An intelligent cabinet monitoring method and system based on cloud-edge collaboration

Abstract

The application discloses a kind of intelligent cabinet monitoring method and system based on cloud edge cooperation, it is related to Internet of Things technical field, and task number is obtained by decomposing monitoring task, according to the key rating rule of preestablished, monitoring task is divided into key chain task and non key chain task;Through interference prediction model and PID controller, real-time prediction intelligent cabinet edge gateway resource contention influence, obtain resource contention state, using MQMSA algorithm on-line adjustment PID parameter, and the front node of key chain task is reserved in buffer area;Using multi-agent deep deterministic policy feature model, obtain spare key path, and migrate key chain task to spare key path;Non key chain task is merged into the node corresponding to key chain task, obtain merging buffer consumption value and judge, and start edge trigger mechanism;Build the digital twin model of intelligent cabinet cluster, real-time update task flow path, corresponding resource contention state and edge trigger mechanism.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, and in particular to a smart cabinet monitoring method and system based on cloud-edge collaboration. Background Technology

[0002] With the development of IoT, edge computing and AI technologies, smart cabinets not only need to monitor internal environmental parameters in real time to ensure the safety of equipment and assets, but also need to undertake increasingly complex edge computing tasks to improve the service interaction experience.

[0003] Smart lockers are typically deployed at the network edge, where the computing resources of the edge gateways are relatively limited. When faced with high-concurrency computing tasks such as facial recognition for package pickup during peak hours, resource contention is likely to occur, leading to the blockage of the underlying environmental monitoring process and the inability to report critical alarm data in a timely manner, posing serious security risks. In addition, the complex electromagnetic environment causes monitoring distortion. High-frequency electromagnetic interference not only affects the accuracy of weak signal acquisition by sensors but also degrades the signal-to-noise ratio of wireless communication links, resulting in data packet loss and false alarms. Existing technologies often lack effective interference prediction and proactive avoidance mechanisms, making it difficult to distinguish between real faults and environmental noise. To ensure the reliability of data transmission, existing smart lockers are usually equipped with both wired and wireless links. However, traditional routing scheduling strategies are usually based on a primary-backup switchover mode, meaning that the system only switches to the backup link when the primary link is completely disconnected. The switchover process is often accompanied by several seconds of service interruption. In addition, once the switchover occurs, all service traffic floods into the primary link, causing a surge in traffic costs. There is a lack of a fine-grained traffic distribution mechanism based on service priority. Finally, in the event of extreme disaster scenarios such as a complete network outage, existing systems often lack effective data disaster recovery and breakpoint resume mechanisms, resulting in the loss of critical on-site data before and after the failure, making it impossible to trace the fault source and determine responsibility afterward. Summary of the Invention

[0004] The technical problem solved by this invention is that existing technologies have difficulty solving the problem of monitoring task blockage caused by limited resources on the edge side, low reliability transmission and high cost operation of communication links, and difficulty in data traceability under extreme working conditions.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: Firstly, a smart cabinet monitoring method based on cloud-edge collaboration includes the following steps: Step S1: Based on the monitoring requirements of the smart cabinet, the monitoring tasks are decomposed, task numbers are generated, and the monitoring tasks are divided into critical chain tasks and non-critical chain tasks according to the preset criticality rating rules. Step S2: Through the interference prediction model and PID controller, the impact of resource contention on the edge gateway of the smart cabinet is predicted in real time. The resource contention status is predicted. Based on the internal environmental parameters of the smart cabinet, the MQMSA algorithm is used to adjust the PID parameters online. Buffer reservations are made for the front nodes of the critical chain tasks based on the resource contention status. Step S3: Based on the current resource status of the smart cabinet, use the multi-agent deep deterministic strategy feature model to predict the backup critical path, and migrate the critical chain task to the backup critical path. Step S4: Inject non-critical chain tasks into the nodes corresponding to critical chain tasks to obtain the injection buffer consumption value. Use the buffer judgment mechanism to judge the injection buffer consumption value and start the edge triggering mechanism. Step S5: Construct a digital twin model of the smart cabinet cluster and update the task flow path, corresponding resource contention status, and edge triggering mechanism in real time.

[0006] Preferably, step S1 includes the following sub-steps: Step S11: Through the operation of the smart cabinet terminal, scan all currently running processes and decompose them into monitoring tasks. Use a distributed ID generation algorithm to generate a task number for each monitoring task and embed the task number into the header of the data packet of the monitoring task. The task number is a unique index for the entire life cycle. Step S12: Preset criticality rating rules, and determine the monitoring tasks according to the criticality rating rules to obtain critical chain tasks and non-critical chain tasks. The critical chain tasks include early warning of environmental changes, theft prevention of core assets, and handling of external emergencies. The non-critical chain tasks include routine operation and maintenance data collection, user behavior analysis, and commercial content delivery.

[0007] Preferably, step S2 includes the following sub-steps: Step S21: Deploy an interference prediction model based on time series analysis on the edge gateway. The interference prediction model collects the CPU load, memory usage and electromagnetic noise floor data of the gateway in real time as the current running data. Based on the historical running data and the current running data, a time window is set to predict the resource contention state of the next time window. The resource contention state includes no contention state, computing resource contention state and electromagnetic interference contention state. Step S22: The internal environmental parameters of the intelligent cabinet are used as the controlled variables. The MQMSA algorithm is used to optimize the proportional coefficient, integral coefficient and derivative coefficient of the PID controller in real time. The MQMSA algorithm uses environmental stability and actuator energy consumption as a joint fitness function, outputs the optimal control parameters, and drives the actuator to perform internal control within the cabinet. Step S23: Based on the predicted resource contention status of the next time window, reserve buffers for the critical task front nodes of the edge gateway. When a computational resource contention state is predicted, the edge gateway immediately allocates a buffer, allowing only critical chain tasks to access it. When the resource contention state is in the computational resource contention state, the edge gateway immediately allocates a buffer, allowing access to the buffer when the task number corresponds to a critical chain task, and prohibiting access to the buffer when the task number corresponds to a non-critical chain task.

[0008] Preferably, step S21 includes the following sub-steps: Step S211: Collect current running data according to the current sampling frequency, and at the same time retrieve historical running data within a preset time period from the local database. The historical running data includes CPU load, memory usage, electromagnetic environment noise floor data, and historical statistical features. The historical statistical features include historical mean, historical variance, historical change slope, historical peak value, and characteristic patterns of resource contention events. Step S212: Set a time window of length T, and align and merge the current running data with the historical running data: Within the current time window, the statistical characteristics of each indicator of the running data are calculated. The statistical characteristics include mean, variance, slope of change and number of peaks. The statistical characteristics of the current time window and the historical statistical characteristics are concatenated to obtain a multidimensional feature vector. Step S213: Input the multidimensional feature vector into the interference prediction model, perform time series analysis, and output the resource contention status of the next time window.

[0009] Preferably, step S3 includes the following sub-steps: The edge gateway collects transmission parameter packets of each communication interface in real time. The transmission parameter packets include a first status parameter and a second status parameter. The first status parameter includes latency jitter, signal quality attenuation rate and retransmission request count, which are used to indicate the link health level. The second status parameter includes task number and corresponding monitoring task, which are used to indicate task priority. Latency jitter is the degree of dispersion in the time interval between consecutive data packets, used to indicate the trend of link congestion; The signal quality attenuation rate is the slope of the change in signal strength or signal-to-noise ratio per unit time, used to represent the degree of deterioration of the physical environment; The retransmission request count is the number of data packets for which no acknowledgment response has been received, used to indicate packet loss in the link. Based on the first state parameters, the predicted failure probability of the current primary communication link is calculated, and a first migration cost function is constructed by combining the cost coefficient of the backup communication link: ; in, This is the predicted failure probability value. This is the cost coefficient. This is the result of the first migration cost function; When any one of the first state parameters is greater than a preset security threshold, and the result of the first migration cost function is greater than the handover determination value, the main communication link is determined to be in an unhealthy state, and the edge gateway executes the first scheduling strategy. When the physical connection of the main communication link is broken, the main communication link is determined to be in a fault state. The first scheduling strategy includes: Parse the transmission parameter packet and identify the task number in the second state parameter. When the task number is a critical chain task, mark the transmission parameter packet with high priority. Extract the transmission parameter packets corresponding to the high-priority coloring marks, and forcibly modify the physical exit to the backup communication link interface. Transmit the remaining transmission parameter packets through the main communication link. When the main communication link fails, temporarily store them in the local cache queue.

[0010] Preferably, step S4 includes the following sub-steps: Step S41: Set up an inflow buffer pool before non-critical chain tasks are inflowed into the main communication link, preset a first threshold and a second threshold, and monitor the buffer occupancy rate of non-critical chain tasks in real time. When the buffer occupancy rate is less than the first threshold, non-critical chain tasks are allowed to transmit normally. When the first threshold < buffer occupancy rate < second threshold, the edge gateway initiates a traffic shaping strategy to reduce the transmission priority of non-critical chain tasks; When the buffer occupancy rate exceeds the second threshold, temporarily discard the lowest priority non-critical chain tasks. Step S42: Set up a project buffer pool, which is a local non-volatile storage space. Write the alarm data and core status snapshots generated by the critical chain task into the project buffer pool. Step S43: When the project buffer pool is detected to be consumed, a local audible and visual alarm is triggered on the edge side. After the network connection is restored, the edge gateway automatically initiates a data reconciliation process with the cloud to retransmit the critical data temporarily stored in the project buffer pool.

[0011] Preferably, step S5 specifically includes: A digital twin model of the smart cabinet is built on a cloud server. Monitoring tasks are received in real time and sent to the corresponding virtual package entity in the digital twin model. The task is marked with color according to the task number to obtain task flow data. The resource contention status is mapped to the edge gateway component of the digital twin model using a heat map to obtain resource status data. Dynamically rendered link lines are constructed, and the main communication link and backup communication link are virtually topologically mapped to obtain link topology data. The system tracks the flow of critical chain tasks in the virtual topology in real time. When a path migration is triggered, the digital twin model synchronously displays the dynamic process of the virtual package entity instantly jumping from the main link line to the backup link line and shows the buffer occupancy rate of the buffer pool and project buffer pool in a dynamic water level map.

[0012] Preferably, the processing logic for inputting the multidimensional feature vector into the interference prediction model, performing time series analysis, and outputting the resource contention state for the next time window is as follows: An interference prediction model is constructed based on a selective state-space model. Multidimensional feature vectors are input into the pre-projection layer of the interference prediction model. Using the zero-order preservation discretization rule, the continuous-time multidimensional feature vectors are transformed into a discretized input sequence through a time-scale parameter. The input sequence is mapped to a high-dimensional latent state space through a state equation. The state equation is as follows: ; in, and The system matrix and input matrix are discretized. Given the input sequence at time k, Let k be the state space at time k-1, i.e., the potential state space. Let k be the state space at time k; A parallel correlation scanning algorithm is used to selectively scan the input sequence. The latent state space is mapped back to the output space through the output matrix, and state decoding is performed to obtain a prediction vector of future resource features. The prediction vector is processed through a root mean square normalization layer and a residual connection layer, and finally input into a fully connected layer. The output is a probability distribution vector processed by the Softmax activation function. The probability distribution vector includes the prediction probability of a contention-free state, the prediction probability of a resource contention state, and the prediction probability of an electromagnetic interference contention state. The sum of each sample in the probability distribution vector is 1. Based on the probability distribution vector, the resource contention state of the next time window is obtained using the maximum a posteriori probability criterion.

[0013] Preferably, the expression for calculating the switching determination value is: ; in, To switch the judgment value, This is the moving average of the results of the first migration cost function over a preset time period. Let be the standard deviation of the result of the first migration cost function, and k be the sensitivity coefficient. This is the load adjustment factor.

[0014] Secondly, a cloud-edge collaborative intelligent cabinet monitoring system includes a task identification module, an edge resource allocation module, a critical chain planning module, a non-critical chain incorporation module, and a digital twin module. The task identification module is used to decompose the monitoring requirements of the smart cabinet into tasks, generate a full life cycle task number using a distributed algorithm, and divide the decomposed monitoring tasks into critical chain tasks and non-critical chain tasks based on preset criticality rating rules. The edge resource allocation module is used to predict the impact of resource contention on the edge gateway of the smart cabinet in real time through the interference prediction model and PID controller, predict the resource contention status, adjust the PID parameters online using the MQMSA algorithm based on the internal environmental parameters of the smart cabinet, and reserve buffers for the front nodes of critical chain tasks based on the resource contention status. The critical chain planning module is used to monitor the current communication resource status of the smart cabinet in real time, run a multi-agent deep deterministic strategy feature model, automatically plan backup critical paths, and migrate critical chain tasks to backup critical paths. The non-critical chain inbound module is used to inbound non-critical chain tasks into the nodes corresponding to critical chain tasks, obtain inbound buffer consumption values, judge the inbound buffer consumption values ​​through a buffer judgment mechanism, and start the edge triggering mechanism. The digital twin module is used to construct a digital twin model of the smart cabinet cluster and update the task flow path, corresponding resource contention status and edge triggering mechanism in real time. The beneficial effects of this invention are as follows: By integrating task lifecycle management, edge-side interference prediction and autonomous control, dynamic routing with multi-dimensional state awareness, and a two-level buffer disaster recovery mechanism, the reliability and intelligence level of the intelligent cabinet monitoring system in complex environments are significantly improved. Firstly, this invention utilizes distributed tags and criticality rating rules to achieve atomic decomposition and full-link tracing of monitoring tasks, ensuring that critical chain tasks involving life and property safety enjoy the highest priority processing channel under any resource-constrained or network congestion conditions, eliminating delays and loss of core alarm data. Secondly, this invention innovatively introduces an interference prediction model based on time series analysis and an MQMSA-optimized PID control algorithm to anticipate the risks of computing resource contention and electromagnetic interference within future time windows, and dynamically reserves state buffers, effectively resolving... This invention solves the problem of blocked underlying monitoring processes in high-concurrency scenarios, while ensuring the stability of the cabinet environment control and energy saving of actuators. It constructs a micro-routing decision mechanism based on cost functions, abandoning the traditional network outage switching mode. By calculating the probability of link failure and cost in real time, it triggers packet-level redirection when the main link shows unhealthy signs. Combined with traffic coloring technology, it seamlessly migrates only critical business to the backup link, significantly reducing traffic costs. Finally, this invention designs a two-level disaster recovery mechanism of inflow buffer and project buffer and a digital twin closed-loop feedback system. It uses local non-volatile storage to save key field data at the moment of extreme network and power outages, and optimizes the system threshold through virtual simulation. It realizes a complete closed loop of monitoring data throughout the entire life cycle and the traceability of faults, significantly reducing the operation and maintenance threshold and labor costs. Attached Figure Description

[0015] Figure 1 A flowchart illustrating the steps of a cloud-edge collaborative smart cabinet monitoring method according to an embodiment of the present invention; Figure 2 This is a basic flowchart of a cloud-edge collaborative intelligent cabinet monitoring system provided as an embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example 1, referring to Figure 1 This paper provides a smart cabinet monitoring method based on cloud-edge collaboration, which includes the following steps: Step S1: Based on the monitoring requirements of the smart cabinet, the monitoring tasks are decomposed, task numbers are generated, and the monitoring tasks are divided into critical chain tasks and non-critical chain tasks according to the preset criticality rating rules. Step S2: Through the interference prediction model and PID controller, the impact of resource contention on the edge gateway of the smart cabinet is predicted in real time. The resource contention status is predicted. Based on the internal environmental parameters of the smart cabinet, the MQMSA algorithm is used to adjust the PID parameters online. Buffer reservations are made for the front nodes of the critical chain tasks based on the resource contention status. Step S3: Based on the current resource status of the smart cabinet, use the multi-agent deep deterministic strategy feature model to predict the backup critical path and migrate the critical chain task to the backup critical path. Step S4: Inject non-critical chain tasks into the nodes corresponding to critical chain tasks to obtain the injection buffer consumption value. Use the buffer judgment mechanism to judge the injection buffer consumption value and start the edge triggering mechanism. Step S5: Construct a digital twin model of the smart cabinet cluster and update the task flow path, corresponding resource contention status, and edge triggering mechanism in real time.

[0018] In this embodiment, the smart cabinet terminal operating system scans all currently running processes and decomposes the monitoring requirements into the following specific tasks: environmental change monitoring (sudden temperature rise), core asset anti-theft monitoring (abnormal opening of cabinet doors), user behavior analysis (collection frequency statistics), and commercial content push (advertisement playback). A distributed ID generation algorithm, such as the snowflake algorithm, is used to generate a 32-bit globally unique task number for each monitoring task, and the task number is embedded in the header of the monitoring task's data packet. The determination is made based on the preset key rating rules: Environmental change warning and core asset theft prevention are classified as critical chain tasks; routine operation and maintenance data collection, user behavior analysis, and commercial content push are classified as non-critical chain tasks. The rating rules are set as follows: tasks involving personal safety, asset security, or requiring sub-second response are classified as critical chain tasks, and the rest are classified as non-critical chain tasks. An interference prediction model based on time series analysis is deployed on the edge gateway. The CPU load, memory usage and electromagnetic noise data of the gateway are collected in real time. A time window of 5 seconds is set, and the current running data is aligned and fused with the historical running data of the past 1 hour. The statistical features are calculated and then input into the interference prediction model. The statistical features include mean, variance and slope of change.

[0019] The interference prediction model is built on a selective state-space model and outputs the resource contention state for the next time window. When a computational resource contention state (CPU load ≥ 70% or memory utilization ≥ 80%) is predicted, the edge gateway immediately allocates a buffer at the kernel layer, reserving 30% of the CPU time slice and memory space. When the task number corresponds to a critical chain task, access to the buffer is allowed; non-critical chain tasks are prohibited from accessing it. Meanwhile, internal environmental parameters are collected by temperature and humidity sensors inside the cabinet as controlled variables. The MQMSA algorithm is used to optimize the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller in real time. Environmental stability and actuator energy consumption are used as a joint fitness function. After outputting the optimal control parameters, the refrigeration or heating module is driven to perform temperature and humidity regulation inside the cabinet to ensure the stability of the storage environment for fresh food packages. The edge gateway collects transmission parameter packets of each communication interface in real time, including first status parameters (latency jitter, signal quality attenuation rate, retransmission request count) and second status parameters (task number and corresponding monitoring task). Based on the first status parameters, it calculates the predicted failure probability of the current main communication link and constructs a migration cost function by combining the cost coefficient of the backup communication link. When the latency jitter in the first state parameter exceeds 50ms and the migration cost function result is greater than the handover judgment value, the main communication link is determined to be in an unhealthy state. At this time, the edge gateway parses the transmission parameter packet, identifies the task number corresponding to the critical chain task, applies high-priority coloring to it, and forces the colored data packet to be transmitted through the backup communication link. The data packets of other non-critical chain tasks continue to be transmitted through the main link. Set up an inflow buffer pool before non-critical chain tasks are inflow into the main communication link, and set up a project buffer pool in the local non-volatile storage space. When a network interruption is detected, write the alarm data and core status snapshot generated by the critical chain task into the project buffer pool, and trigger the local audible and visual alarm on the edge side. After the network connection is restored, the edge gateway automatically initiates a data reconciliation process with the cloud and retransmits the critical data temporarily stored in the project buffer pool. Finally, a digital twin model of the smart cabinet cluster is built on the cloud server. Monitoring tasks are received in real time and sent to the corresponding virtual package entities in the digital twin model. The tasks are distinguished and marked by color according to their numbers. Dynamically rendered link lines are constructed, and the main communication link and backup communication link are mapped in a virtual topology. The flow position of critical chain tasks in the virtual topology is tracked in real time. When a path migration is triggered, the digital twin model synchronously displays the dynamic process of the virtual package entity jumping from the main link line to the backup link line. The buffer occupancy rate of the buffer pool and project buffer pool is displayed in a dynamic water level chart. The operators can monitor the task flow status and resource health status of the entire cluster in real time through a visual dashboard.

[0020] Step S1 includes the following sub-steps: Step S11: Through the operation of the smart cabinet terminal, scan all currently running processes and decompose them into monitoring tasks. Use the distributed ID generation algorithm to generate a task number for each monitoring task and embed the task number into the header of the data packet of the monitoring task. The task number is a unique index for the entire life cycle. Step S12: Preset criticality rating rules, and determine the monitoring tasks according to the criticality rating rules to obtain critical chain tasks and non-critical chain tasks. Critical chain tasks include early warning of environmental changes, theft prevention of core assets, and handling of unexpected external events; Non-critical chain tasks include routine operation and maintenance data collection, user behavior analysis, and commercial content delivery.

[0021] This embodiment primarily addresses the technical challenges of chaotic monitoring data and indistinguishable priorities in complex smart cabinet business scenarios. By introducing a unique identifier (task number) throughout the entire lifecycle and a criticality rating mechanism, it achieves refined management and differentiated scheduling of monitoring tasks. The specific implementation process is as follows: First, the smart cabinet terminal operating system starts a monitoring daemon process, which periodically (e.g., every 5 seconds) scans all currently running processes and services in the system kernel and application layer. Based on a preset monitoring task decomposition table, the daemon process decomposes the scanned macro-business processes into indivisible monitoring tasks. For example, the environmental monitoring service is decomposed into temperature acquisition tasks, humidity acquisition tasks, and smoke detection tasks. For each monitoring task, the system calls a distributed ID generation algorithm (such as Snowflake algorithm or UUIDv4), and combines the current timestamp, machine MAC address, and serial number to generate a 64-bit or 128-bit globally unique task number (Trace ID). The task number is embedded as a custom field into the header of all data packets generated by the monitoring task (e.g., MQTT protocol Topic or HTTP request header). This task number runs through the entire lifecycle of data acquisition, edge processing, network transmission, and cloud storage, serving as a unique index key for end-to-end tracing. A pre-defined criticality rating rule table is used, which sets different weight thresholds. The monitoring tasks generated in step S11 are iterated through, and the business type field is matched against the criticality rating rule table for determination. Critical Chain Task Determination: Environmental change warning category: rapid temperature rise task with a temperature change rate exceeding 2℃ / min, humidity change task, smoke concentration exceeding the standard alarm task; Core asset anti-theft measures: cabinet door unauthorized opening monitoring tasks, vibration / displacement sensor triggered tasks, and camera obstruction alarm tasks; External emergency events include: water immersion sensor triggering tasks, mains power outage alarm tasks, and network link physical disconnection tasks. Non-critical chain task determination: Routine maintenance data collection includes tasks such as timed temperature and humidity monitoring every 10 minutes, device heartbeat data collection, and fan speed log collection. User behavior analysis tasks include user screen click heatmap tasks and pickup code input error retry log tasks. Commercial content push: This includes tasks such as downloading advertising materials and downloading firmware upgrade packages (non-emergency security patches). This embodiment solves the problem of data loss of context information after passing through multiple nodes in traditional monitoring systems. By checking the task number, the processing status and time of the task at any time and any node can be quickly restored, greatly shortening the fault location time. In addition, this embodiment avoids low-value data from occupying valuable computing and network resources by separating non-critical tasks such as routine maintenance data and advertising push from the core processing path.

[0022] Step S2 includes the following sub-steps: Step S21: Deploy an interference prediction model based on time series analysis on the edge gateway. The interference prediction model collects the CPU load, memory usage and electromagnetic noise floor data of the gateway in real time as the current running data. Based on the historical running data and the current running data, a time window is set to predict the resource contention state of the next time window. The resource contention state includes no contention state, computing resource contention state and electromagnetic interference contention state. Step S22: The internal environmental parameters of the intelligent cabinet are used as the controlled variables. The MQMSA algorithm is used to optimize the proportional coefficient, integral coefficient and derivative coefficient of the PID controller in real time. The MQMSA algorithm uses environmental stability and actuator energy consumption as a joint fitness function, outputs the optimal control parameters, and drives the actuator to perform the cabinet control. Step S23: Based on the predicted resource contention status of the next time window, reserve buffers for the critical task front nodes of the edge gateway. When a computational resource contention state is predicted, the edge gateway immediately allocates a buffer, allowing only critical chain tasks to access it. When the resource contention state is in the computational resource contention state, the edge gateway immediately allocates a buffer, allowing access to the buffer when the task number corresponds to a critical chain task, and prohibiting access to the buffer when the task number corresponds to a non-critical chain task.

[0023] In this embodiment, by deploying an interference prediction model and utilizing time series analysis technology to anticipate resource bottlenecks within the next 10 seconds, dedicated CPU and memory buffers are pre-locked for critical chain tasks before computing resources are exhausted. This avoids the risk of critical alarm data being blocked or discarded due to sudden high loads in the AI-identified concurrent system, reducing the latency of critical services. By introducing the MQMSA algorithm for real-time online optimization of PID parameters, not only is the control accuracy of environmental parameters considered, but also the energy consumption and mechanical wear of actuators (fans / air conditioners). Compared with traditional fixed-parameter PID or simple on / off control, this embodiment effectively suppresses overshoot and oscillation, making environmental fluctuations smoother and actuators more energy-efficient, thereby extending the service life of the core components of the smart cabinet and reducing long-term power consumption. By including electromagnetic environmental noise in the monitoring range and defining electromagnetic interference contention states, complex operating conditions such as wireless signal shielding and strong magnetic field interference can be identified.

[0024] Step S21 includes the following sub-steps: Step S211: Collect current running data according to the current sampling frequency, and at the same time retrieve historical running data within a preset time period from the local database. The historical running data includes CPU load, memory usage, electromagnetic environment noise floor data and historical statistical features. The historical statistical features include historical mean, historical variance, historical change slope, historical peak value and characteristic patterns of resource contention events. Step S212: Set a time window of length T, and align and merge the current running data with the historical running data: Within the current time window, the statistical characteristics of each indicator of the running data before calculation are obtained. The statistical characteristics include mean, variance, slope of change and number of peaks. The statistical characteristics of the current time window and the historical statistical characteristics are concatenated to obtain a multi-dimensional feature vector. Step S213: Input the multidimensional feature vector into the interference prediction model and perform time series analysis to output the resource contention status of the next time window.

[0025] Existing technologies typically only trigger alarms based on the current CPU utilization threshold (e.g., >80%), which can easily lead to false alarms due to momentary reasonable spikes or slow memory leaks. This invention introduces historical operating data and historical statistical features, looking not only at the current high level but also at the historical high levels at corresponding time points. This allows for the differentiation between normal periodic loads and abnormal sudden attacks, significantly reducing the false alarm rate. In this embodiment, the system does not wait for the load to reach 100% before triggering a response. Instead, it calculates the slope of change. When the CPU load or electromagnetic noise floor shows a positive high slope (rapid increase), even if the current value has not exceeded the standard, the model can predict that it will be overloaded in the next few seconds. This allows the system to be notified in advance to reserve resources. The multi-dimensional feature vector generated by this invention can comprehensively characterize the health status of the edge gateway and identify the coupling relationship between computing resource contention and electromagnetic interference contention, providing a precise decision-making basis for subsequent differentiated regulation.

[0026] Step S3 includes the following sub-steps: The edge gateway collects transmission parameter packets of each communication interface in real time. The transmission parameter packets include first status parameters and second status parameters. The first status parameters include latency jitter, signal quality attenuation rate and retransmission request count, which are used to indicate the link health level. The second status parameters include task number and corresponding monitoring task, which are used to indicate task priority. Latency jitter is the degree of dispersion in the time interval between consecutive data packets, used to indicate the trend of link congestion; The signal quality attenuation rate is the slope of the change in signal strength or signal-to-noise ratio per unit time, used to represent the degree of deterioration of the physical environment; The retransmission request count is the number of data packets for which no acknowledgment response has been received, used to indicate packet loss in the link. The predicted failure probability of the current primary communication link is calculated based on the first state parameters, and the first migration cost function is constructed by combining the cost coefficient of the backup communication link: ; in, This is the predicted failure probability value. This is the cost coefficient. This is the result of the first migration cost function; When any one of the first state parameters is greater than the preset security threshold, and the result of the first migration cost function is greater than the handover determination value, the main communication link is determined to be in an unhealthy state, and the edge gateway executes the first scheduling strategy. When the physical connection of the main communication link is broken, the main communication link is determined to be in a fault state. The first scheduling strategy includes: Parse the transmission parameter packet and identify the task number in the second state parameter. When the task number is a critical chain task, mark the transmission parameter packet with high priority. Extract the transmission parameter packets corresponding to the high-priority coloring marks, and forcibly modify the physical exit to the backup communication link interface. Transmit the remaining transmission parameter packets through the main communication link. When the main communication link fails, temporarily store them in the local cache queue.

[0027] In this embodiment, the predicted failure probability of the current main communication link is calculated based on the collected first state parameters. The predicted failure probability comprehensively considers the normalized value of latency jitter, signal attenuation rate, and retransmission request accumulation factor. At the same time, the current tariff unit price and remaining data package of the backup communication link are queried to calculate the tariff cost coefficient, and the first migration cost function is constructed and calculated in real time. ; in, The business sensitivity factor is set to 0.9 for critical chain tasks and 0.2 for non-critical chain tasks in this embodiment. It is an exponential amplification function of the risk term. The logarithmic suppression function for the cost term is then used to determine the status of the main communication link and trigger a response based on the following logic: Unhealthy state determination: When any one of the first state parameters (such as latency jitter) exceeds the preset safety threshold and the result of the first migration cost function is greater than the preset handover determination value, the main communication link is determined to be in an unhealthy state. At this time, the edge gateway immediately activates the first scheduling strategy. Fault Status Determination: When the physical layer connection of the main communication link is detected to be disconnected or more than 3 consecutive heartbeats are lost, the main communication link is determined to be in a fault state. At this time, the default strategy is to perform full traffic switching or local caching. The specific execution process of the first scheduling strategy: The gateway parses data packets at the IP layer, identifies the task number in the second state parameter, and if the number corresponds to a critical chain task, it uses differential service code points to mark the data packet as high priority. Packet-level redirection: The gateway's network layer intercepts all packets with high-priority coloring marks, ignoring the operating system's default routing table, and forcibly redirects their physical exit to an alternative communication link interface, achieving seamless migration of critical data; Non-critical task isolation: For non-critical chain task data packets that are not tagged, continue to be transmitted on the original main communication link. If the main communication link fails, write these data packets to the local or local buffer queue and retransmit them after the network is restored, without occupying expensive backup link resources.

[0028] Step S4 includes the following sub-steps: Step S41: Set up an inflow buffer pool before non-critical chain tasks are inflowed into the main communication link, preset a first threshold and a second threshold, and monitor the buffer occupancy rate of non-critical chain tasks in real time. Among them, after non-critical chain tasks are imported into the buffer pool, the storage space occupied by all non-critical chain task data packets in the current buffer pool is counted as the import buffer consumption value, and the ratio of the import buffer consumption value to the total capacity of the import buffer pool is used as the buffer occupancy rate. When the buffer occupancy rate is less than the first threshold, non-critical chain tasks are allowed to transmit normally. When the first threshold < buffer occupancy rate < second threshold, the edge gateway initiates a traffic shaping strategy to reduce the transmission priority of non-critical chain tasks; When the buffer occupancy rate exceeds the second threshold, temporarily discard the lowest priority non-critical chain tasks. Step S42: Set up the project buffer pool. The project buffer pool is a local non-volatile storage space. Write the alarm data and core status snapshots generated by the critical chain task into the project buffer pool. Step S43: When the project buffer pool is detected to be consumed, a local audible and visual alarm is triggered on the edge side. After the network connection is restored, the edge gateway automatically initiates a data reconciliation process with the cloud to retransmit the critical data temporarily stored in the project buffer pool.

[0029] In this embodiment, the connectivity status of the main communication link and the backup communication link is monitored in real time. When both links are unavailable or an unexpected power outage signal is detected, the alarm data and core status snapshot generated by the critical link task are written into the project buffer pool. At the network layer entry point of the edge gateway, a buffer pool is set up for all data streams identified as non-critical chain tasks, and mapped to a dedicated memory queue. Two key thresholds are preset: the first threshold is 50% capacity, and the second threshold is 90% capacity. By setting up the buffer pool, transmission delays caused by non-critical services preempting resources are eliminated, effectively preventing critical data delays caused by non-critical service congestion. In addition, through local audible and visual alarms and automatic reconciliation and retransmission mechanisms, this invention can automatically identify and upload critical data accumulated during the network outage after communication is restored, without the need for manual intervention to export or compare data.

[0030] Step S5 specifically includes: A digital twin model of the smart cabinet is built on a cloud server. Monitoring tasks are received in real time and sent to the corresponding virtual package entity in the digital twin model. The task is marked with color according to the task number to obtain task flow data. The resource contention status is mapped to the edge gateway component of the digital twin model using a heat map to obtain resource status data. Dynamically rendered link lines are constructed to perform virtual topology mapping on the main communication link and the backup communication link to obtain link topology data. The system tracks the flow of critical chain tasks in the virtual topology in real time. When a path migration is triggered, the digital twin model synchronously displays the dynamic process of the virtual package entity instantly jumping from the main link line to the backup link line and shows the buffer occupancy rate of the buffer pool and project buffer pool in a dynamic water level map.

[0031] In this embodiment, a three-dimensional digital twin model is deployed on a cloud server to receive task flow data, resource status data, and link topology data in real time; For task flow data, parse the criticality rating field in the task number. If it is a critical chain task, render the virtual package entity as a highlighted red; if it is a non-critical chain task, render it as a semi-transparent gray.

[0032] By constructing a digital twin model, the monitoring task resource status and network topology are transformed into an intuitive 3D visualization view. Operation and maintenance personnel can perceive the overall health status and micro-operational details of the smart cabinet cluster simply by looking at the heat map colors, line thicknesses, and virtual packet flow trajectories on the screen. This greatly reduces the operation and maintenance threshold and decision-making difficulty. By using the task number to connect the entire lifecycle flow trajectory, combined with the instant jump display function for path migration, transient faults such as network jitter, route switching, and packet loss retransmission can be traced. Operation and maintenance personnel can replay the historical status of the digital twin model to accurately reproduce the link switching process and buffer pool level changes when the fault occurred, quickly locating whether the performance bottleneck is caused by network congestion or insufficient computing power. The dynamic water level map displays the occupancy rate of the incoming buffer pool and the project buffer pool, and the heat map displays the resource load of the edge gateway. It can capture traffic surges or resource exhaustion in real time. When the water level map turns red or the heat map shows an alarm, operation and maintenance personnel can intervene manually in advance, significantly improving the long-term operational stability of the smart cabinet system.

[0033] The logic for processing the resource contention state of the next time window by inputting multidimensional feature vectors into the interference prediction model and performing time series analysis is as follows: An interference prediction model is constructed based on a selective state-space model. Multidimensional feature vectors are input into the pre-projection layer of the interference prediction model. Using the zero-order preservation discretization rule, the continuous-time multidimensional feature vectors are transformed into a discretized input sequence through a time-scale parameter. The input sequence is mapped to a high-dimensional latent state space through a state equation. The state equation is as follows: ; in, and The system matrix and input matrix are discretized. Given the input sequence at time k, Let k be the state space at time k-1, i.e., the potential state space. Let k be the state space at time k; A parallel correlation scanning algorithm is used to selectively scan the input sequence. The latent state space is mapped back to the output space through the output matrix, and state decoding is performed to obtain the prediction vector of future resource features. The prediction vector is processed through a root mean square normalization layer and a residual connection layer, and finally input into a fully connected layer. The output is a probability distribution vector processed by the Softmax activation function. The probability distribution vector includes the prediction probability of the contention-free state, the prediction probability of the resource contention state, and the prediction probability of the electromagnetic interference contention state. The sum of each sample in the probability distribution vector is 1. Based on the probability distribution vector, the resource contention state of the next time window is obtained by using the maximum a posteriori probability criterion.

[0034] In this embodiment, an interference prediction model based on the Mamba architecture is constructed. The concatenated multidimensional feature vector is input into the model's pre-projection layer. Using the zero-order preservation discretization rule, a learnable time scale parameter is introduced to convert the continuous-time multidimensional feature vector into a discretized input sequence. The linear projection layer dynamically generates the input matrix B and the system matrix A. The discretized input sequence is mapped to a high-dimensional latent state space. The latent state after selective scanning is mapped back to the output space through the output matrix and the state is decoded to obtain the resource contention state of the next time window. In this embodiment, by adopting an interference prediction architecture based on a selective state-space model, the traditional Transformer model, whose computational complexity increases quadratically with sequence length, is abandoned. A parallel correlation scanning algorithm is used to achieve linear computational complexity. Since the computing power of the embedded edge gateway is limited, it is necessary to quickly process historical monitoring data that lasts for several minutes, which cannot rely on traditional deep learning models. Therefore, the parallel correlation scanning algorithm is adopted, which significantly reduces inference latency and memory usage. By introducing learnable time scale parameters and dynamically generated input and system matrices, this invention enables the model to adapt to the high-frequency jitter of the CPU and the low-frequency drift fluctuations of electromagnetic noise. This allows the model to selectively focus on key interference features while ignoring irrelevant noise, thereby maintaining extremely high prediction accuracy when facing sudden high loads or electromagnetic interference.

[0035] The expression for calculating the switching decision value is: ; in, To switch the judgment value, This is the moving average of the results of the first migration cost function over a preset time period. Let be the standard deviation of the result of the first migration cost function, and k be the sensitivity coefficient. This is the load adjustment factor.

[0036] This embodiment details the dynamic calculation process of the switching decision value, with a sensitivity coefficient k=3. A load adjustment factor is introduced; if the current total gateway traffic > 80% of the bandwidth, ... <1, otherwise =1. This invention establishes a negative correlation mechanism between network load and switching threshold. When the total gateway traffic is detected to be >80% of the bandwidth, μ<1 will significantly reduce the switching judgment value. This means that in high-load congestion scenarios, the system will become more sensitive and more inclined to migrate critical chain tasks to backup links as early as possible, thereby avoiding queuing delays and packet loss caused by main link congestion, effectively filtering out normal jitter in the network, avoiding resource waste caused by oversensitivity, and ensuring rapid response in the face of real sudden failures.

[0037] Example 2, refer to Figure 2 This paper presents a smart cabinet monitoring system based on cloud-edge collaboration, including a task identification module, an edge resource allocation module, a critical chain planning module, a non-critical chain incorporation module, and a digital twin module. The task identification module is used to decompose the monitoring needs of the smart cabinet, generate a full life cycle task number using a distributed algorithm, and divide the decomposed monitoring tasks into critical chain tasks and non-critical chain tasks based on preset criticality rating rules. The edge resource allocation module is used to predict the impact of resource contention on the edge gateway of the smart cabinet in real time through the interference prediction model and PID controller, obtain the predicted resource contention status, adjust the PID parameters online using the MQMSA algorithm based on the internal environmental parameters of the smart cabinet, and reserve buffers for the front nodes of critical chain tasks based on the resource contention status. The critical chain planning module is used to monitor the current communication resource status of the smart cabinet in real time, run a multi-agent deep deterministic strategy feature model, automatically plan backup critical paths, and migrate critical chain tasks to backup critical paths. The non-critical chain inbound module is used to inbound non-critical chain tasks into the nodes corresponding to critical chain tasks, obtain the inbound buffer consumption value, judge the inbound buffer consumption value through the buffer judgment mechanism, and start the edge triggering mechanism. The digital twin module is used to build a digital twin model of the smart cabinet cluster and update the task flow path, corresponding resource contention status and edge triggering mechanism in real time.

[0038] By integrating task lifecycle management, edge-side interference prediction and autonomous control, multi-dimensional state-aware dynamic routing, and a two-level buffer disaster recovery mechanism, this invention significantly improves the reliability and intelligence level of the intelligent cabinet monitoring system in complex environments. Firstly, by utilizing distributed tags and criticality rating rules, this invention achieves atomic decomposition and full-link tracing of monitoring tasks, ensuring that critical chain tasks involving life and property safety enjoy the highest priority processing channel under any resource-constrained or network congestion conditions, eliminating delays and loss of core alarm data. Secondly, this invention innovatively introduces an interference prediction model based on time series analysis and an MQMSA-optimized PID control algorithm to proactively detect computing resource contention and electromagnetic interference risks within future time windows and dynamically reserve state buffers, effectively solving high-concurrency... To address the challenge of blocked underlying monitoring processes in emergency scenarios, while ensuring the stability of cabinet environment control and actuator energy efficiency, this invention constructs a micro-routing decision mechanism based on a cost function. It abandons the traditional network outage switching mode and triggers packet-level redirection when the main link shows unhealthy signs by real-time calculation of link failure probability and cost. Combined with traffic coloring technology, it seamlessly migrates only critical services to the backup link, significantly reducing traffic costs. Finally, this invention designs a two-level disaster recovery mechanism with ingress buffer and project buffer, as well as a digital twin closed-loop feedback system. It utilizes local non-volatile storage to save critical on-site data during extreme network and power outages and optimizes system thresholds through virtual simulation, achieving a complete closed loop of monitoring data throughout the entire lifecycle and fault traceability, significantly reducing the operation and maintenance threshold and labor costs.

[0039] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. 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 be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A smart cabinet monitoring method based on cloud-edge collaboration, characterized in that, Includes the following steps: Step S1: Based on the monitoring requirements of the smart cabinet, the monitoring tasks are decomposed, task numbers are generated, and the monitoring tasks are divided into critical chain tasks and non-critical chain tasks according to the preset criticality rating rules. Step S2: Through the interference prediction model and PID controller, the impact of resource contention on the edge gateway of the smart cabinet is predicted in real time. The resource contention status is predicted. Based on the internal environmental parameters of the smart cabinet, the MQMSA algorithm is used to adjust the PID parameters online. Buffer reservations are made for the front nodes of the critical chain tasks based on the resource contention status. Step S3: Based on the current resource status of the smart cabinet, use the multi-agent deep deterministic strategy feature model to predict the backup critical path, and migrate the critical chain task to the backup critical path. Step S4: Inject non-critical chain tasks into the nodes corresponding to critical chain tasks to obtain the injection buffer consumption value. Use the buffer judgment mechanism to judge the injection buffer consumption value and start the edge triggering mechanism. Step S5: Construct a digital twin model of the smart cabinet cluster and update the task flow path, corresponding resource contention status, and edge triggering mechanism in real time.

2. The intelligent cabinet monitoring method based on cloud-edge collaboration as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11: Through the operation of the smart cabinet terminal, scan all currently running processes and decompose them into monitoring tasks. Use a distributed ID generation algorithm to generate a task number for each monitoring task and embed the task number into the header of the data packet of the monitoring task. The task number is a unique index for the entire life cycle. Step S12: Preset criticality rating rules, and determine the monitoring tasks according to the criticality rating rules to obtain critical chain tasks and non-critical chain tasks. The critical chain tasks include early warning of environmental changes, theft prevention of core assets, and handling of external emergencies. The non-critical chain tasks include routine operation and maintenance data collection, user behavior analysis, and commercial content delivery.

3. The intelligent cabinet monitoring method based on cloud-edge collaboration as described in claim 1, characterized in that, Step S2 includes the following sub-steps: Step S21: Deploy an interference prediction model based on time series analysis on the edge gateway. The interference prediction model collects the CPU load, memory usage and electromagnetic noise floor data of the gateway in real time as the current running data. Based on the historical running data and the current running data, a time window is set to predict the resource contention state of the next time window. The resource contention state includes no contention state, computing resource contention state and electromagnetic interference contention state. Step S22: The internal environmental parameters of the intelligent cabinet are used as the controlled variables. The MQMSA algorithm is used to optimize the proportional coefficient, integral coefficient and derivative coefficient of the PID controller in real time. The MQMSA algorithm uses environmental stability and actuator energy consumption as a joint fitness function, outputs the optimal control parameters, and drives the actuator to perform internal control within the cabinet. Step S23: Based on the predicted resource contention status of the next time window, reserve buffers for the critical task front nodes of the edge gateway. When a computational resource contention state is predicted, the edge gateway immediately allocates a buffer, allowing only critical chain tasks to access it. When the resource contention state is in the computational resource contention state, the edge gateway immediately allocates a buffer, allowing access to the buffer when the task number corresponds to a critical chain task, and prohibiting access to the buffer when the task number corresponds to a non-critical chain task.

4. The intelligent cabinet monitoring method based on cloud-edge collaboration as described in claim 3, characterized in that, Step S21 includes the following sub-steps: Step S211: Collect current running data according to the current sampling frequency, and at the same time retrieve historical running data within a preset time period from the local database. The historical running data includes CPU load, memory usage, electromagnetic environment noise floor data, and historical statistical features. The historical statistical features include historical mean, historical variance, historical change slope, historical peak value, and characteristic patterns of resource contention events. Step S212: Set a time window of length T, and align and merge the current running data with the historical running data: Within the current time window, the statistical characteristics of each indicator of the running data are calculated. The statistical characteristics include mean, variance, slope of change and number of peaks. The statistical characteristics of the current time window and the historical statistical characteristics are concatenated to obtain a multidimensional feature vector. Step S213: Input the multidimensional feature vector into the interference prediction model, perform time series analysis, and output the resource contention status of the next time window.

5. The intelligent cabinet monitoring method based on cloud-edge collaboration as described in claim 1, characterized in that, Step S3 includes the following sub-steps: The edge gateway collects transmission parameter packets of each communication interface in real time. The transmission parameter packets include a first status parameter and a second status parameter. The first status parameter includes latency jitter, signal quality attenuation rate and retransmission request count, which are used to indicate the link health level. The second status parameter includes task number and corresponding monitoring task, which are used to indicate task priority. Latency jitter is the degree of dispersion in the time interval between consecutive data packets, used to indicate the trend of link congestion; The signal quality attenuation rate is the slope of the change in signal strength or signal-to-noise ratio per unit time, used to represent the degree of deterioration of the physical environment; The retransmission request count is the number of data packets for which no acknowledgment response has been received, used to indicate packet loss in the link. Based on the first state parameters, the predicted failure probability of the current primary communication link is calculated, and a first migration cost function is constructed by combining the cost coefficient of the backup communication link: ； in, This is the predicted failure probability value. This is the cost coefficient. This is the result of the first migration cost function; When any one of the first state parameters is greater than a preset security threshold, and the result of the first migration cost function is greater than the handover determination value, the main communication link is determined to be in an unhealthy state, and the edge gateway executes the first scheduling strategy. When the physical connection of the main communication link is broken, the main communication link is determined to be in a fault state. The first scheduling strategy includes: Parse the transmission parameter packet and identify the task number in the second state parameter. When the task number is a critical chain task, mark the transmission parameter packet with high priority. Extract the transmission parameter packets corresponding to the high-priority coloring marks, and forcibly modify the physical exit to the backup communication link interface. Transmit the remaining transmission parameter packets through the main communication link. When the main communication link fails, temporarily store them in the local cache queue.

6. The intelligent cabinet monitoring method based on cloud-edge collaboration as described in claim 1, characterized in that, Step S4 includes the following sub-steps: Step S41: Set up an inflow buffer pool before non-critical chain tasks are inflowed into the main communication link, preset a first threshold and a second threshold, and monitor the buffer occupancy rate of non-critical chain tasks in real time. When the buffer occupancy rate is less than the first threshold, non-critical chain tasks are allowed to transmit normally. When the first threshold < buffer occupancy rate < second threshold, the edge gateway initiates a traffic shaping strategy to reduce the transmission priority of non-critical chain tasks; When the buffer occupancy rate exceeds the second threshold, temporarily discard the lowest priority non-critical chain tasks. Step S42: Set up a project buffer pool, which is a local non-volatile storage space. Write the alarm data and core status snapshots generated by the critical chain task into the project buffer pool. Step S43: When the project buffer pool is detected to be consumed, a local audible and visual alarm is triggered on the edge side. After the network connection is restored, the edge gateway automatically initiates a data reconciliation process with the cloud to retransmit the critical data temporarily stored in the project buffer pool.

7. The intelligent cabinet monitoring method based on cloud-edge collaboration as described in claim 1, characterized in that, Step S5 specifically includes: A digital twin model of the smart cabinet is built on a cloud server. Monitoring tasks are received in real time and sent to the corresponding virtual package entity in the digital twin model. The task is marked with color according to the task number to obtain task flow data. The resource contention status is mapped to the edge gateway component of the digital twin model using a heat map to obtain resource status data. Dynamically rendered link lines are constructed, and the main communication link and backup communication link are virtually topologically mapped to obtain link topology data. The system tracks the flow of critical chain tasks in the virtual topology in real time. When a path migration is triggered, the digital twin model synchronously displays the dynamic process of the virtual package entity instantly jumping from the main link line to the backup link line and shows the buffer occupancy rate of the buffer pool and project buffer pool in a dynamic water level map.

8. The intelligent cabinet monitoring method based on cloud-edge collaboration as described in claim 4, characterized in that, The multidimensional feature vector is input into the interference prediction model, and time series analysis is performed. The processing logic for outputting the resource contention state of the next time window is as follows: An interference prediction model is constructed based on a selective state-space model. Multidimensional feature vectors are input into the pre-projection layer of the interference prediction model. Using the zero-order preservation discretization rule, the continuous-time multidimensional feature vectors are transformed into a discretized input sequence through a time-scale parameter. The input sequence is mapped to a high-dimensional latent state space through a state equation. The state equation is as follows: ； in, and The system matrix and input matrix are discretized. Given the input sequence at time k, Let k be the state space at time k-1, i.e., the potential state space. Let k be the state space at time k; A parallel correlation scanning algorithm is used to selectively scan the input sequence. The latent state space is mapped back to the output space through the output matrix, and state decoding is performed to obtain a prediction vector of future resource features. The prediction vector is processed through a root mean square normalization layer and a residual connection layer, and finally input into a fully connected layer. The output is a probability distribution vector processed by the Softmax activation function. The probability distribution vector includes the prediction probability of a contention-free state, the prediction probability of a resource contention state, and the prediction probability of an electromagnetic interference contention state. The sum of each sample in the probability distribution vector is 1. Based on the probability distribution vector, the resource contention state of the next time window is obtained using the maximum a posteriori probability criterion.

9. The intelligent cabinet monitoring method based on cloud-edge collaboration as described in claim 5, characterized in that, The expression for calculating the switching determination value is: ； in, To switch the judgment value, This is the moving average of the results of the first migration cost function over a preset time period. Let be the standard deviation of the result of the first migration cost function, and k be the sensitivity coefficient. This is the load adjustment factor.

10. A cloud-edge collaborative intelligent cabinet monitoring system, applied in any one of the cloud-edge collaborative intelligent cabinet monitoring methods as described in claims 1-9, characterized in that, It includes a task identification module, an edge resource allocation module, a critical chain planning module, a non-critical chain incorporation module, and a digital twin module; The task identification module is used to decompose the monitoring requirements of the smart cabinet into tasks, generate a full life cycle task number using a distributed algorithm, and divide the decomposed monitoring tasks into critical chain tasks and non-critical chain tasks based on preset criticality rating rules. The edge resource allocation module is used to predict the impact of resource contention on the edge gateway of the smart cabinet in real time through the interference prediction model and PID controller, predict the resource contention status, adjust the PID parameters online using the MQMSA algorithm based on the internal environmental parameters of the smart cabinet, and reserve buffers for the front nodes of critical chain tasks based on the resource contention status. The critical chain planning module is used to monitor the current communication resource status of the smart cabinet in real time, run a multi-agent deep deterministic strategy feature model, automatically plan backup critical paths, and migrate critical chain tasks to backup critical paths. The non-critical chain inbound module is used to inbound non-critical chain tasks into the nodes corresponding to critical chain tasks, obtain inbound buffer consumption values, judge the inbound buffer consumption values ​​through a buffer judgment mechanism, and start the edge triggering mechanism. The digital twin module is used to construct a digital twin model of the smart cabinet cluster and update the task flow path, corresponding resource contention status and edge triggering mechanism in real time.

Images