A millimeter wave dynamic diffractive coverage method based on collaborative control of ris and multi-modal occlusion prediction of IoT devices

Abstract

The application relates to a millimeter wave dynamic diffraction coverage method based on IoT device cooperative control of RIS and multi-modal shielding prediction, which comprises the following steps: collecting human body movement, door opening and closing, vehicle passing and obstacle displacement through an IoT device to perceive that a smart terminal has a shielding environment; feeding back channel data in real time through the smart terminal to determine that the smart terminal has a shielding risk, wherein the channel data comprises millimeter wave channel RSRP, signal strength, link error rate and interference state; two lightweight AI models based on a gateway deployed on an edge cloud device work cooperatively to output an optimal control strategy in real time; and an RIS device with a printed passive metasurface structure is deployed at one or more places in a 5G millimeter wave signal coverage blind area, and the RIS device is distributedly controlled by the IoT device nearby to reflect relevant millimeter wave signals to the smart terminal device. The present application has the problems of high hardware cost, no pre-judgment ability, lack of dynamic diffraction capability and inability to reuse stock resources.

Description

Technical Field

[0001] This invention relates to the field of millimeter-wave communication technology. In particular, it relates to a millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction. Background Technology

[0002] Millimeter-wave communication boasts advantages such as ultra-high bandwidth, ultra-low latency, and ultra-high transmission rates, making it a core technology for 5G millimeter-wave and 6G communication. However, millimeter waves have extremely short wavelengths, resulting in inherent defects such as weak penetration, poor diffraction, and susceptibility to obstruction by people, walls, doors, vehicles, and greenery. This leads to numerous signal blind spots in indoor corners, corridors, elevator entrances, and street canyons, as well as severe link breakage and jitter issues.

[0003] Existing millimeter-wave anti-massage solutions mainly fall into three categories: The first category is dedicated active repeaters and multi-base station relay networking. This solution requires a large number of new dedicated communication devices, resulting in high deployment costs, high power consumption, and complex networking, making it unsuitable for large-scale deployment. The second category is traditional intelligent metasurface (RIS) reflection enhancement solutions. Current RIS solutions rely on dedicated controllers and dedicated radio frequency sensing devices, making it impossible to reuse existing equipment resources. Moreover, they are mostly static reflection optimizations, unable to predict dynamic massage, and can only passively repair link breaks, resulting in poor resistance to dynamic massage. The third category is cloud-based AI path optimization solutions. These solutions rely on cloud computing power to complete path and beam strategy calculations, and then issue commands to adjust the electromagnetic phase of existing RIS hardware, thereby changing the signal propagation path. Due to the inherent latency of cloud data interaction, the system's real-time performance is insufficient, making it difficult to adapt to the millisecond-level link switching requirements in millimeter-wave dynamic massage scenarios. Furthermore, the long-term cloud computing power and bandwidth overhead are significant, making it unsuitable for large-scale deployment of massive edge IoT nodes.

[0004] In summary, existing technologies suffer from technical defects such as high hardware costs, reliance on specialized equipment, lack of predictive capabilities, lack of dynamic diffraction capabilities, and inability to reuse existing resources. The industry urgently needs a low-cost, fully reusable, predictive, and adaptive millimeter-wave dynamic diffraction coverage solution. Summary of the Invention

[0005] This invention provides a millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction. This addresses the problems of existing technologies, such as high hardware costs, reliance on dedicated equipment, lack of predictive capabilities, absence of dynamic diffraction capabilities, and inability to reuse existing resources.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction, for tracking a smart terminal with 5G millimeter-wave signal coverage blind zone, comprising: The smart terminal is detected by collecting data on human movement, door opening and closing, vehicle passage, and obstacle displacement through IoT devices. The smart terminal is then fed back channel data in real time to determine if there is a risk of obstruction. The channel data includes millimeter-wave channel RSRP, signal strength, link bit error rate, and interference status. Two lightweight AI models deployed on the edge cloud device work together: one lightweight AI model is a lightweight LSTM time-series prediction model, which predicts the probability and range of dynamic occlusion generation in the next 1 to 3 seconds based on the continuous visual frame sequence of the IoT device; the other lightweight AI model is a lightweight PPO reinforcement learning model, which uses link throughput, signal stability, device power consumption and occlusion risk as a comprehensive reward function to output the optimal control strategy in real time. The optimal control strategy includes RIS phase adjustment parameters, IoT relay enable / disable commands and millimeter-wave diffraction path switching strategies. A RIS device employing a printed passive metasurface structure is deployed in one or more locations within the 5G millimeter-wave signal coverage blind zone. It is distributedly controlled by nearby IoT devices. The distributed control includes: receiving the optimal control strategy output by the lightweight AI model to dynamically adjust the phase parameters of the metasurface unit to change the propagation direction of the millimeter-wave electromagnetic wave, thereby achieving directional reflection, diffraction, and filling of shadow blind zones, and reflecting the relevant millimeter-wave signal to the smart terminal device.

[0007] To achieve the above objectives, in a second aspect, the present invention also relates to a millimeter-wave dynamic diffraction coverage system for tracking a smart terminal carrying 5G millimeter waves when it operates in a 5G millimeter-wave signal coverage blind zone, comprising: The full-domain multimodal perception module is used to collect data on human movement, door opening and closing, vehicle passage, and obstacle displacement through IoT devices to sense if the smart terminal has an obstructed environment; and to determine if the smart terminal has an obstruction risk by feeding back channel data in real time through the smart terminal. The channel data includes millimeter-wave channel RSRP, signal strength, link bit error rate, and interference status. The edge cloud AI intelligent scheduling and decision-making module is used to coordinate two lightweight AI models deployed on the gateway of the edge cloud device: one lightweight AI model is a lightweight LSTM time-series prediction model, which predicts the probability and range of dynamic occlusion generation in the next 1 to 3 seconds based on the continuous visual frame sequence of the IoT device; the other lightweight AI model is a lightweight PPO reinforcement learning model, which uses link throughput, signal stability, device power consumption and occlusion risk as a comprehensive reward function to output the optimal control strategy in real time. The optimal control strategy includes RIS phase adjustment parameters, IoT relay enable / disable commands and millimeter-wave diffraction path switching strategies. An intelligent distributed control module is used in RIS devices employing printed passive metasurface structures, deployed in one or more locations within the 5G millimeter-wave signal coverage blind spots. Distributed control is performed by nearby IoT devices. The distributed control includes: receiving the optimal control strategy output by the lightweight AI model to dynamically adjust the phase parameters of the metasurface unit to change the propagation direction of the millimeter-wave electromagnetic wave, thereby achieving directional reflection, diffraction, and filling of shadow blind spots, and reflecting the relevant millimeter-wave signal to the intelligent terminal device.

[0008] To achieve the above objectives, in a third aspect, the present invention also relates to a computer-readable storage medium storing instructions that, when executed, perform the above-described millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction.

[0009] The present invention relates to a millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction, which has the following advantages compared with the prior art: This invention proposes a low-cost integrated solution that combines existing IoT device collaboration with lightweight edge AI and a passive RIS-based minimalist driver.

[0010] 1. For the first time, existing ordinary IoT devices can serve as both RIS sensing nodes and distributed controllers, eliminating the need for traditional dedicated RIS controllers, requiring zero new hardware deployment, and significantly reducing the cost of network replenishment in millimeter-wave blind spots. Significant cost advantages: Existing ordinary IoT devices are fully reused, eliminating the need for new dedicated RIS controllers and communication equipment. Passive RIS panels can be mass-produced at low cost, reducing overall deployment costs by over 80%.

[0011] 2. Employing multimodal AI occlusion prediction and pre-diffraction reconstruction, this approach breaks through the traditional passive link breakage repair mode, achieving path reconstruction even before occlusion occurs, thus completely resolving link jitter and link breakage issues caused by dynamic occlusion. Significantly improved anti-occlusion capability: Breaking through the traditional passive repair mode, it achieves AI-predictive pre-diffraction, combined with a three-mode hybrid diffraction mechanism, completely solving the problems of link breakage and blind zone coverage in millimeter-wave dynamic occlusion. 3. A three-mode hybrid diffraction mechanism of millimeter wave reflection-diffraction-IoT relay is proposed, which is different from the traditional single RIS reflection scheme. It adaptively adapts the diffraction strategy for different occlusion levels, has stronger blind zone coverage capability and wider application scenarios. It can be adapted to millimeter wave coverage in all scenarios such as power grid substations, power supply stations, and factory buildings. It is compatible with 5G millimeter wave and 6G metasurface communication systems and has broad industrialization prospects.

[0012] 4. Adopting a lightweight edge AI architecture, all decisions are completed locally without relying on the cloud, with millisecond-level response, low latency, low power consumption, no cloud transmission latency, and is suitable for low-power IoT networking scenarios and massive edge IoT scenarios. Attached Figure Description

[0013] Figure 1 This is a flowchart of a millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction according to the present invention.

[0014] Figure 2 This is a schematic diagram of an application scenario in a specific substation.

[0015] Figure 3 This is a diagram of the overall working mode architecture.

[0016] Figure 4 This is a schematic diagram of the structure of a millimeter-wave dynamic diffraction coverage system based on IoT device collaborative control RIS and multimodal occlusion prediction according to the present invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0018] Example 1 This invention discloses a millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction, used for tracking smart terminals operating in 5G millimeter-wave signal coverage blind spots. Please refer to [link to relevant documentation]. Figure 1-3 The algorithm in this embodiment is implemented based on a millimeter-wave system, including a switch, a 5G millimeter-wave base station, and edge cloud devices, wherein... Figure 3 Mid-edge cloud devices are implemented through electronic devices such as personal computers, host computers, servers, and smart terminals, typically through servers or server clusters. In one example, such as a 750kV substation in a power grid, the area is approximately 100,000 square meters (300m long and wide). 300m), see appendix for details. Figure 2In some areas, there are buildings that obstruct the view. In open areas, smart terminals such as inspection robots equipped with 5G millimeter wave modules can directly connect to 5G millimeter wave base stations. However, when the smart terminals operate in areas with buildings that obstruct the view, these areas are 5G millimeter wave signal coverage blind spots. In this case, it is necessary to use IoT devices such as cameras for real-time identification. Multiple cameras can identify moving smart terminals and track the moving smart terminals (inspection robots) in real time through the cameras in the area.

[0019] Step S1: Collect human movement, door opening and closing, vehicle passage and obstacle displacement through IoT devices to sense that the smart terminal has an obstructed environment; determine that the smart terminal has an obstruction risk by real-time feedback of channel data through the smart terminal, the channel data including millimeter wave channel RSRP, signal strength, link bit error rate and interference status.

[0020] Specifically, this step is full-domain multimodal perception: real-time synchronous collection of dynamic environmental video information through various existing IoT devices (such as the cameras of IoT devices), corresponding to the millimeter-wave channel status data of the inspection robot's 5G millimeter-wave module that has arrived at the location, and uploading it to the AI ​​model of the edge cloud device; Step S2: Two lightweight AI models deployed on the gateway of the edge cloud device work together: one lightweight AI model is a lightweight LSTM time-series prediction model, which predicts the probability and range of dynamic occlusion generation in the next 1 to 3 seconds based on the continuous visual frame sequence of the IoT device; the other lightweight AI model is a lightweight PPO reinforcement learning model, which uses link throughput, signal stability, device power consumption and occlusion risk as a comprehensive reward function to output the optimal control strategy in real time. The optimal control strategy includes RIS phase adjustment parameters, IoT relay enable / disable commands and millimeter wave diffraction path switching strategies. Specifically, this step is the edge cloud AI intelligent scheduling decision-making step. The lightweight LSTM time series prediction model predicts the occlusion trend and predicts the probability and range of dynamic occlusion generation in the next 1 to 3 seconds. Specifically, the lightweight LSTM time series prediction model of the edge cloud device analyzes the environmental dynamics in real time, predicts the short-term occlusion risk, and marks the millimeter wave line-of-sight links that are about to fail. The core of the model is to use Conv-LSTM to remember the spatiotemporal dependence of motion between frames, deduce the staggered temporal evolution of objects, and achieve multi-duration occlusion prediction in three steps: 1: Extracting spatial occlusion precursors in a single frame CNN extracts key features for each frame: relative distance to the target, direction of motion, relative velocity, spatial location of obstacles, and precursor pixels of local overlap.

[0021] 2: LSTM Temporal Memory Modeling (Core) The historical features of 10 frames are fed into a lightweight Conv-LSTM step by step, and the gating mechanism continuously updates the status. The forgetting gate filters static backgrounds and objects without moving interference. The input gate continuously stores temporal features of occlusion precursors such as inter-frame optical flow, target velocity changes, and objects gradually approaching and overlapping. The state preserves the long-term evolution trend of multi-target movement, distinguishing between transient interlaced movement patterns and continuous occlusion movement patterns.

[0022] 3: Multi-step future time series prediction (parallel output of 1 / 2 / 3 seconds): The LSTM's output at the final time step, incorporating all hidden features from historical memories, is fed into a multi-layer temporal decoding module to perform parallel occlusion predictions for three future durations, including: Short-term future 1 second: Based on linear extrapolation of instantaneous velocity, focusing on instantaneous local occlusion; Mid-term forecast (2 seconds): Integrating acceleration and changes in motion direction to predict large-area overlapping occlusion; Long-term future (3 seconds): Learn non-linear motions such as object deceleration / turning to predict long-distance occlusion; the decoding branch synchronously outputs the probability of occlusion occurrence and the pixel / box occlusion range for the corresponding duration, completing dynamic occlusion prediction.

[0023] Step S3: The RIS device with a printed passive metasurface structure is deployed in one or more locations in the 5G millimeter-wave signal coverage blind zone. It is distributedly controlled by the nearest IoT device. The distributed control includes: receiving the optimal control strategy output by the lightweight AI model to dynamically adjust the phase parameters of the metasurface unit to change the propagation direction of the millimeter-wave electromagnetic wave in order to achieve directional reflection, diffraction, and filling of shadow blind zones of the signal, and reflecting the relevant millimeter-wave signal to the smart terminal device.

[0024] Specifically, this step is an intelligent distributed control step, including S31-S34.

[0025] S31 uses a lightweight model combined with occlusion probability and channel quality to output optimal RIS phase parameters. IoT devices and camera devices adjust their respective external RIS diffraction antennas. Multiple RIS units dynamically adjust the metasurface unit phase, and relay scheduling and diffraction path planning are implemented.

[0026] Among them, the RIS device, which adopts a printed passive metasurface structure, is deployed in the corners of walls, corridors, elevator entrances and wall blind spots in the 5G millimeter wave signal coverage blind area. It reflects the wireless signal by being mounted on a pole or fixed to the wall.

[0027] S32 Distributed RIS Control: Nearby IoT devices, such as cameras, can replace dedicated controllers by sending phase control commands via the network line to drive the RIS metasurface to reconstruct the electromagnetic wave propagation path, thereby achieving millimeter-wave directional diffraction. S33, Adaptive Link Switching: Adaptively activates a hybrid mode of diffraction, reflection, and IoT multi-hop relay based on the level of occlusion, and completes the pre-establishment of a new diffraction link before the original link is interrupted, achieving seamless switching; S34. Real-time dynamic iteration: Continuously collect link status, and use AI to iterate and optimize strategies in real time to adapt to dynamic changes in the environment.

[0028] In this embodiment, as a preferred embodiment, it also includes S4, real-time dynamic iteration: continuously collecting the link status from S1 to S3, and using AI real-time iterative optimization strategies to adapt to dynamic changes in the environment.

[0029] Example 2 like Figure 4 As shown, a millimeter-wave dynamic diffraction coverage system is described in this embodiment. The algorithm system of this embodiment is based on a millimeter-wave system and includes a switch, a 5G millimeter-wave base station, and edge cloud devices. Some areas are obstructed by buildings. In open areas, intelligent terminals such as inspection robots with 5G millimeter-wave modules can directly connect to 5G millimeter-wave base stations. However, when intelligent terminals operate in areas with building obstructions, these areas become 5G millimeter-wave signal coverage blind spots. In such cases, real-time identification is required using IoT devices such as cameras. Multiple cameras can identify moving intelligent terminals, and the cameras in the area can track the moving intelligent terminals in real time. Figure 3 Mid-edge cloud devices are implemented through electronic devices such as personal computers, host computers, servers, and smart terminals, and are generally implemented through servers or server clusters. In this embodiment, a millimeter-wave dynamic diffraction coverage system also includes: a full-domain multimodal perception module 61, an edge cloud AI intelligent scheduling and decision-making module 62, and an intelligent distributed control module 63.

[0030] The full-domain multimodal perception module 61 is used to collect data on human movement, door opening and closing, vehicle passage, and obstacle displacement through IoT devices to sense that the smart terminal is in an obstructed environment; and to determine the risk of obstruction of the smart terminal by real-time feedback of channel data through the smart terminal. The channel data includes millimeter wave channel RSRP, signal strength, link bit error rate, and interference status. The edge cloud AI intelligent scheduling and decision-making module 62 is used to coordinate two lightweight AI models based on the gateway deployed on the edge cloud device: one lightweight AI model is a lightweight LSTM time-series prediction model, which predicts the probability and range of dynamic occlusion generation in the next 1 to 3 seconds based on the continuous visual frame sequence of IoT devices; the other lightweight AI model is a lightweight PPO reinforcement learning model, which uses link throughput, signal stability, device power consumption and occlusion risk as a comprehensive reward function to output the optimal control strategy in real time. The optimal control strategy includes RIS phase adjustment parameters, IoT relay enable / disable commands and millimeter wave diffraction path switching strategies. The intelligent distributed control module 63 is used in RIS devices employing printed passive metasurface structures, deployed in one or more locations within 5G millimeter-wave signal coverage blind spots. Distributed control is performed by nearby IoT devices. The distributed control includes: receiving the optimal control strategy output by a lightweight AI model to dynamically adjust the phase parameters of the metasurface unit to change the propagation direction of millimeter-wave electromagnetic waves, thereby achieving directional reflection, diffraction, and filling of shadow blind spots, and reflecting the relevant millimeter-wave signals to the intelligent terminal device.

[0031] In this embodiment, the edge cloud AI intelligent scheduling and decision-making module 62 uses a RIS device with a printed passive metasurface structure, which is deployed in the corners, corridors, elevator entrances and wall blind spots of the 5G millimeter wave signal coverage blind area. It reflects the wireless signal by sticking it on a pole or fixing it to the wall.

[0032] In this embodiment, the edge cloud AI intelligent scheduling decision module 62 predicts the probability and range of dynamic occlusion generation in the next 1 to 3 seconds based on the continuous visual frame sequence of IoT devices, specifically including: CNN extracts key features for each frame, including: relative distance to the target, direction of motion, relative velocity, spatial location of obstacles, and pre-overlapping pixels. LSTM temporal memory modeling: 10 frames of historical features are fed into a lightweight Conv-LSTM step by step, and the gating mechanism continuously updates the state; the forget gate filters static background and objects without moving interference; the input occlusion precursor temporal features include: the gate continuously stores inter-frame optical flow, target velocity changes and objects gradually approaching and overlapping; the state preserves the long-term trend of multi-target motion evolution, distinguishing between short-term interlaced motion patterns and continuous occlusion motion patterns; Multi-step future time series prediction: The output of the LSTM at the last time step is fused with hidden features from all historical memories and fed into a multi-layer time series decoding module to perform parallel extrapolation of occlusion predictions for three future durations, including: 1 second in the future: linear extrapolation based on instantaneous velocity, focusing on instantaneous local occlusion; 2 seconds in the future: incorporating acceleration changes and motion direction changes to predict large-area overlapping occlusion; 3 seconds in the future: learning the nonlinear motion of objects to predict long-distance merging occlusion. The decoding branch synchronously outputs the probability of occlusion occurring and the occlusion range for the corresponding duration, thus completing dynamic occlusion prediction.

[0033] In this embodiment, the directional diffraction of the signal in the intelligent distributed control module 63 specifically includes: the nearest IoT device sends a phase modulation command to the RIS device through the network line, driving the metasurface of the RIS device to reconstruct the electromagnetic wave propagation path, thereby realizing millimeter wave directional diffraction; In this embodiment, the real-time output optimal control strategy of the lightweight PPO reinforcement learning model in the edge cloud AI intelligent scheduling decision module 62 also includes: an adaptive link switching strategy: adaptively enabling diffraction, reflection, and IoT multi-hop relay hybrid mode according to the occlusion level, and completing the control strategy of pre-establishing a new diffraction link before the original link is interrupted.

[0034] The millimeter-wave dynamic diffraction coverage system of this embodiment is the same as the millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction described in Embodiment 1, and will not be repeated here.

[0035] Example 3 This invention relates to a computer-readable storage medium storing instructions. When the instructions are executed, they perform a millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction, as described in Embodiment 1. The execution process and effects are the same as those of the millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction described in Embodiment 1, and will not be repeated here.

[0036] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0037] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction, characterized in that, Tracking of a 5G millimeter wave-enabled smart terminal when it operates in a 5G millimeter wave signal coverage blind spot includes: The smart terminal is detected by collecting data on human movement, door opening and closing, vehicle passage, and obstacle displacement through IoT devices. The smart terminal is then fed back channel data in real time to determine if there is a risk of obstruction. The channel data includes millimeter-wave channel RSRP, signal strength, link bit error rate, and interference status. Two lightweight AI models deployed on the edge cloud device work together: one lightweight AI model is a lightweight LSTM time-series prediction model, which predicts the probability and range of dynamic occlusion generation in the next 1 to 3 seconds based on the continuous visual frame sequence of the IoT device; the other lightweight AI model is a lightweight PPO reinforcement learning model, which uses link throughput, signal stability, device power consumption and occlusion risk as a comprehensive reward function to output the optimal control strategy in real time. The optimal control strategy includes RIS phase adjustment parameters, IoT relay enable / disable commands and millimeter-wave diffraction path switching strategies. A RIS device employing a printed passive metasurface structure is deployed in one or more locations within the 5G millimeter-wave signal coverage blind zone. It is distributedly controlled by nearby IoT devices. The distributed control includes: receiving the optimal control strategy output by the lightweight AI model to dynamically adjust the phase parameters of the metasurface unit to change the propagation direction of the millimeter-wave electromagnetic wave, thereby achieving directional reflection, diffraction, and filling of shadow blind zones, and reflecting the relevant millimeter-wave signal to the smart terminal device.

2. The millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction according to claim 1, characterized in that, The RIS device, which employs a printed passive metasurface structure, is deployed in corners, corridors, elevator entrances, and wall blind spots in the 5G millimeter-wave signal coverage blind areas. It reflects wireless signals by being mounted on poles or fixed to the wall.

3. The millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction according to claim 2, characterized in that, Based on the continuous visual frame sequence of the IoT device, the probability of dynamic occlusion generation and the occlusion range in the next 1 to 3 seconds are predicted, specifically including: CNN extracts key features for each frame, including: relative distance to the target, direction of motion, relative velocity, spatial location of obstacles, and pre-overlapping pixels. LSTM temporal memory modeling: 10 frames of historical features are fed into a lightweight Conv-LSTM step by step, and the state is continuously updated by a gating mechanism; the forget gate filters static backgrounds and objects without moving interference; the input occlusion precursor temporal features include: the input gate continuously stores inter-frame optical flow, target velocity changes, and objects gradually approaching and overlapping; the state is preserved for a long time to show the evolution trend of multi-target motion, distinguishing between short-term interlaced motion patterns and continuous occlusion motion patterns; Multi-step future time series prediction: The LSTM's output at the last time step incorporates hidden features from all historical memories and is fed into a multi-layer time series decoding module to perform parallel extrapolation of occlusion predictions for three future durations, including: 1 second in the future: linear extrapolation based on instantaneous velocity, focusing on instantaneous local occlusion; 2 seconds in the future: incorporating acceleration changes and motion direction changes to predict large-area overlapping occlusion; 3 seconds in the future: learning the nonlinear motion of objects to predict long-distance confluence occlusion; the decoding branch synchronously outputs the probability of occlusion occurrence and occlusion range for the corresponding duration, completing dynamic occlusion prediction.

4. The millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction according to claim 1, characterized in that, The directional diffraction of the signal specifically includes: a nearby IoT device sending a phase modulation command to the RIS device via a network line, driving the RIS device's metasurface to reconstruct the electromagnetic wave propagation path, thereby achieving millimeter-wave directional diffraction.

5. A millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction according to claim 1, characterized in that, The real-time output optimal control strategy of the lightweight PPO reinforcement learning model also includes: an adaptive link switching strategy: based on the occlusion level, an adaptive control strategy of enabling a hybrid mode of diffraction, reflection, and IoT multi-hop relay is adopted to complete the pre-establishment of a new diffraction link before the original link is interrupted.

6. A millimeter-wave dynamic diffraction coverage system based on IoT device collaborative control RIS and multimodal occlusion prediction, characterized in that, Tracking of a 5G millimeter wave-enabled smart terminal when it operates in a 5G millimeter wave signal coverage blind spot includes: The full-domain multimodal perception module is used to collect data on human movement, door opening and closing, vehicle passage, and obstacle displacement through IoT devices to sense if the smart terminal has an obstructed environment; and to determine if the smart terminal has an obstruction risk by feeding back channel data in real time through the smart terminal. The channel data includes millimeter-wave channel RSRP, signal strength, link bit error rate, and interference status. The edge cloud AI intelligent scheduling and decision-making module is used to coordinate two lightweight AI models deployed on the gateway of the edge cloud device: one lightweight AI model is a lightweight LSTM time-series prediction model, which predicts the probability and range of dynamic occlusion generation in the next 1 to 3 seconds based on the continuous visual frame sequence of the IoT device; the other lightweight AI model is a lightweight PPO reinforcement learning model, which uses link throughput, signal stability, device power consumption and occlusion risk as a comprehensive reward function to output the optimal control strategy in real time. The optimal control strategy includes RIS phase adjustment parameters, IoT relay enable / disable commands and millimeter-wave diffraction path switching strategies. An intelligent distributed control module is used in RIS devices employing printed passive metasurface structures, deployed in one or more locations within the 5G millimeter-wave signal coverage blind spots. Distributed control is performed by nearby IoT devices. The distributed control includes: receiving the optimal control strategy output by the lightweight AI model to dynamically adjust the phase parameters of the metasurface unit to change the propagation direction of the millimeter-wave electromagnetic wave, thereby achieving directional reflection, diffraction, and filling of shadow blind spots, and reflecting the relevant millimeter-wave signal to the intelligent terminal device.

7. A millimeter-wave dynamic diffraction coverage system based on IoT device collaborative control RIS and multimodal occlusion prediction according to claim 6, characterized in that, In the edge cloud AI intelligent scheduling and decision-making module, a RIS device with a printed passive metasurface structure is deployed in the corners, corridors, elevator entrances, and wall blind spots of the 5G millimeter wave signal coverage blind area. It reflects the wireless signal by means of a pole or fixed to the wall.

8. A millimeter-wave dynamic diffraction coverage system based on IoT device collaborative control RIS and multimodal occlusion prediction according to claim 6, characterized in that, In the edge cloud AI intelligent scheduling and decision-making module, based on the continuous visual frame sequence of the IoT device, the probability and range of dynamic occlusion generation in the next 1 to 3 seconds are predicted, specifically including: CNN extracts key features for each frame, including: relative distance to the target, direction of motion, relative velocity, spatial location of obstacles, and pre-overlapping pixels. LSTM temporal memory modeling: 10 frames of historical features are fed into a lightweight Conv-LSTM step by step, and the gating mechanism continuously updates the state; the forget gate filters static background and objects without moving interference; the input occlusion precursor temporal features include: the gate continuously stores inter-frame optical flow, target velocity changes and objects gradually approaching and overlapping; the state preserves the long-term trend of multi-target motion evolution, distinguishing between short-term interlaced motion patterns and continuous occlusion motion patterns; Multi-step future time series prediction: The output of the LSTM at the last time step is fused with hidden features from all historical memories and fed into a multi-layer time series decoding module to perform parallel extrapolation of occlusion predictions for three future durations, including: 1 second in the future: linear extrapolation based on instantaneous velocity, focusing on instantaneous local occlusion; 2 seconds in the future: incorporating acceleration changes and motion direction changes to predict large-area overlapping occlusion; 3 seconds in the future: learning the nonlinear motion of objects to predict long-distance merging occlusion. The decoding branch synchronously outputs the probability of occlusion occurring and the occlusion range for the corresponding duration, thus completing dynamic occlusion prediction.

9. A millimeter-wave dynamic diffraction coverage system based on IoT device collaborative control RIS and multimodal occlusion prediction according to claim 6, characterized in that, In the intelligent distributed control module, the implementation of directional diffraction of the signal specifically includes: the nearest IoT device sends a phase modulation command to the RIS device through the network line, driving the metasurface of the RIS device to reconstruct the electromagnetic wave propagation path, thereby realizing millimeter wave directional diffraction; In the edge cloud AI intelligent scheduling decision module, the real-time output optimal control strategy of the lightweight PPO reinforcement learning model also includes: an adaptive link switching strategy: based on the occlusion level, adaptively enable the diffraction, reflection, and IoT multi-hop relay hybrid mode, and complete the pre-establishment of a new diffraction link before the original link is interrupted.

10. A computer-readable storage medium, characterized in that: The storage medium stores instructions that, when executed, perform a millimeter-wave dynamic diffraction coverage method based on IoT device collaborative control RIS and multimodal occlusion prediction as described in any one of claims 1-5.

Images