiver card's debugging port via a standard UART interface. The sensor board includes an ultra-low-cost ARM Cortex-M0+ series MCU, an NTC temperature sensor, a photoresistor, and a voltage detection circuit, used to collect temperature, illuminance, and power supply voltage data from the LED display, and report this data every 3 seconds in a 6-byte fixed frame format via the UART communication channel at a baud rate of 115200bps. The central control host includes an ARM... The core board with a Cortex-A7 or higher processor and the expansion board integrating multiple physical interfaces communicate with the LED receiver card via a TCP / IP network. The central control host runs Buildroot lightweight Linux and an MQTT message bus. The MQTT message bus publishes / subscribes to sensor data and device control commands according to a hierarchical topic structure. An edge AI inference engine, deployed on the central control host, runs TFLite or ONNXRuntime, performs real-time anomaly detection and health score calculation based on sensor data, and directly issues a brightness reduction protection command when extreme anomalies are detected. An AI training pipeline, deployed in the cloud, uses the Scikit-learn or LightGBM framework to retrain the Isolation Forest anomaly detection model weekly using accumulated sensor data, and sends the optimized model to the central control host via MQTT. An AI adaptive scene orchestration module, running on the central control host, adopts a four-stage progressive strategy from P0 to P3, gradually transitioning from a rule engine to intelligent scene scheduling based on time series prediction in a cold start state without historical data. A front-end interaction module, built using the React+Ant Design Pro framework, achieves real-time data push with the central control host via WebSocket. The system specifically includes the following six core subsystems: Subsystem 1: Ultra-low cost dual-morphology smart sensor layer.

[0027] An ultra-low-cost intelligent sensor board for LED displays is embedded inside the LED housing and connects to the LED receiver card via a UART interface through a standard debug port (J6 debug port). The main control MCU uses an STM32G030 series (ARM Cortex-M0+, 64MHz, 8KB SRAM, 32KB Flash), integrating an NTC temperature sensor (10kΩ@25℃, B value 3950), a photosensor (GL5516), and a voltage detection circuit (100kΩ+10kΩ voltage divider network). Two physical forms are provided: Version A, the standard board, uses a 25×20mm PCB size and connects to the LED receiver card's J6 debug port via a 4-pin header; the MCU is in a TSSOP-20 package. Version B, the mini dongle, uses a 12×10mm PCB size, a USB-A male connector, and connects to the MCU internally via a USB-to-serial chip; the MCU is in a TSSOP-8 package. Version A and Version B share the same firmware source code. Hardware mode adaptation is achieved through compilation macro definitions or runtime GPIO pin level detection. Both modes use the exact same communication protocol frame format, and the data source is completely transparent to the host computer.

[0028] Subsystem 2: 6-byte lightweight communication protocol.

[0029] A 6-byte fixed-frame communication protocol for resource-constrained MCUs reports data every 3 seconds at a baud rate of 115200bps. The specific frame structure is as follows: Byte0 = Frame header 0xAA; Byte1 = Integer code of temperature value multiplied by 2; Byte2 = Integer value of photosensitivity 0-255; Byte3 = Integer code of voltage value multiplied by 10; Byte4 = Status bits (bit0 temperature warning, bit1 temperature alarm, bit2 voltage warning, bit3 voltage alarm, bit4 communication error); Byte5 = Checksum (0xFF - sum of Byte1 to Byte4) & 0xFF. The total frame length is only 6 bytes, adapting to an 8-byte UART hardware FIFO depth, eliminating the need for buffer overflow handling.

[0030] Subsystem 3: Modular central control host (edge ​​computing + multi-protocol control).

[0031] The core board uses an Allwinner T113 processor (Cortex-A7 dual-core @ 1.2GHz, 512MB DDR3, 4GB eMMC), runs Buildroot Linux (kernel 5.4), and integrates a TFLite 2.10+ custom C++ inference engine. The expansion board integrates the following physical interfaces: 4 RS485 interfaces (SP3485EN transceiver with TVS protection), 1 RS232 interface (MAX3232 level converter), 4 relay output interfaces (G5V-1 with EL357 optocoupler isolation), 4 DI dry contact input interfaces (EL357 optocoupler isolation), 2 Gigabit Ethernet interfaces (RTL8211F), and a 4-port USB 2.0 hub (SL2.1A). The core board and expansion board are connected via an 80-pin board-to-board connector.

[0032] Subsystem 4: Real-time data pipeline based on MQTT.

[0033] The real-time data processing architecture is based on the EMQX MQTT Broker, and the topic structure adopts a four-level hierarchical namespace: cloudnexus / {site_id} / {device_id} / sensor topic (raw sensor data, published every 3 seconds), cloudnexus / {site_id} / {device_id} / status topic (device status changes), cloudnexus / {site_id} / {device_id} / command topic (downlink control commands), and cloudnexus / {site_id} / alert topic (alarm events). Local SQLite caching of 7 days of data supports offline working mode; optional PostgreSQL cloud time-series storage is also available. A single central control host supports simultaneous access from 200 sensors, with end-to-end data latency <200ms.

[0034] Subsystem 5: AI-driven predictive maintenance engine.

[0035] (1) LED display screen comprehensive health scoring model: ; ; ; ; in, The score is based on temperature (deducting points for deviations from the ideal temperature of 35℃). Score for voltage. Score based on runtime (50,000-hour design life). Score historical faults. This is the current real-time operating temperature. This is the current real-time operating voltage. Rated voltage, For the cumulative number of operating days, This represents the number of times a failure has occurred in the past 30 days.

[0036] Health status is divided into normal (90-100), attention (70-89), warning (40-69), and alert (<40).

[0037] (2) Multi-type fault prediction model: Based on the Isolation Forest algorithm, using a 9-dimensional feature vector (mean temperature / standard deviation / rate of change, mean voltage / standard deviation / rate of change, mean photosensitive value, runtime, historical fault count), with contamination=0.05, the model is retrained in the cloud every 7 days and a new model is distributed via MQTT (INT8 quantization). It can predict power module faults (24-48 hours, target accuracy ≥85%), accelerated LED decay (1-2 weeks, ≥70%), receiver card abnormalities (4-8 hours, ≥80%), and fan blockage (12-24 hours, ≥75%).

[0038] (3) Three-level automatic response mechanism: Level 1 (early warning level) reduces brightness to a specified ratio + notification + troubleshooting steps; Level 2 (alarm level) reduces brightness to a lower ratio + AI root cause analysis + screen off suggestion; Level 3 (protection level, temperature > 85℃ or voltage < 180V) reduces brightness directly locally, without relying on cloud service connection. Brightness reduction protection command has higher priority than all other control commands.

[0039] Subsystem 6: AI Adaptive Scene Orchestration Engine.

[0040] A progressive AI adaptive scene orchestration method for smart exhibition halls includes four progressive stages: P0-P3. P0 Basic Rules Phase (Day 1): Based on the rule engine, the photosensitive adaptive LED screen brightness adjustment and delayed exit standby mechanism, fixed timed scene switching, and the use of the photosensitive adaptive brightness adjustment function: ; ; in, For the final output brightness, Based on the baseline brightness, Real-time ambient light intensity, The sensitivity coefficient is adjusted adaptively for brightness. For general ambient light input variables, As a reference luminance for ambient light, The value is constrained to [0.6, 1.4], and an unattended standby rule is superimposed. When no one is detected on site, the standby brightness is reduced. P1 Weekly Learning Phase (Day 7): Aggregate statistics by week and time slice (30-minute granularity), automatically identify regular / irregular days using Jaccard similarity coefficient, and output a natural language AI learning report; P2 parameter adaptive phase (day 14): When the rate of change of ambient light exceeds the threshold, a gradual transition is executed, and when the screen temperature difference is >10℃, an automatic shading / fan failure alarm is pushed. P3 Prediction and Warm-up Phase (Day 30): Based on the AR time series model, predict the probability of activity in the next 2 hours. When the probability is ≥70%, automatically warm up 15 minutes in advance.

[0041] Cold start strategy: Rule engine available on day 1, weekly mode learning completed on day 7, parameters adaptive on day 14, and full AI mode mature on day 30. User feedback closed loop: Users can choose to accept AI recommendations with one click, manually adjust them, or reject them (not use them). The system records users' manual actions as personalized preference data. If a user rejects the same type of advice repeatedly, similar advice will not be pushed to them again for 7 days. Users can mark AI alarms as false alarms, and the corresponding false alarm samples are stored in the negative sample pool for iterative optimization of the fault prediction model.

[0042] Example 1: Standard LED screen sensor board (Version A).

[0043] Hardware configuration: The main control MCU is STM32G030F6P6 (TSSOP-20 package), the temperature sensor is MF52 10kΩ B3950 NTC thermistor (0805 package), the photosensitive sensor is GL5516 photosensitive resistor (5mm through-hole), the voltage divider resistor is 100kΩ+10kΩ (0603 package), the filter capacitor is 100nF+10μF (0603 package), the connector is 2×2Pin 2.54mm header, and the PCB is a 25×20mm FR4 2-layer board.

[0044] Firmware logic: Initialize the system clock to 48MHz and UART1 115200-8N1; the main loop executes temperature acquisition (NTC voltage divider → 12-bit ADC → 100-point LUT lookup table → actual temperature) every 3 seconds, photosensor acquisition (10-bit ADC → 0-255 mapping), voltage acquisition (10-bit ADC → voltage divider ratio conversion → integer encoding), status bit encoding → checksum calculation → UART transmission. Thresholds can be remotely configured and written to Flash via UART.

[0045] Example 2: Mini Dongle (Version B, Touchscreen All-in-One).

[0046] Hardware configuration: The main control MCU is STM32G030J6 (TSSOP-8 package), the USB to serial port chip is CH340G (SOP-16 package), the temperature sensor is MF52 10kΩ B3950 NTC thermistor (0805 package), the photosensitive sensor is GL5516 photosensitive resistor (5mm through-hole), the voltage divider resistor is 100kΩ+10kΩ (0603 package), the filter capacitor is 100nF+10μF (0603 package), the USB header is 2×4Pin 2.0mm, the USB female connector is USB-A with 90° bent pins, and the PCB is a 12×10mm FR4 2-layer board.

[0047] Example 3: Cross-brand receiver card compatibility, as shown in Table 1 below.

[0048] Table 1 Cross-brand receiver card compatibility

[0049] Example 4: Actual deployment test in the exhibition hall (60 days).

[0050] Test environment: A medium-sized enterprise showroom, approximately 300 square meters, with 46 LED screens, 20 touch screen all-in-one machines, 16 lighting circuits, 4 air conditioners, and 2 motorized curtains. The test results are shown in Table 2 below.

[0051] Table 2. Test Results of Actual Deployment in the Exhibition Hall

[0052] 60-day fault prediction results: power module malfunctions were predicted 3 times in advance (100%), LED chip attenuation trends were predicted 1 time in advance (50%), receiver card communication malfunctions were predicted 4 times in advance (80%), and fan blockages were predicted 2 times in advance (100%). The overall accuracy rate was 83.3%.

[0053] Therefore, this invention employs the aforementioned intelligent monitoring, predictive maintenance, and adaptive scene orchestration system for LED displays. The self-developed CloudNexus system overcomes the limitations of existing technologies, capable of collecting multiple parameters such as temperature, illumination, and voltage. It leverages AI to achieve fault prediction and adaptive scene control, while also supporting compatible access to LED displays from different brands, resulting in superior overall performance. This invention achieves full-stack intelligent management of LED displays at extremely low hardware costs, filling the technological gap in low-cost state perception and adaptive scene orchestration in the LED display field.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A smart monitoring, predictive maintenance, and adaptive scene arrangement system for LED displays, characterized in that: It includes a sensor layer, a central control host, a communication module, an MQTT message bus, an AI predictive maintenance engine, an AI adaptive scene orchestration module, and a front-end interaction module; The sensor layer includes at least one ultra-low-cost smart sensor board embedded inside the LED display cabinet. The smart sensor board is equipped with an ARM Cortex-M0+ architecture MCU, an NTC temperature sensor, a photosensitive sensor, and a voltage detection circuit. The smart sensor board is connected to the LED receiver card debugging port through a UART interface to collect the temperature, illuminance, and power supply voltage data of the LED display in real time. The communication module is configured with a 6-byte lightweight communication protocol. It uploads sensor-collected data through the UART communication channel at a baud rate of 115200bps and a reporting cycle of once every 3 seconds, using a 6-byte fixed frame format. The central control unit consists of a core board and an expansion baseboard. The core board is equipped with an ARM Cortex-A7 architecture processor, runs the Buildroot lightweight Linux system, builds an MQTT message bus, and has a built-in AI inference engine. The expansion baseboard integrates a variety of physical interfaces. The MQTT message bus adopts the real-time data processing architecture of EMQX MQTT Broker and uses a four-level hierarchical namespace topic structure; The AI ​​predictive maintenance engine consists of an AI inference engine and an AI training pipeline. The AI ​​inference engine is deployed on the central control unit, equipped with TFLite or ONNX Runtime, and is used to perform anomaly detection and health score calculation based on data collected by the number of sensors. It also adopts a hierarchical automatic response mechanism, and directly issues a brightness reduction protection command when an extreme anomaly is detected. The AI ​​training pipeline is deployed in the cloud, using the Scikit-learn or LightGBM framework. It retrains the IsolationForest anomaly detection model weekly based on the accumulated sensor data, and sends the optimized model to the central control unit via the MQTT message bus to complete the update. The AI ​​adaptive scene orchestration module runs on the central control host and adopts a four-stage progressive cold start strategy, gradually iterating from rule-based scheduling to intelligent scene scheduling based on time-series prediction. The front-end interaction module is built using the React+Ant Design Pro framework and pushes data to the central control host in real time via WebSocket.

2. The LED display intelligent monitoring, predictive maintenance, and adaptive scene arrangement system according to claim 1, characterized in that, The smart sensor board is available in two physical forms: Version A standard small board: adopts a 25×20mm PCB size, connects to the J6 debugging port of the LED receiver card through a 4Pin header, and the MCU adopts a TSSOP-20 package; Version B Mini Dongle: It adopts a 12×10mm PCB size, USB-A male connector, and internally connects to the MCU via a USB to serial chip. The MCU is in a TSSOP-8 package. Version A and Version B share the same firmware source code. Hardware mode adaptation is achieved through compilation macro definitions or runtime GPIO pin level detection. Both modes use the same communication protocol frame format.

3. The LED display screen intelligent monitoring, predictive maintenance, and adaptive scene arrangement system according to claim 1, characterized in that, The 6-byte fixed frame format is specifically as follows: Byte0 = Frame header 0xAA; Byte1 = the integer encoding of the temperature value multiplied by 2; Byte2 = Photosensitive value, an integer value between 0 and 255; Byte3 = the integer encoding of the voltage value multiplied by 10; Byte4 = Status bits, which include bit0 temperature warning, bit1 temperature alarm, bit2 voltage warning, bit3 voltage alarm and bit4 communication error; Byte5 = checksum & 0xFF, where the checksum is 0xFF - the sum of Byte1 to Byte4.

4. The LED display screen intelligent monitoring, predictive maintenance, and adaptive scene arrangement system according to claim 1, characterized in that, The expansion baseboard of the central control host integrates the following physical interfaces: 4 RS485 interfaces, 1 RS232 interface, 4 relay output interfaces, 4 DI dry contact input interfaces, 2 Gigabit Ethernet interfaces, and 4 USB 2.0 HUBs.

5. The intelligent monitoring, predictive maintenance, and adaptive scene arrangement system for LED displays according to claim 1, characterized in that, The topic structure of the MQTT message bus adopts a four-level hierarchical namespace, including: The cloudnexus / {site_id} / {device_id} / sensor topic is used to publish sensor data every 3 seconds; The cloudnexus / {site_id} / {device_id} / status topic is used to report device status change information; The cloudnexus / {site_id} / {device_id} / command theme is used to issue control commands; The cloudnexus / {site_id} / alert topic is used for alarm broadcasting.

6. The LED display screen intelligent monitoring, predictive maintenance, and adaptive scene arrangement system according to claim 1, characterized in that, The formula for calculating the health score is as follows: ； ； ； ； in, Rate your health. Score for temperature. Score for voltage. Score based on runtime. Score historical faults. This is the current real-time operating temperature. This is the current real-time operating voltage. Rated voltage, For the cumulative number of operating days, This represents the number of failures that occurred in the past 30 days. Health status is classified as follows: 90≤HD≤100: Normal state; 70≤HD≤89: Under observation; 40≤HD≤69: Warning status; HD < 40: Alarm status.

7. The intelligent monitoring, predictive maintenance, and adaptive scene arrangement system for LED displays according to claim 1, characterized in that, The hierarchical automatic response mechanism is divided into three levels, as detailed below: The first level is the warning level, which involves reducing the brightness to a preset ratio, sending a notification, and outputting standardized investigation steps. The second level is the alarm level, which involves significantly reducing screen brightness, conducting AI root cause analysis, and pushing a screen-off suggestion. The third level is the protection level. When the screen temperature is detected to be greater than 85℃ or the power supply voltage is lower than 180V, the central control host directly issues a brightness reduction command locally. This protection action can be executed without relying on the cloud network. Among them, the brightness reduction protection command has a higher priority than all other control commands.

8. The intelligent monitoring, predictive maintenance, and adaptive scene arrangement system for LED displays according to claim 1, characterized in that, The Isolation Forest anomaly detection model uses a 9-dimensional feature vector, with the 9 features being, in order: mean temperature, standard deviation of temperature, rate of temperature change, mean voltage, standard deviation of voltage, rate of voltage change, mean photosensitive value, cumulative runtime, and historical fault count. The Isolation Forest anomaly detection model predicts four types of faults: It can predict power module failures 24 to 48 hours in advance with an accuracy rate of no less than 85%. It can predict accelerated decay of LED beads 1-2 weeks in advance with an accuracy rate of no less than 70%. Predicting receiver card anomalies 4-8 hours in advance with an accuracy rate of no less than 80%; It can predict fan blockage 12 to 24 hours in advance with an accuracy rate of no less than 75%.

9. The intelligent monitoring, predictive maintenance, and adaptive scene arrangement system for LED displays according to claim 1, characterized in that, The four-stage progressive cold start strategy includes four stages, P0-P3, as follows: P0 Basic Rules Phase: Day 1, based on the rule engine, the brightness adjustment of the photosensitive adaptive LED screen and the delayed exit standby mechanism; P1 Weekly Pattern Learning Phase: Day 7, Weekly Pattern Learning and Natural Language AI Report Based on Jaccard Similarity Coefficient; P2 parameter adaptation phase: Day 14, parameter adaptive optimization and gradual transition mechanism based on ambient light and ambient temperature; P3 Prediction and Warm-up Phase: On the 30th day, the prediction and scheduling of the activity probability in the next 2 hours is based on the AR autoregressive model, and the prediction and scheduling of automatic warm-up is set before the expected period.

10. The intelligent monitoring, predictive maintenance, and adaptive scene arrangement system for LED displays according to claim 1, characterized in that, It also includes a closed-loop user feedback mechanism, as detailed below: Users can choose to accept the AI ​​suggestions with one click, adjust them manually, or leave them unprocessed. Manual actions are recorded as user preferences after AI decisions are overridden. If the same type of suggestion is rejected repeatedly, it will not be triggered again within a set period. AI alerts can be flagged as false alarms, and the corresponding samples are stored in a negative sample pool for use in optimizing subsequent models.

Images