Multi-cloud cooperative system based on edge computing

Abstract

The invention relates to the technical field of cloud computing and edge computing, in particular to a multi-cloud cooperative system based on edge computing. Comprising a data acquisition module for acquiring communication characteristic parameters and a current control topology state; the first processing module is used for identifying intermittent communication disturbance and calling a digital twinborn model to predict and control topological entropy increment; the second processing module is used for determining a lagging failure risk value for keeping the current control right affiliation unchanged; the third processing module is used for calculating an entropy increase-risk ratio for measuring the survival stability of the system; and the cooperative control module is used for generating a degradation operation instruction for switching the control node or maintaining the current connection based on a comparison result of the entropy increase-risk ratio and a survival threshold value. According to the method, routing oscillation and control right split brain are effectively avoided, physical collision accidents are prevented, and the survival probability of the system in a severe network environment is improved.

Description

Technical Field

[0001] This invention relates to the fields of cloud computing and edge computing technology, specifically to a multi-cloud collaborative system based on edge computing. Background Technology

[0002] In cross-regional industrial control and edge computing collaborative application scenarios, edge execution units rely on real-time interaction with remote cloud control nodes to obtain high-precision instructions. The system usually needs to combine multi-cloud architecture and network situational awareness technology to ensure the survivability of the control link and the timeliness of instruction issuance. For high availability assurance in heterogeneous network environments, existing solutions generally adopt a linear primary / backup failover architecture based on heartbeat detection. This involves simply monitoring link connectivity and, upon detecting packet loss or timeout, directly transferring full control to a backup node to mitigate the failure. While this approach is effective in hard failure scenarios involving complete physical link outages, it lacks a quantitative assessment mechanism for network gray-scale failures and the cost of switching. When encountering intermittent communication disturbances or link congestion, traditional linear logic is prone to frequent routing oscillations and control splits due to oversensitivity to network jitter, leading to an unstable and chaotic system. Furthermore, the lack of mathematical prediction of the risks of maintaining the current connection causes the system to forcibly maintain interaction even when it should degrade due to congestion, resulting in command arrival times exceeding the physical device's safety window and causing serious delayed failures. Therefore, establishing an adaptive decision-making mechanism based on entropy increase and risk game theory, which can accurately identify sub-healthy network states while balancing the topological chaos costs of control switching with the delayed failure risks of maintaining the status quo, has become an urgent technical problem to be solved.

[0003] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention discloses a multi-cloud collaborative system based on edge computing. Specifically, the technical solution of this invention is as follows: The data acquisition module is used to acquire communication characteristic parameters and current control topology status between the edge execution unit and multiple heterogeneous remote control nodes. The communication characteristic parameters include: heartbeat packet loss rate, API response delay jitter variance, routing hop count and link congestion. The first processing module is used to identify whether there is intermittent communication disturbance in the heterogeneous remote control node based on the heartbeat packet loss rate and the API response delay jitter variance, and when there is intermittent communication disturbance, to call the digital twin model to simulate the process of executing the control switch action, so as to predict the increase in control topology entropy that characterizes the change in control topology orderliness. The second processing module is used to determine, based on the link congestion level and the current control topology state, the risk value of hysteresis failure where the command state is inconsistent between the edge execution unit and the heterogeneous remote control node, while maintaining the current control ownership unchanged; The third processing module is used to calculate the entropy increase-risk ratio, which is used to measure the survivability stability of the system, based on the increase in the control topology entropy and the hysteresis failure risk value. The collaborative control module is used to generate adaptive collaborative control instructions for the edge execution unit based on the comparison result between the entropy increase-risk ratio and the preset survival threshold. The adaptive collaborative control instructions include switching control node instructions or degraded operation instructions to maintain the current connection.

[0005] Preferably, based on the heartbeat packet loss rate and the API response latency jitter variance, identifying whether there is intermittent communication disturbance in the heterogeneous remote control nodes includes: The heartbeat packet loss rate and the API response latency jitter variance were invoked. Construct an anomaly detection model based on the isolated forest algorithm; The heartbeat packet loss rate and the API response latency jitter variance are input into the anomaly detection model, and an anomaly score is output. In response to the abnormal score being between a preset normal threshold and a downtime threshold, it is determined that the heterogeneous remote control node has intermittent communication disturbances, and this state is defined as a gray-scale fault state; wherein, the downtime threshold is greater than the normal threshold.

[0006] Preferably, the increase in control topology entropy resulting from the execution of control handover actions, based on a digital twin model, includes: Collect the amount of metadata fragments of the edge execution unit and the routing complexity of the alternative control nodes; The process of transferring control from the current control node to a candidate control node is simulated in the digital twin model. The number of routing oscillations and metadata reorganization costs generated during the statistical simulation process; The control topology entropy increment generated by performing the control handover action is calculated by weighted summation of the number of routing oscillations and the metadata reorganization overhead.

[0007] Preferably, based on the link congestion level and the current control topology state, a hysteresis failure risk value is determined for the occurrence of instruction state inconsistency between the edge execution unit and the heterogeneous remote control node while maintaining the current control ownership unchanged. This includes: The link congestion level and the safe window time for instruction execution of the edge execution unit are invoked; The corresponding network transmission delay distribution model is determined based on the link congestion level; Calculate the probability integral of the delay time being greater than the instruction execution safety window time in the network transmission delay distribution model; The probability integral is defined as the hysteresis failure risk value under the condition that the current ownership of control remains unchanged.

[0008] Preferably, the entropy increase-risk ratio, used to measure the survivability stability of the system, is calculated based on the increase in entropy of the control topology and the hysteresis failure risk value, including: Invoke the control topology entropy increment and the hysteresis failure risk value; The control topology entropy increment is normalized to obtain the switching cost factor; The hysteresis failure risk value is weighted to obtain the maintenance risk factor; Calculate the quotient obtained by dividing the switching cost factor by the maintenance risk factor, and determine the quotient as the entropy increase-risk ratio.

[0009] Preferably, based on the comparison result between the entropy increase-risk ratio and the preset survival threshold, adaptive cooperative control instructions are generated for the edge execution unit, including: Invoke the entropy increase-risk ratio and the preset survival threshold; In response to the entropy increase-risk ratio being less than the preset survival threshold, a switching control node instruction is generated to trigger the edge execution unit to transfer control to an alternative heterogeneous remote control node. In response to the entropy increase-risk ratio being greater than or equal to the preset survival threshold, a degraded operation instruction is generated to maintain the current connection, thereby locking the current control topology state and entering a local autonomous mode.

[0010] Preferably, the downgrade operation instruction is specifically configured as follows: The edge execution unit is prohibited from performing any routing operations that change control attribution; Start the local cached instruction queue of the edge execution unit; The instruction execution cycle of the edge execution unit is adjusted to a preset safety redundancy cycle until the intermittent communication disturbance disappears; wherein the duration of the safety redundancy cycle is longer than the instruction execution cycle of the edge execution unit in normal communication state.

[0011] Preferably, the edge execution unit is an industrial robot controller, the heterogeneous remote control node is a public cloud or private cloud server distributed in different geographical areas, and the control topology entropy increment is used to characterize the ambiguity of control ownership caused by routing oscillations during cross-cloud switching.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention utilizes an unsupervised learning algorithm based on isolated forests and dynamic dual-threshold logic to achieve accurate perception and definition of network sub-health status. Unlike traditional solutions that rely solely on linear judgment based on connectivity detection, this solution inputs the heartbeat packet loss rate and API response latency jitter variance into the anomaly detection model, and ensures through dynamic boundary constraint logic that the normal threshold is always lower than the crash threshold, thereby effectively distinguishing between normal network fluctuations, complete crashes, and intermittent communication disturbances. This mechanism avoids misjudgments caused by the failure of fixed thresholds in harsh network environments, provides a precise trigger signal for whether to initiate digital twin simulation, and solves the problem of traditional methods being overly sensitive to network jitter.

[0013] 2. This invention introduces a control topology entropy increase prediction mechanism based on digital twin simulation, quantifying the system chaos cost caused by cross-cloud switching actions; by mapping physical heterogeneous networks into virtual directed graphs and running a virtual BGP routing convergence algorithm controlled by routing complexity parameters, the system can pre-simulate the number of routing oscillations and metadata reorganization overhead during the control transfer process; this method breaks the blind logic of switching upon failure in traditional multi-cloud disaster recovery, effectively preventing control split-brain and communication storms caused by frequent routing switching during network oscillations, and ensuring the topological orderliness of the system in cross-regional scheduling.

[0014] 3. This invention constructs a hysteresis failure risk assessment model based on probability distribution integral, clarifying the impact boundary of maintaining the current connection on the security of physical devices; the system determines the distribution model of network transmission delay based on link congestion and calculates the probability integral of the delay time exceeding the instruction execution safety window, thereby accurately quantifying the risk of instruction state inconsistency that may be caused by not switching; this scheme not only focuses on the average delay, but also on the probability of long-tail delay breaking through the physical safety window, effectively avoiding physical accidents such as industrial robot collisions caused by slow response leading to late instruction arrival.

[0015] 4. This invention establishes a game-theoretic decision-making system and a degraded autonomous survival strategy based on the entropy increase-risk ratio, which significantly improves the system's survival probability in extreme network environments. By calculating the ratio of switching costs to maintenance risks, the system can make the optimal choice between cutting off an arm to survive and surviving with defects. In particular, when the switching cost is too high, by generating degraded operation instructions, the current topology is forcibly locked, the local cache queue is started, and the instruction execution cycle is lengthened, using time to trade for space stability, avoiding systemic avalanche caused by over-automated decision-making, and realizing the extreme survival of industrial control systems in gray-scale fault scenarios. Attached Figure Description

[0016] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0018] Example 1: Please see Figure 1 Multi-cloud collaborative systems based on edge computing include: The data acquisition module is used to acquire communication characteristic parameters and current control topology status between the edge execution unit and multiple heterogeneous remote control nodes. The communication characteristic parameters include: heartbeat packet loss rate, API response latency jitter variance, routing hop count and link congestion. The first processing module is used to identify whether there are intermittent communication disturbances in heterogeneous remote control nodes based on the heartbeat packet loss rate and API response delay jitter variance. When there are intermittent communication disturbances, it calls the digital twin model to simulate the process of executing the control switch action, so as to predict the increase in control topology entropy that represents the change in control topology order. The second processing module is used to determine the risk value of delayed failure due to inconsistency of command state between the edge execution unit and the heterogeneous remote control node, based on the link congestion degree and the current control topology state, while maintaining the current control ownership unchanged. The third processing module is used to calculate the entropy increase-risk ratio, which is used to measure the survivability and stability of the system, based on the increase in control topology entropy and the risk value of hysteresis failure. The collaborative control module is used to generate adaptive collaborative control instructions for edge execution units based on the comparison results between the entropy increase-risk ratio and the preset survival threshold. The adaptive collaborative control instructions include instructions to switch control nodes or instructions to maintain the current connection and perform degraded operation. This embodiment details the core architecture and operating logic of the system, aiming to solve the decision-making challenges caused by asymmetric network interference in cross-regional industrial control scenarios. The data acquisition module is deployed on the edge gateway side, and builds a full-dimensional network situational awareness capability through the SNMP protocol and custom probes to obtain the communication characteristic parameters between the edge execution unit and the heterogeneous remote control nodes distributed on AWS, Azure and private clouds in real time. The first processing module identifies whether there is a gray-scale fault that is not a complete shutdown, i.e., intermittent communication disturbance, based on the obtained heartbeat packet loss rate and API response delay jitter variance. When a disturbance is identified, the digital twin model is called to simulate the switching process in virtual space, thereby predicting the increase in control topology entropy. The second processing module quantifies the risk of delayed failure caused by network congestion leading to instructions exceeding the physical device's safety window, without switching, based on link congestion and the current control topology status. Based on this, the third processing module calculates the entropy increase-risk ratio according to the increase in control topology entropy and the risk value of hysteresis failure, and constructs a mathematical benchmark for dynamic decision-making; the collaborative control module generates adaptive collaborative control instructions that include switching control nodes or maintaining the current connection in degraded operation based on the comparison result of the entropy increase-risk ratio and the preset survival threshold. This embodiment breaks the linear logic of switching upon failure in traditional multi-cloud disaster recovery by introducing a game-theoretic calculation mechanism that controls the increase of topology entropy and the risk value of hysteresis failure. In the face of gray-scale failure scenarios common in the industrial Internet, the system effectively avoids routing oscillations and split-brain control caused by overly sensitive switching, while preventing physical collision accidents caused by slow response, thus improving the survival probability of multinational industrial control systems in harsh network environments.

[0019] Example 2: Based on heartbeat packet loss rate and API response latency jitter variance, identify whether intermittent communication disturbances exist in heterogeneous remote control nodes, including: The heartbeat packet loss rate and API response latency jitter variance are used to construct an anomaly detection model based on the isolated forest algorithm. The heartbeat packet loss rate and API response latency jitter variance are input into the anomaly detection model, and the anomaly score is output. In response to an anomaly score falling between a preset normal threshold and a crash threshold, it is determined that there is intermittent communication disturbance in the heterogeneous remote control node, and this state is defined as a gray-scale fault state; where the crash threshold is greater than the normal threshold. This embodiment is a further specification of the process of identifying intermittent communication disturbances, which utilizes an unsupervised learning algorithm to process outlier detection in high-dimensional data; Key indicators of network sub-health status in system calls are heartbeat packet loss rate and API response latency jitter variance; An anomaly detection model based on the Isolation Forest algorithm is constructed. This model is trained using historical normal communication data to build multiple binary isolation trees. Then, the above communication feature parameters are input into the model, the average path length of data points in the isolation trees is calculated, and anomaly scores are output. The system executes gray-scale fault determination logic, responding when the anomaly score falls between a preset normal threshold and a crash threshold, where the normal threshold... The 99th percentile value of the distribution of outlier scores in the historical normal data of the training set, such as 0.65, is used to define the boundary of statistically significant outliers. To prevent excessive fluctuations in historical data under poor network conditions... This renders the grayscale fault determination range ineffective. This embodiment introduces dynamic boundary constraint logic: real-time detection. and The difference, if Then force Revised to ,in A preset minimum grayscale bandwidth, such as 0.05, is used to ensure that the normal threshold is always strictly less than the downtime threshold, guaranteeing the physical feasibility of the judgment logic in any network environment; while the downtime threshold... Defined as 0.95, this value is determined based on the theoretical score lower bound when the physical link is completely interrupted, thus clarifying the numerical boundary of the system startup simulation, determining that the node is in an intermittent communication disturbance state where the connection is not broken but the performance is severely degraded, and defining it as a gray-scale fault state. This embodiment utilizes the isolated forest algorithm and dual threshold design to accurately capture the sub-healthy state of the network. This method effectively distinguishes between three states: normal network fluctuations, complete downtime, and gray-scale faults. It avoids the misjudgment of the traditional fixed threshold method when facing complex network jitter, and provides a precise trigger signal for whether to start the expensive digital twin simulation, ensuring the efficient use of computing resources.

[0020] Example 3: The prediction of the control topology entropy increase caused by the execution of control handover actions based on the digital twin model includes: collecting the amount of metadata fragments of the edge execution unit and the routing complexity of the alternative control nodes; simulating the process of migrating control from the current control node to the alternative control node in the digital twin model; counting the number of routing oscillations and metadata reorganization costs generated during the simulation; and calculating the control topology entropy increase caused by the execution of control handover actions by weighted summation of the number of routing oscillations and metadata reorganization costs. This embodiment further specifies the process of predictive control topology entropy increment, aiming to quantify the systemic disorder caused by the switching action; the system collects the amount of metadata fragments that the edge execution unit needs to synchronize across clouds during the switching process, as well as the routing complexity of the network topology from the edge to the alternative cloud node; here it is clarified that the routing complexity is a dimensionless parameter obtained by normalizing the number of routing hops obtained by the data acquisition module, specifically defined as the ratio of the number of routing hops to the network diameter, used to uniformly measure the depth of heterogeneous network structures. To address the issue that scanning only active links might cause the network diameter to equal the number of routing hops, thus rendering the ratio invalid, the calculation scope of the network diameter here is explicitly defined as: the diameter of the largest connected subgraph stored in the digital twin model database, which includes all discovered nodes in the area where the current control node and the candidate control node are located, i.e., the maximum value of the shortest path between any two nodes in the graph, thereby ensuring that this parameter can effectively reflect the overall scale characteristics of the potential network topology. The control transfer process is simulated in the constructed digital twin model. This digital twin model is constructed using virtual topology mapping technology, specifically mapping routing nodes of the physical heterogeneous network to vertices in a virtual directed graph, mapping physical links to edges, and mapping real-time link congestion to the dynamic impedance weights of the edges. The specific calculation formula for this mapping is as follows: in, The preset numerical stability constant is set to a value of [value missing]. This boundary protection logic is used to prevent link congestion. Approaching 1.0, that is, when there is complete congestion, the floating-point number overflow or infinity anomaly caused by the denominator approaching zero is ensured to ensure that the calculation of virtual path weights is bounded in the entire domain. This is the impedance coefficient, typically 5.0; in the routing logic of the virtual BGP algorithm, this impedance weight... It is added to the length value of the AS-Path path attribute, i.e., the virtual path length. This causes highly congested links to be treated as longer paths and thus have lower priority in the BGP routing algorithm; In this process, the collected route complexity is used as an environmental parameter input model and mapped to the route convergence damping coefficient in the virtual BGP algorithm; the specific mapping function expression is as follows: ,in To normalize routing complexity, The baseline penalty value, such as 500. This is an adjustment factor, such as 2.0; the symbol is deliberately used here. To distinguish it from the expression of displacement parameters This ensures the uniqueness and rigor of the mathematical symbols defined throughout the text; the expression quantitatively specifies that the larger the routing complexity value, the longer the routing node suppresses path changes in the simulation environment; In order to make this coefficient This embodiment clarifies that it can be recognized by the standard BGP protocol stack. Mapping relationship with the half-life parameter of the RouteFlapDamping algorithm as defined in RFC2439: in, Using a base time constant, such as 5 minutes, this means that the calculated The value is transformed into a specific half-life time value through a logarithmic function, which directly corrects the virtual router's forgetting speed of route rollover behavior, thereby directly affecting the oscillation convergence speed in the simulation results through algorithm logic; During the simulation, a random packet loss sequence consistent with the identified communication disturbance characteristics is injected into the virtual edges of the current connection. This sequence is generated using a Gilbert-Elliott two-state Markov model, and its center-hop packet loss rate is... Used to define the state transition probability matrix of the model, satisfying , This represents the probability of transitioning from a good state to a bad state. The probability of transitioning from a bad state to a good state; used to solve the underdetermined problem of the equation; This embodiment introduces the average burst length. As the second constraint, define Based on historical statistics, the value is preset to 3.0, and thus the simultaneous solution yields... and The state transition matrix is ​​fully defined, and the API response delay jitter variance is used to set the additional Gaussian white noise variance parameter in bad states. This transforms the scalar characteristics into a packet loss simulation process with time-dependent characteristics, and runs a virtual BGP routing convergence algorithm controlled by the routing complexity parameter to reproduce the path selection behavior of the control flow between the primary and backup nodes. The simulation window counts the number of routing oscillations caused by asynchronous routing table convergence, and it also calculates the metadata reorganization cost through end-to-end simulation. This cost is calculated by performing variable mapping to determine the amount of fragmented input metadata. The unit is bytes. Convert it into a virtual data packet sequence. Set the MTU of the virtual transport protocol to 1500 bytes and the protocol header overhead to 40 bytes. Calculate the number of virtual data packets: The payload size of a single data packet is 1460 bytes; this is simulated in the digital twin model. The encapsulation, transmission, and reception process of each virtual data packet is analyzed. The trajectory of each virtual data packet is tracked, and the out-of-order arrival time difference caused by simulated network jitter is calculated. The buffer holding time required to complete sequence reassembly while waiting for missing fragments is calculated in the reassembly buffer at the virtual receiver. The sum of this dynamic buffer holding time and transmission delay is calculated as the metadata reassembly overhead. The control topology entropy increment is calculated according to the following formula: in: The source is historical failure case statistics. Its physical meaning is the routing stability weight coefficient, and its value includes a unit conversion factor. The unit is 1 / time, which is used to eliminate the number of routing oscillations. The physical dimensions of the unit; its specific numerical value is determined by counting the cumulative number of communication interruptions caused by routing oscillations in the network area where the edge execution unit is located over the past 365 days. And according to the formula Calculated; It is hereby clarified that the constant 0.5 in the formula has a physical dimension, namely 1 / degree, to ensure that the calculation result on the right side of the equation is consistent with that on the left side. The dimensions remain consistent; this formula shows that the more frequent the historical failures, the greater the weight of the contribution of routing oscillations to entropy increase. The source is the business system rating, and its physical meaning is the data accuracy sensitivity weighting coefficient. Its value includes a unit conversion factor, and the unit is 1 / millisecond. It is used to eliminate metadata reorganization overhead. The physical dimensions, thus the result This is a dimensionless value; its specific value is determined by consulting the business level mapping table. Set the data consistency level of the business system Levels range from 1 to 5, with 1 being the lowest and 5 the highest. The values ​​are 0.1, 0.2, 0.4, 0.8, and 1.6; for example, when the current service level is L=4, The value is 0.8; The source is the output of digital twin simulation. Its numerical distribution characteristics are affected by the input routing complexity parameter. The physical meaning is the number of routing oscillations generated by the simulation, and the unit is times. The source is dynamic statistics from digital twin simulation; the physical meaning is metadata reorganization overhead, and the unit is milliseconds. Example 4: Based on link congestion and the current control topology, the risk value of delayed failure due to inconsistency in instruction state between the edge execution unit and the heterogeneous remote control node is determined while maintaining the current control ownership. This includes: invoking link congestion and the instruction execution safety window time of the edge execution unit; determining the corresponding network transmission delay distribution model based on link congestion; calculating the probability integral of the delay time being greater than the instruction execution safety window time in the network transmission delay distribution model; and defining the probability integral as the risk value of delayed failure under the condition of maintaining the current control ownership. This embodiment further specifies the process of determining the risk value of hysteresis failure, focusing on the impact of the long-tail distribution of network latency on physical security; the system call link congestion degree and the instruction execution safety window time provided by the equipment manufacturer's technical manual; then, based on the link congestion degree, a shifted gamma distribution model describing the network latency characteristics under congestion conditions is determined. Calculate the probability integral of the delay time exceeding the instruction execution safety window time in this distribution model, taking into account the probability density function. The domain is restricted to In order to avoid when Errors in complex number operations caused by exceeding the lower limit of integration should be addressed by taking the physical minimum delay as the lower limit of integration. With security window The larger value, the corrected calculation formula is as follows: in, The source is the result of integral calculation, and its physical meaning is the risk value of delayed failure; The source is a preset parameter, which is the instruction execution safety window time, in milliseconds; The source is a probabilistic model, representing the current link congestion level. The network transmission delay probability density function is defined mathematically as follows: In this distributed model, the parameters are related to the link congestion. The specific functional relationship is as follows, with a value range of [0,1]: Displacement parameters : Represents the theoretical minimum latency of the current physical link. ,in This is the baseline physical delay constant under congestion-free conditions. Its value is derived and calculated by automatically retrieving all idle time periods (i.e., link congestion levels) of the edge execution unit and the current control node over the past 30 days. Historical heartbeat round-trip time data were used to calculate the arithmetic mean after removing outliers in the upper and lower 5%, and 50% of this mean was determined as the one-way physical baseline delay. This ensures that the lower limit of the integration score is set based on the actual physical link capabilities. Shape parameters : Indicates the central tendency of the delayed distribution. ,in This is the initial shape factor, typically 2.0; Scale parameters : Indicates the degree of dispersion of the delay. ,in, This is the initial scaling factor, typically 5ms. For the gamma function, in the formula Shape parameters as defined above For the same variable; the calculated probability integral is defined as the hysteresis failure risk value; This embodiment uses probability theory to accurately quantify the potential danger of the no-switching strategy. The method not only focuses on the average delay, but also on whether the long tail of the delay distribution has breached the safety window of the physical device. By clarifying the mapping relationship between congestion and distribution parameters, it provides a risk basis for decision-making that meets the real-time requirements of industry, and prevents physical safety accidents caused by late arrival of instructions.

[0021] Example 5: Based on the increase in control topology entropy and the risk value of hysteresis failure, the entropy increase-risk ratio used to measure the survivability and stability of the system is calculated, including: calling the increase in control topology entropy and the risk value of hysteresis failure; normalizing the increase in control topology entropy to obtain the switching cost factor; weighting the risk value of hysteresis failure to obtain the maintenance risk factor; calculating the quotient obtained by dividing the switching cost factor by the maintenance risk factor, and determining the quotient as the entropy increase-risk ratio. This embodiment further specifies the process of calculating the entropy increase-risk ratio, aiming to construct a unified decision-making dimension. The system calls the control topology entropy increase and hysteresis failure risk value calculated in the preceding steps. The Min-Max standardization method is used to normalize the control topology entropy increase, mapping it to the zero-to-one interval to obtain the switching cost factor. Considering the differences in failure tolerance among different business scenarios, the hysteresis failure risk value is weighted to obtain the maintenance risk factor. The entropy increase-risk ratio is calculated according to the following formula: in, The source is the calculation result, and the physical meaning is entropy increase minus risk ratio; The source is normalized, and its physical meaning is the switching cost factor, which is calculated using the following formula: ; Regarding the current entropy increase in real-time stream computing It may break the historical extreme value, that is or ,lead to For issues exceeding the [0,1] interval, the system adds numerical truncation and dynamic update logic: if the calculation result... Then a mandatory order ; like Then a mandatory order Simultaneously, triggering an asynchronous background process will resolve the current out-of-bounds situation. The value is written to the historical database to update the next calculation cycle. or This ensures the adaptability of the normalized benchmark; in this calculation, the system has built-in circuit breaker protection logic for a denominator of zero: If detected ,in This is the minimum value of machine precision, such as This indicates that the system is in an extreme steady state with a constant historical entropy value or in the early stages of a cold start. At this point, a forced command... This means balancing default values ​​to avoid division by zero anomalies and maintain the continuity of subsequent ratio calculations; where... Add value to the currently predicted control topology entropy. and These are the historical maximum entropy increase and historical minimum entropy increase recorded by the system within the past 24-hour sliding window, respectively. The source is weighted processing, and its physical meaning is the maintenance risk factor. Its calculation formula is: ;in The value of the hysteresis failure risk calculated above. This is the business importance weighting coefficient, ranging from 1.0 to 3.0. It is determined by looking up a table based on the physical security level involved in the edge execution unit, with a larger coefficient for higher security levels. The source is a preset constant, and its physical meaning is to prevent the minimum value where the denominator is zero. It is defined as... Dimensionless values ​​with the same dimensions are used to ensure consistency in the physical meaning of addition operations in the denominator; typical values ​​are [value missing]. ; The entropy increase-risk ratio proposed in this embodiment intuitively reflects the chaotic cost that must be paid to eliminate the current lag risk system. As a highly condensed decision indicator, this ratio can quantify whether switching is worthwhile. That is, the smaller the ratio, the higher the benefit of switching, and the larger the ratio, the better it is to maintain the status quo, even though there is risk. This provides scientific mathematical support for subsequent adaptive control.

[0022] Example 6: Based on the comparison between the entropy increase-risk ratio and a preset survival threshold, adaptive collaborative control instructions are generated for the edge execution unit, including: invoking the entropy increase-risk ratio and the preset survival threshold; in response to the entropy increase-risk ratio being less than the preset survival threshold, generating a control node switching instruction to trigger the edge execution unit to transfer control to an alternative heterogeneous remote control node; in response to the entropy increase-risk ratio being greater than or equal to the preset survival threshold, generating a degraded operation instruction to maintain the current connection, thereby locking the current control topology state and entering a local autonomous mode. This embodiment further specifies the process of generating adaptive cooperative control instructions, establishing the system's survival strategy logic. The system calls the entropy increase-risk ratio calculated in real time and the preset survival threshold derived from regression analysis of historical operating data. The comparison and decision logic is executed: in response to the entropy increase-risk ratio being less than the survival threshold, indicating that the switching cost is lower than the imminent failure risk, the system generates a switching control node instruction, triggering the edge execution unit to transfer control to the alternative node; conversely, in response to the entropy increase-risk ratio being greater than or equal to the survival threshold, indicating that the topology chaos caused by the switching action is extremely high, the system generates a degraded operation instruction to maintain the current connection, locks the current control topology state, and enters a local autonomous mode. This embodiment implements a rational survival strategy in harsh network environments; the system no longer blindly pursues the optimal connection, but makes the choice most conducive to the system's continuity between cutting off an arm to survive and surviving with defects based on the real-time cost-risk ratio, effectively avoiding systemic avalanche caused by over-automated decision-making.

[0023] Example 7: The specific configuration of the downgrade operation instruction is as follows: prohibit the edge execution unit from performing any routing addressing operations that change control ownership; start the local cache instruction queue of the edge execution unit; adjust the instruction execution cycle of the edge execution unit to a preset safety redundancy cycle until the intermittent communication disturbance disappears; wherein, the duration of the safety redundancy cycle is longer than the instruction execution cycle of the edge execution unit in normal communication state; This embodiment further specifies the configuration of the downgraded operation instructions, and describes in detail the execution details of the local autonomy mode; the system forcibly prohibits the edge execution unit from performing any routing addressing operations that change control ownership, freezes the current topology at the network level, and prevents routing oscillations caused by the automatic reconnection mechanism due to heartbeat loss; it starts the local cached instruction queue of the edge execution unit and executes the pre-downloaded standard action sequence that does not rely on real-time cloud interaction; it adjusts the instruction execution cycle of the edge execution unit to a preset safety redundancy cycle until the intermittent communication disturbances disappear, and this safety redundancy cycle is significantly longer than the instruction execution cycle under normal communication conditions; This embodiment uses a combination of topology locking and slow-down operation to trade time for spatial stability. This slow-down strategy artificially extends the tolerance window for network commands, ensuring that physical devices will not go out of control and collide due to the momentary interruption of the command queue during gray-scale failures, thus achieving industrial-grade extreme survivability.

[0024] Example 8: The edge execution unit is an industrial robot controller, and the heterogeneous remote control node is a public cloud or private cloud server distributed in different geographical areas. The control topology entropy increment is used to characterize the ambiguity of control ownership caused by routing oscillation during cross-cloud switching. This embodiment defines the specific physical carrier and application background of the system; clarifies that the edge execution unit is specifically an industrial robot controller with extremely high real-time requirements, such as a six-axis robotic arm controller deployed on an automobile assembly line; defines heterogeneous remote control nodes as public or private cloud servers distributed in different geographical regions, such as AWS Frankfurt data center and Azure Ireland data center; and defines the control topology entropy increment to characterize the transient masterless or dual-master state caused by DNS resolution delay and BGP routing convergence in a multi-cloud environment, i.e., the ambiguity of control ownership. This embodiment clarifies the applicability of the technical solution in addressing the unique issue of ambiguous control ownership in cross-border and cross-cloud industrial internet scenarios. By specifying physical objects, it emphasizes the necessity of the system in handling high-precision, high-risk industrial control tasks, and demonstrates its control capabilities for extreme risk scenarios where two clouds may simultaneously command a robot or an unmanned robot.

[0025] 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.

Claims

1. A multi-cloud collaboration system based on edge computing, characterized in that, include: The data acquisition module is used to acquire communication characteristic parameters and current control topology status between the edge execution unit and multiple heterogeneous remote control nodes. The communication characteristic parameters include: heartbeat packet loss rate, API response delay jitter variance, routing hop count and link congestion. The first processing module is used to identify whether there is intermittent communication disturbance in the heterogeneous remote control node based on the heartbeat packet loss rate and the API response delay jitter variance, and when there is intermittent communication disturbance, to call the digital twin model to simulate the process of executing the control switch action, so as to predict the increase in control topology entropy that characterizes the change in control topology orderliness. The second processing module is used to determine, based on the link congestion level and the current control topology state, the risk value of hysteresis failure where the command state is inconsistent between the edge execution unit and the heterogeneous remote control node, while maintaining the current control ownership unchanged; The third processing module is used to calculate the entropy increase-risk ratio, which is used to measure the survivability stability of the system, based on the increase in the control topology entropy and the hysteresis failure risk value. The collaborative control module is used to generate adaptive collaborative control instructions for the edge execution unit based on the comparison result between the entropy increase-risk ratio and the preset survival threshold. The adaptive collaborative control instructions include switching control node instructions or degraded operation instructions to maintain the current connection.

2. The edge-computing-based multi-cloud collaboration system of claim 1, wherein, Based on the heartbeat packet loss rate and the API response latency jitter variance, identify whether the heterogeneous remote control nodes experience intermittent communication disturbances, including: The heartbeat packet loss rate and the API response latency jitter variance were invoked. Construct an anomaly detection model based on the isolated forest algorithm; The heartbeat packet loss rate and the API response latency jitter variance are input into the anomaly detection model, and an anomaly score is output. In response to the abnormal score being between a preset normal threshold and a downtime threshold, it is determined that the heterogeneous remote control node has intermittent communication disturbances, and this state is defined as a gray-scale fault state; wherein, the downtime threshold is greater than the normal threshold.

3. The edge-computing-based multi-cloud collaboration system of claim 1, wherein, Based on the digital twin model, the increase in control topology entropy resulting from the execution of control handover actions is predicted, including: Collect the amount of metadata fragments of the edge execution unit and the routing complexity of the alternative control nodes; The process of transferring control from the current control node to a candidate control node is simulated in the digital twin model. The number of routing oscillations and metadata reorganization costs generated during the statistical simulation process; The control topology entropy increment generated by performing the control handover action is calculated by weighted summation of the number of routing oscillations and the metadata reorganization overhead.

4. The multi-cloud collaborative system based on edge computing according to claim 1, characterized in that, Based on the link congestion level and the current control topology state, a hysteresis failure risk value is determined for the edge execution unit and the heterogeneous remote control node, under the condition of maintaining the current control ownership unchanged, including: The link congestion level and the safe window time for instruction execution of the edge execution unit are invoked; The corresponding network transmission delay distribution model is determined based on the link congestion level; Calculate the probability integral of the delay time being greater than the instruction execution safety window time in the network transmission delay distribution model; The probability integral is defined as the hysteresis failure risk value under the condition that the current ownership of control remains unchanged.

5. The multi-cloud collaborative system based on edge computing according to claim 1, characterized in that, Based on the increase in entropy of the control topology and the hysteresis failure risk value, the entropy increase-risk ratio, used to measure the survivability stability of the system, is calculated, including: Invoke the control topology entropy increment and the hysteresis failure risk value; The control topology entropy increment is normalized to obtain the switching cost factor; The hysteresis failure risk value is weighted to obtain the maintenance risk factor; Calculate the quotient obtained by dividing the switching cost factor by the maintenance risk factor, and determine the quotient as the entropy increase-risk ratio.

6. The multi-cloud collaborative system based on edge computing according to claim 1, characterized in that, Based on the comparison between the entropy increase-risk ratio and the preset survival threshold, adaptive cooperative control instructions are generated for the edge execution unit, including: Invoke the entropy increase-risk ratio and the preset survival threshold; In response to the entropy increase-risk ratio being less than the preset survival threshold, a switching control node instruction is generated to trigger the edge execution unit to transfer control to an alternative heterogeneous remote control node. In response to the entropy increase-risk ratio being greater than or equal to the preset survival threshold, a degraded operation instruction is generated to maintain the current connection, thereby locking the current control topology state and entering a local autonomous mode.

7. The multi-cloud collaborative system based on edge computing according to claim 6, characterized in that, The specific configuration of the downgrade operation command is as follows: The edge execution unit is prohibited from performing any routing operations that change control attribution; Start the local cached instruction queue of the edge execution unit; The instruction execution cycle of the edge execution unit is adjusted to a preset safety redundancy cycle until the intermittent communication disturbance disappears; wherein the duration of the safety redundancy cycle is longer than the instruction execution cycle of the edge execution unit in normal communication state.

8. The multi-cloud collaborative system based on edge computing according to any one of claims 1 to 7, characterized in that, The edge execution unit is an industrial robot controller, and the heterogeneous remote control node is a public cloud or private cloud server distributed in different geographical areas. The control topology entropy increment is used to characterize the ambiguity of control ownership caused by routing oscillations during cross-cloud switching.

Images