Cloud edge cooperative computing system for household elevator digital factory

Abstract

The invention relates to the technical field of cloud-side cooperative computing, and discloses a cloud-side cooperative computing system for a household elevator digital factory, which comprises a cloud computing center, a factory edge computing platform and a plurality of edge computing nodes deployed on a production line, the cloud computing center is configured to execute a non-real-time global task, and the non-real-time global task comprises the steps of training and updating a global artificial intelligence model for quality prediction and process optimization by utilizing a federated learning framework based on aggregated data from a plurality of factory edge computing platforms; and executing cross-factory production data mining and long-term efficiency analysis. The edge computing node is introduced to be directly connected with the production equipment and the sensor, localization processing of tasks is achieved, the edge computing node executes atomic computing tasks such as equipment real-time closed-loop control and high-frequency sensing data flow processing, and delay caused by remote transmission of data to the cloud is avoided.

Description

Technical Field

[0001] This invention relates to the field of cloud-edge collaborative computing technology, specifically to a cloud-edge collaborative computing system for a digital factory for home elevators. Background Technology

[0002] With the rapid development of industrial internet and intelligent manufacturing technologies, the home elevator manufacturing industry is actively transforming into digital and networked factories. Currently, this field generally adopts a digital factory architecture centered on cloud computing, relying on the powerful storage and computing capabilities of remote data centers to achieve production data aggregation, process simulation optimization, and enterprise-level resource management, providing effective support for macro-planning and decision-making in the production process.

[0003] However, when facing the diverse production scenarios of home elevators, centralized cloud architecture suffers from significant physical distances between the cloud and on-site equipment, resulting in substantial network transmission latency. This makes it difficult to meet the stringent requirements for millisecond-level response in elevator assembly precision control, real-time quality inspection, and equipment fault diagnosis. On the other hand, uploading all the massive amounts of real-time data generated at the edge to the cloud for processing not only consumes a large amount of bandwidth and increases costs but also poses risks to data security and privacy. Based on this, the present invention designs a cloud-edge collaborative computing system for digital factories of home elevators to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a cloud-edge collaborative computing system for a digital factory of home elevators, which solves the problems of long physical distance and large amount of real-time data in the background technology.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A cloud-edge collaborative computing system for a digital factory of home elevators includes: a cloud computing center, a factory edge computing platform, and multiple edge computing nodes deployed on the production line;

[0007] The cloud computing center is configured to perform non-real-time global tasks, including: training and updating a global artificial intelligence model for quality prediction and process optimization using a federated learning framework based on aggregated data from multiple factory edge computing platforms; and performing cross-factory production data mining and long-term performance analysis.

[0008] The factory edge computing platform is configured to: manage the registration, status monitoring, and resource orchestration of all edge computing nodes within the factory; receive and interpret models and macro instructions from the cloud computing center; decompose the received composite production tasks into multiple atomic computing tasks, and allocate the atomic computing tasks to multiple target edge computing nodes based on a dynamic offloading decision algorithm aimed at minimizing the total system cost.

[0009] The edge computing node is directly connected to the production equipment and sensors, and is configured to: execute atomic computing tasks assigned by the factory edge computing platform, the tasks including real-time closed-loop control of equipment, high-frequency sensor data stream processing, and real-time quality anomaly detection based on a lightweight model; and feed back the processing results and summary data to the factory edge computing platform in real time.

[0010] Preferably, the objective of the dynamic unloading decision algorithm is to minimize the total system cost J. total Its objective function is defined as:

[0011] ;

[0012] The constraint is: for any real-time atomic computation task i, the following condition is satisfied. ;

[0013] N is the total number of atomic computing tasks currently awaiting allocation; i is the task index; T i For the estimated end-to-end latency of task i, E i To estimate energy costs, S i Score the safety confidence of the allocation scheme; λ t , λ e With λ s These are the dynamic weighting coefficients for latency, energy consumption, and security, respectively, and they satisfy λ. t +λ e +λ s =1; T max This represents the maximum allowed response latency for task i.

[0014] Preferably, the estimated end-to-end delay T i The calculation formula is:

[0015] ;

[0016] Among them, T ci T represents the communication latency of task i. qi T represents the queuing delay of task i on the candidate edge computing node. ei This represents the computation execution latency of task i on the candidate edge computing node.

[0017] ;

[0018] Among them, D in and D out B represents the input data volume and the result data volume of task i, respectively. i,n The effective bandwidth of the network link allocated to task i, where τ0 is the fixed propagation delay;

[0019] ;

[0020] Among them, Q n Let μ be the total number of tasks currently pending for candidate node n. n The average processing rate of node n;

[0021] ;

[0022] Among them, C i f is the number of computation cycles required for task i. n The available computational frequency for node n.

[0023] Preferably, the estimated energy cost E i The calculation formula is:

[0024] ;

[0025] in, ;

[0026] P total P represents the total power consumption of candidate edge computing node n when it performs a task. static Let k be the static power consumption of node n. n The effective capacitance coefficient, f, associated with the node n chip architecture n Its operating frequency;

[0027] ρt is the unit energy price coefficient at the current moment;

[0028] T ei This represents the computation execution latency of task i on the candidate edge computing node.

[0029] Preferably, the security confidence score S i The calculation formula is:

[0030] ;

[0031] Where: J is the total number of preset security assessment dimensions; j is the security dimension index; Vji is the required threshold for task i in the j-th security dimension; Vjn is the guarantee capability provided by candidate edge computing node n in the j-th security dimension; ω j Let be the weight coefficient of the j-th dimension, and satisfy . .

[0032] Preferably, the dynamic weighting coefficient λ t , λ e With λ s The adjustment method is as follows:

[0033] Obtain the basic weights based on the current production stage strategy configuration;

[0034] λbase t is the base weight for time delay, λbase e is the base weight for energy consumption, and λbase s is the base weight for security items;

[0035] Fine-tuning is performed based on the real-time system status, using the following formula:

[0036] ;

[0037] ;

[0038] ;

[0039] Among them, L c L represents the current average network load rate. avg Its historical average; Z c Z represents the current level of network threat detected. avg It is its historical average level; α and β are sensitivity adjustment coefficients.

[0040] Preferably, the factory edge computing platform includes a fault tolerance and migration module. When a failure is detected in the target edge computing node, this module automatically executes the following fault tolerance and migration process:

[0041] Step S01: Immediately freeze the state and execution context of all atomic computation tasks currently being executed on the faulty node;

[0042] Step S02: Based on the latest system resource status and node load information, rerun the dynamic unloading decision algorithm to calculate and select a new optimal target edge computing node for each frozen task;

[0043] Step S03: Completely migrate the frozen task state image and related context data to the newly selected edge computing node, and resume task execution from the point of interruption.

[0044] Preferably, the cloud computing center and the factory edge computing platform adopt a model collaborative update mechanism based on federated learning, and the operation process of this mechanism is as follows:

[0045] The factory edge computing platform periodically collects the model gradients generated by each edge computing node under its jurisdiction during local training, and adds random noise that conforms to the differential privacy standard for privacy processing before the gradients are uploaded.

[0046] On the platform side, multiple local gradients that have undergone privacy processing are securely aggregated, and the aggregated gradient is then uploaded to the cloud computing center.

[0047] The cloud computing center receives aggregated gradients from multiple factory edge computing platforms and performs a weighted average of the aggregated gradients based on the proportion of data contributed by each factory, thereby updating the parameters of the global artificial intelligence model.

[0048] Preferably, when the factory edge computing platform decomposes composite production tasks into atomic computing tasks, the task classification and priority rules it uses comprehensively consider the following dimensions:

[0049] The real-time level of a task is determined based on the maximum allowed end-to-end response time threshold. The more urgent the response time requirement, the higher the real-time level.

[0050] The computational complexity of a task is quantified by the scale of computational operations required to perform that task.

[0051] The security level of task data is classified according to the sensitivity of the production data processed by the task;

[0052] The overall priority of a task is calculated by weighting the above-mentioned indicators. Tasks with high real-time requirements, high computational complexity, and high data security level will be given higher overall priority and will be allocated computing resources in subsequent scheduling.

[0053] Preferably, the aforementioned.

[0054] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0055] 1. In this invention, by introducing edge computing nodes to directly connect production equipment and sensors, localized task processing is achieved. Edge computing nodes execute atomic computing tasks, avoiding the delay caused by remote data transmission to the cloud. At the same time, the factory edge computing platform adopts a dynamic offloading decision algorithm to optimize task allocation with the goal of minimizing the total system cost, ensuring that real-time tasks are processed first on the edge side, and meeting the millisecond-level response requirements of the production process.

[0056] 2. In this invention, edge computing nodes directly process high-frequency sensor data streams and real-time quality anomaly detection on the production line side, only feeding back the processing results and summary data to the factory edge computing platform, rather than the original massive data, thereby greatly reducing network bandwidth usage and transmission costs. The factory edge computing platform further aggregates and simplifies the data before uploading it to the cloud computing center for non-real-time global tasks. This data filtering mechanism optimizes resource utilization, reduces operating costs, and solves the bandwidth waste problem of centralized architecture.

[0057] 3. In this invention, edge computing nodes process sensitive production data locally, reducing the chance of data leakage. The factory edge computing platform employs a security confidence scoring mechanism to dynamically assess the security of task allocation and adds noise to the uploaded model gradients using differential privacy technology to achieve privacy processing. The cloud computing center and platform utilize a federated learning-based collaborative update mechanism, aggregating only encrypted gradient data and avoiding the sharing of raw data. This effectively protects sensitive enterprise information and improves the overall system's security and compliance. Attached Figure Description

[0058] Figure 1 This is a diagram of the overall system architecture of the present invention;

[0059] Figure 2 This is a flowchart of the fault-tolerant migration module of the present invention;

[0060] Figure 3 This is a flowchart illustrating the task decomposition and priority allocation process of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example 1;

[0063] Please see Figures 1-3 A cloud-edge collaborative computing system for a digital factory of home elevators includes: a cloud computing center, a factory edge computing platform, and multiple edge computing nodes deployed on the production line;

[0064] The cloud computing center is configured to perform non-real-time global tasks, including: training and updating a global artificial intelligence model for quality prediction and process optimization using a federated learning framework based on aggregated data from multiple factory edge computing platforms; and performing cross-factory production data mining and long-term performance analysis.

[0065] The factory edge computing platform is configured to: manage the registration, status monitoring and resource orchestration of all edge computing nodes within the factory; receive and interpret models and macro instructions from the cloud computing center; decompose the received composite production tasks into multiple atomic computing tasks, and allocate the atomic computing tasks to multiple target edge computing nodes based on a dynamic offloading decision algorithm aimed at minimizing the total system cost.

[0066] Edge computing nodes are directly connected to production equipment and sensors and are configured to execute atomic computing tasks assigned by the factory edge computing platform. These tasks include real-time closed-loop control of equipment, high-frequency sensor data stream processing, and real-time quality anomaly detection based on lightweight models. The processing results and summary data are fed back to the factory edge computing platform in real time.

[0067] The objective of the dynamic unloading decision algorithm is to minimize the total system cost J. total Its objective function is defined as:

[0068] ;

[0069] The constraint is: for any real-time atomic computation task i, the following condition is satisfied. ;

[0070] N is the total number of atomic computing tasks currently awaiting allocation; i is the task index; T i For the estimated end-to-end latency of task i, E i To estimate energy costs, S i Score the safety confidence of the allocation scheme; λ t , λ e With λ s These are the dynamic weighting coefficients for latency, energy consumption, and security, respectively, and they satisfy λ. t +λ e +λ s =1; T max This represents the maximum allowed response latency for task i.

[0071] Estimated end-to-end delay T i The calculation formula is:

[0072] ;

[0073] Among them, T ci T represents the communication latency of task i. qi T represents the queuing delay of task i on the candidate edge computing node. ei This represents the computation execution latency of task i on the candidate edge computing node.

[0074] ;

[0075] Among them, D in and D out B represents the input data volume and the result data volume of task i, respectively. i,n The effective bandwidth of the network link allocated to task i, where τ0 is the fixed propagation delay;

[0076] ;

[0077] Among them, Qn Let μ be the total number of tasks currently pending for candidate node n. n The average processing rate of node n;

[0078] ;

[0079] Among them, C i f is the number of computation cycles required for task i. n The available computational frequency for node n.

[0080] Estimated energy cost E i The calculation formula is:

[0081] ;

[0082] in, ;

[0083] P total P represents the total power consumption of candidate edge computing node n when it performs a task. static Let k be the static power consumption of node n. n The effective capacitance coefficient, f, associated with the node n chip architecture n Its operating frequency;

[0084] ρt is the unit energy price coefficient at the current moment;

[0085] T ei This represents the computation execution latency of task i on the candidate edge computing node.

[0086] Security confidence score S i The calculation formula is:

[0087] ;

[0088] Where: J is the total number of preset security assessment dimensions; j is the security dimension index; Vji is the required threshold for task i in the j-th security dimension; Vjn is the guarantee capability provided by candidate edge computing node n in the j-th security dimension; ω j Let be the weight coefficient of the j-th dimension, and satisfy . .

[0089] Dynamic weighting coefficient λ t , λ e With λ s The adjustment method is as follows:

[0090] Obtain the basic weights based on the current production stage strategy configuration;

[0091] λbase t is the base weight for time delay, λbase e is the base weight for energy consumption, and λbase s is the base weight for security items;

[0092] Fine-tuning is performed based on the real-time system status, using the following formula:

[0093] ;

[0094] ;

[0095] ;

[0096] Among them, L c L represents the current average network load rate. avg Its historical average; Z c Z represents the current level of network threat detected. avg It is its historical average level; α and β are sensitivity adjustment coefficients.

[0097] The working principle of this invention is as follows: The cloud-edge collaborative computing system for digital factories of home elevators achieves efficient collaboration through a three-tier architecture consisting of a cloud computing center, a factory edge computing platform, and edge computing nodes. During system operation, the cloud computing center first aggregates data from multiple factory edge computing platforms based on a federated learning framework, training and updating a global artificial intelligence model for non-real-time tasks such as quality prediction and process optimization; simultaneously, it performs cross-factory data mining and long-term performance analysis to support macro-level decision-making. The factory edge computing platform, acting as an intermediate layer, manages the registration, status monitoring, and resource orchestration of all edge computing nodes within the factory, receiving and interpreting models and instructions issued from the cloud; when receiving complex production tasks, the platform decomposes them into multiple atomic computing tasks and allocates them according to a dynamic offloading decision algorithm.

[0098] Dynamic weighting coefficients are adjusted based on the system's real-time status, such as fine-tuning based on network load and threat level, ensuring adaptive optimization of the algorithm. Edge computing nodes are directly connected to production equipment and sensors, executing atomic tasks such as real-time closed-loop control, data stream processing, and anomaly detection, and feeding the results back to the platform to form a closed loop. Overall, the system achieves a balance between real-time response and resource efficiency through layered collaboration, solving the problems of high latency and high bandwidth consumption in cloud processing.

[0099] Example 2;

[0100] Please see Figures 1-3 In this embodiment of the invention, the factory edge computing platform includes a fault tolerance and migration module. When a fault is detected in the target edge computing node, this module automatically executes the following fault tolerance and migration process:

[0101] Step S01: Immediately freeze the state and execution context of all atomic computing tasks currently executing on the faulty node; Step S02: Based on the latest system resource status and node load information, rerun the dynamic unloading decision algorithm to calculate and select a new optimal target edge computing node for each frozen task; Step S03: Completely migrate the frozen task state image and related context data to the newly selected edge computing node, and resume task execution from the interruption point.

[0102] The cloud computing center and the factory edge computing platform adopt a model collaborative update mechanism based on federated learning. The operation process of this mechanism is as follows:

[0103] The factory edge computing platform regularly collects the model gradients generated by each edge computing node under its jurisdiction during local training, and adds random noise that meets the differential privacy standard for privacy processing before the gradients are uploaded.

[0104] On the platform side, multiple local gradients that have undergone privacy processing are securely aggregated, and the aggregated gradient is then uploaded to the cloud computing center.

[0105] The cloud computing center receives aggregated gradients from multiple factory edge computing platforms and performs a weighted average of the aggregated gradients based on the proportion of data contributed by each factory, thereby updating the parameters of the global artificial intelligence model.

[0106] When decomposing complex production tasks into atomic computing tasks, the factory edge computing platform considers the following dimensions in its task classification and priority rules:

[0107] The real-time level of a task is determined based on the maximum allowed end-to-end response time threshold. The more urgent the response time requirement, the higher the real-time level.

[0108] The computational complexity of a task is quantified by the scale of computational operations required to perform that task.

[0109] The security level of task data is classified according to the sensitivity of the production data processed by the task;

[0110] The overall priority of a task is calculated by weighting the above-mentioned indicators. Tasks with high real-time requirements, high computational complexity, and high data security level will be given higher overall priority and will be allocated computing resources in subsequent scheduling.

[0111] The working principle of this invention embodiment is as follows: In this embodiment, the system further enhances robustness and intelligence through fault tolerance and migration modules, federated learning model collaborative update mechanisms, and task classification priority rules. When the factory edge computing platform detects a failure in the target edge computing node, the fault tolerance and migration module automatically starts: First, in step S01, the state and context of all tasks currently executing on the failed node are immediately frozen; second, in step S02, the dynamic unloading decision algorithm is re-run based on the latest system resources to calculate a new optimal node for each frozen task; finally, in step S03, the task state image is completely migrated to the new node and restored from the point of interruption to ensure business continuity. Simultaneously, a federated learning collaborative update mechanism is adopted between the cloud and the platform: the platform periodically collects the local model gradients of the edge nodes, adds differential privacy noise for privacy processing, and uploads them to the cloud after secure aggregation on the platform side; the cloud aggregates gradients by weighted average according to the proportion of data volume in each factory, updates global model parameters, and achieves privacy-preserving distributed learning. Furthermore, when decomposing composite tasks, the platform calculates a comprehensive priority based on multiple dimensions such as task real-time level, computational complexity, and data security level, with high-priority tasks receiving priority resource allocation. This design ensures system reliability through fault tolerance, improves model accuracy through federated learning, and optimizes resource utilization through priority scheduling, effectively addressing uncertainties and security risks in the production environment.

[0112] Example 3;

[0113] Please see Figures 1-3 In the embodiments of the present invention, specific embodiments are described below.

[0114] This embodiment describes a specific implementation of a cloud-edge collaborative computing system for a digital factory of home elevators. This system achieves efficient task processing through the collaborative operation of a cloud computing center, a factory edge computing platform, and edge computing nodes. In an example scenario, the factory edge computing platform receives a complex production task involving real-time quality inspection and equipment control on an elevator assembly line. During system initialization, the cloud computing center pre-trains a global artificial intelligence model to predict quality anomalies. The factory edge computing platform manages 10 edge computing nodes, each with registration information including a computing frequency of 2 GHz, static power consumption of 10 watts, and an effective capacitance coefficient of 0.5. The platform monitors node status in real time, such as an average network load rate of 0.8, a historical average load rate of 0.7, a current network threat level of 2, and a historical average threat level of 1.5.

[0115] When a complex task arrives, the factory edge computing platform breaks it down into five atomic computing tasks, with task indices i ranging from 1 to 5. Each task has specific parameters: Task 1 has 50 megabytes of input data and 10 megabytes of result data, Task 2 has 60 megabytes of input data and 12 megabytes of result data, and so on. The platform assigns tasks based on a dynamic offloading decision algorithm that minimizes the total system cost J. total For the goal. J total The calculation formula is to sum the weighted sum of the delay term, energy consumption term, and safety term from i equal to 1 to 5, where the delay weight λ t The base value is 0.4, and the energy consumption weight λ e The base value is 0.3, and the safety weight is λ. s The base value is 0.3, and it is fine-tuned according to the real-time status: the sensitivity coefficients α and β are 0.1, and the adjusted λ t The result is 0.414; λ s This equals 0.3 plus 0.1 multiplied by the current threat level of 2 minus the historical average threat level of 1.5, resulting in 0.35; λ e This equals 1 minus 0.414 minus 0.35, resulting in 0.236. The constraint is the estimated end-to-end delay T for each task i. i The response time T must not exceed the maximum response time. max .

[0116] Estimated end-to-end delay T i The calculation includes communication delay T ci Queuing delay T qi And the calculation execution delay T ei Taking Task 1 as an example, T ci The result is equal to the input data size of 50 megabytes plus the result data size of 10 megabytes, divided by the effective network link bandwidth of 100 megabits per second plus a fixed propagation delay of 5 milliseconds, resulting in 60 milliseconds; T qi The result is equal to the total number of pending tasks at the candidate node (8) divided by the average processing rate of 2 tasks per second, which equals 4 seconds, or 4000 milliseconds; T ei The result is equal to the number of computation cycles required for the task divided by the available computing frequency of the node, which is 2 gigahertz, resulting in 500 milliseconds. Therefore, T i The sum is 60 plus 4000 plus 500, which equals 4560 milliseconds, but T must be satisfied. i Less than or equal to T max Otherwise, the allocation will be adjusted.

[0117] Energy cost E i The calculation is based on the total power consumption P. total Multiply by T ei Multiply by the unit energy price coefficient ρt, where P totalThis equals a static power consumption of 10 watts plus an effective capacitance factor of 0.5 multiplied by a calculation frequency of 2 gigahertz, which equals 11 watts. ρt is 0.1 yuan per kilowatt-hour, and T... ei E is 500 milliseconds, or 0.5 seconds. i 11 multiplied by 0.5 multiplied by 0.1 equals 0.55 yuan. Safety confidence score S i The dimensional weight ω is calculated using three security dimensions. j The values ​​are 0.3, 0.3, and 0.4 respectively.

[0118] During the task execution process, the platform continuously monitors the node status; if a node fails, the fault-tolerant migration module freezes the task status, re-runs the algorithm, and assigns the task to a new node to ensure recovery.

[0119] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud-edge collaborative computing system for a digital factory of home elevators, characterized in that, include: Cloud computing center, factory edge computing platform, and multiple edge computing nodes deployed on the production line; The cloud computing center is configured to perform non-real-time global tasks, including: training and updating a global artificial intelligence model for quality prediction and process optimization using a federated learning framework based on aggregated data from multiple factory edge computing platforms; and performing cross-factory production data mining and long-term performance analysis. The factory edge computing platform is configured to: manage the registration, status monitoring, and resource orchestration of all edge computing nodes within the factory; receive and interpret models and macro instructions from the cloud computing center; decompose the received composite production tasks into multiple atomic computing tasks, and allocate the atomic computing tasks to multiple target edge computing nodes based on a dynamic offloading decision algorithm aimed at minimizing the total system cost. The edge computing node is directly connected to the production equipment and sensors, and is configured to: execute atomic computing tasks assigned by the factory edge computing platform, the tasks including real-time closed-loop control of equipment, high-frequency sensor data stream processing, and real-time quality anomaly detection based on a lightweight model; and feed back the processing results and summary data to the factory edge computing platform in real time.

2. The cloud-edge collaborative computing system for a digital factory of home elevators according to claim 1, characterized in that, The objective of the dynamic unloading decision algorithm is to minimize the total system cost J. total Its objective function is defined as: ； The constraint is: for any real-time atomic computation task i, the following condition is satisfied. ; N is the total number of atomic computing tasks currently awaiting allocation; i is the task index; T i For the estimated end-to-end latency of task i, E i To estimate energy costs, S i Score the safety confidence of the allocation scheme; λ t , λ e With λ s These are the dynamic weighting coefficients for latency, energy consumption, and security, respectively, and they satisfy λ. t +λ e +λ s =1; T max This represents the maximum allowed response latency for task i.

3. The cloud-edge collaborative computing system for a digital factory of home elevators according to claim 2, characterized in that, The estimated end-to-end delay T i The calculation formula is: ； Among them, T ci T represents the communication latency of task i. qi T represents the queuing delay of task i on the candidate edge computing node. ei This represents the computation execution latency of task i on the candidate edge computing node.

4. The cloud-edge collaborative computing system for a digital factory of home elevators according to claim 3, characterized in that: ； Among them, D in and D out B represents the input data volume and the result data volume of task i, respectively. i,n The effective bandwidth of the network link allocated to task i, where τ0 is the fixed propagation delay; ； Among them, Q n Let μ be the total number of tasks currently pending for candidate node n. n The average processing rate of node n; ； Among them, C i f is the number of computation cycles required for task i. n The available computational frequency for node n.

5. A cloud-edge collaborative computing system for a digital factory of home elevators according to claim 2, characterized in that: The estimated energy cost E i The calculation formula is: ； in, ; P total P represents the total power consumption of candidate edge computing node n when it performs a task. static Let k be the static power consumption of node n. n The effective capacitance coefficient, f, associated with the node n chip architecture n Its operating frequency; ρt is the unit energy price coefficient at the current moment; T ei This represents the computation execution latency of task i on the candidate edge computing node.

6. A cloud-edge collaborative computing system for a digital factory of home elevators according to claim 2, characterized in that, The security confidence score S i The calculation formula is: ； Where: J is the total number of preset security assessment dimensions; j is the security dimension index; Vji is the required threshold for task i in the j-th security dimension; Vjn is the guarantee capability provided by candidate edge computing node n in the j-th security dimension; ω j Let be the weight coefficient of the j-th dimension, and satisfy . .

7. A cloud-edge collaborative computing system for a digital factory of home elevators according to claim 1, characterized in that, The dynamic weighting coefficient λ t , λ e With λ s The adjustment method is as follows: Obtain the basic weights based on the current production stage strategy configuration; λbase t is the base weight for time delay, λbase e is the base weight for energy consumption, and λbase s is the base weight for security items; Fine-tuning is performed based on the real-time system status, using the following formula: ； ； ； Among them, L c L represents the current average network load rate. avg Its historical average; Z c Z represents the current level of network threat detected. avg It is its historical average level; α and β are sensitivity adjustment coefficients.

8. A cloud-edge collaborative computing system for a digital factory of home elevators according to claim 1, characterized in that: The factory edge computing platform includes a fault tolerance and migration module. When a failure is detected in the target edge computing node, this module automatically executes the following fault tolerance and migration process: Step S01: Immediately freeze the state and execution context of all atomic computation tasks currently being executed on the faulty node; Step S02: Based on the latest system resource status and node load information, rerun the dynamic unloading decision algorithm to calculate and select a new optimal target edge computing node for each frozen task; Step S03: Completely migrate the frozen task state image and related context data to the newly selected edge computing node, and resume task execution from the point of interruption.

9. A cloud-edge collaborative computing system for a digital factory of home elevators according to claim 1, characterized in that, The cloud computing center and the factory edge computing platform adopt a model collaborative update mechanism based on federated learning. The operation process of this mechanism is as follows: The factory edge computing platform periodically collects the model gradients generated by each edge computing node under its jurisdiction during local training, and adds random noise that conforms to the differential privacy standard for privacy processing before the gradients are uploaded. On the platform side, multiple local gradients that have undergone privacy processing are securely aggregated, and the aggregated gradient is then uploaded to the cloud computing center. The cloud computing center receives aggregated gradients from multiple factory edge computing platforms and performs a weighted average of the aggregated gradients based on the proportion of data contributed by each factory, thereby updating the parameters of the global artificial intelligence model.

10. A cloud-edge collaborative computing system for a digital factory of home elevators according to claim 1, characterized in that, When the factory edge computing platform decomposes complex production tasks into atomic computing tasks, it comprehensively considers the following dimensions based on task classification and priority rules: The real-time level of a task is determined based on the maximum allowed end-to-end response time threshold. The more urgent the response time requirement, the higher the real-time level. The computational complexity of a task is quantified by the scale of computational operations required to perform that task. The security level of task data is classified according to the sensitivity of the production data processed by the task; The overall priority of a task is calculated by weighting the above-mentioned indicators. Tasks with high real-time requirements, high computational complexity, and high data security level will be given higher overall priority and will be allocated computing resources in subsequent scheduling.

Images