A cloud-edge collaboration-based printer anomaly judgment method and system

Abstract

The application relates to the fields of cloud-edge collaboration technology and network edge computing technology, in particular to a printer abnormality judgment method and system based on cloud-edge collaboration, which comprises the following steps: state sensing: acquiring network state signals in a connected state or an isolated state; a sensing unit acquires resource load levels in an idle state, a busy state or an exhausted state; data monitoring and regular reasoning: an edge state monitoring unit collects multi-dimensional data of a printer to form real-time state data streams; a regular abnormality reasoning unit receives the real-time state data streams, outputs known abnormality alarms or unknown data stream signals; autonomous learning: outputting an updated local cognitive model and task priority evaluation; dynamic scheduling: outputting resource regulation instructions to adjust data collection frequency and analysis depth; cognitive reporting: a cognitive data reporting unit reports the local cognitive model as high-value cognitive data packets to the cloud; the application realizes flexible management and anti-fragility under extreme constraints, and ensures the analysis capability of core threats.

Description

Technical Field

[0001] This invention relates to the fields of cloud-edge collaboration technology and network edge computing technology, specifically to a printer anomaly detection method and system based on cloud-edge collaboration. Background Technology

[0002] With the application of cloud-edge collaborative architecture in edge scenarios such as printer cluster monitoring, the system's reliance on network connectivity and cloud computing power is increasing.

[0003] In the current cloud master-edge-slave architecture, edge nodes typically play the role of passive executors, responsible for data collection and executing cloud commands, while complex cognitive and analytical functions rely on the cloud center. However, this rigid division of labor exposes serious flaws in extreme situations such as network isolation or edge resource depletion. Once disconnected from the cloud, edge nodes become cognitively frozen and unable to autonomously identify new anomalies. At the same time, when resources are depleted, edge nodes also lack elastic management capabilities, failing to effectively distinguish between benign data storms and malicious threats and allocate resources, resulting in poor system resilience. Therefore, how to enable edge nodes to still possess autonomous cognitive and elastic management capabilities under the dual constraints of disconnection and resource depletion is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] The purpose of this invention is to provide a printer anomaly detection method and system based on cloud-edge collaboration, to address the shortcomings of existing technologies where edge nodes rely on cloud-based cognition and lack autonomy, preventing them from becoming cognitively frozen during network isolation, and enabling them to maintain elastic management capabilities to distinguish between benign data storms and malicious threats even when resources are depleted. Specifically, the technical solution of this invention is as follows:

[0005] A method for printer anomaly detection based on cloud-edge collaboration includes the following steps:

[0006] Step 1, State Awareness: Obtain network state signals in connected or isolated states through the edge connection sensing unit; obtain the resource load level of idle, busy, or depleted resources through the local resource sensing unit;

[0007] Step 2, Data Monitoring and Conventional Reasoning: The edge state monitoring unit collects multi-dimensional data from the printer to form a real-time state data stream; the conventional anomaly reasoning unit receives the real-time state data stream, uses a local cognitive model containing feature vector centers and pattern signatures of known normal tasks to perform reasoning, and outputs known anomaly alarms or unknown data stream signals.

[0008] Step 3, Autonomous Learning: When the network state signal is in an isolated state, the autonomous learning and cognition unit is activated. The autonomous learning and cognition unit performs online learning on the unknown data stream signal, distinguishes between benign statistical anomalies and malignant novel anomalies, and outputs an updated local cognitive model and task priority evaluation.

[0009] Step 4, Dynamic Scheduling: When the resource load level is exhausted, the dynamic resource scheduling unit is activated; the dynamic resource scheduling unit executes a differentiated resource reallocation strategy based on the network status signal and the task priority assessment, and outputs resource regulation instructions to adjust the data acquisition frequency and analysis depth;

[0010] Step 5, Cognitive Reporting: When the network status signal returns to the connected state, the cognitive data reporting unit reports the local cognitive model as a high-value cognitive data packet to the cloud.

[0011] Furthermore, the local resource awareness unit collects the CPU utilization, memory usage, and cache queue length of the local node; the determination of the resource load level is obtained by comparing the CPU utilization, memory usage, and cache queue length with the resource security baseline; the resource security baseline is dynamically set through statistical analysis based on the hardware configuration and historical normal operation data of the edge node.

[0012] Furthermore, the conventional anomaly reasoning unit loads the local cognitive model; the unknown data stream signal refers to the real-time state data stream that cannot be matched or understood by the current local cognitive model;

[0013] In step 3, the autonomous learning and cognition unit only receives the data stream corresponding to the unknown data stream signal as input for online learning; the updated local cognitive model output by the autonomous learning and cognition unit is sent to the regular anomaly reasoning unit to update its reasoning rule base.

[0014] Furthermore, when the autonomous learning and cognitive unit performs the online learning, it distinguishes between benign statistical anomalies and malignant novel anomalies by comprehensively evaluating the burstiness, periodicity, and characteristic distribution of the unknown data stream;

[0015] The benign statistical anomalies were identified as having high suddenness and strong periodicity, and their characteristic distribution was consistent with that of known normal tasks.

[0016] The malignant novel anomaly is determined to be one whose characteristic distribution is inconsistent with or deviates from that of a known normal task.

[0017] Furthermore, when the network status signal is in an isolated state and the resource load level is exhausted, the dynamic resource scheduling unit executes the differentiated resource reallocation strategy, including:

[0018] Based on the task priority assessment, data streams assessed as low-priority - benign storms are proactively discarded or downgraded; and the released computing and caching resources are centrally preserved and allocated to data streams assessed as high-priority - suspected malicious.

[0019] Furthermore, when the network status signal is in a connected state and the resource load level is depleted, the dynamic resource scheduling unit executes the differentiated resource reallocation strategy, which is an emergency resource protection strategy, including:

[0020] Prioritize the basic operation of the edge connection sensing unit and the regular anomaly inference unit; and proactively limit the resources used to process the unknown data stream signals.

[0021] Furthermore, the resource regulation command output in step 4 is sent to the edge state monitoring unit and the conventional anomaly inference unit;

[0022] The edge status monitoring unit dynamically adjusts the collection frequency of specific data sources according to the resource regulation instructions;

[0023] The conventional anomaly reasoning unit dynamically adjusts its analysis depth based on the resource control instructions.

[0024] A printer anomaly detection system based on cloud-edge collaboration includes:

[0025] Edge connectivity sensing units are configured to detect connections to the cloud center and output network status signals in either a connected or isolated state.

[0026] The local resource sensing unit is configured to collect the load of local nodes and output the resource load level as idle, busy, or depleted.

[0027] An edge status monitoring unit is configured to collect printer data and form a real-time status data stream.

[0028] A conventional anomaly inference unit is configured to receive the real-time state data stream, perform inference using a local cognitive model that includes feature vector centers and pattern signatures of known normal tasks, and output known anomaly alarms or unknown data stream signals.

[0029] The autonomous learning and cognition unit is configured to receive the network status signal and the unknown data stream signal; when the network status signal is in an isolated state, the autonomous learning and cognition unit is activated, performs online learning on the unknown data stream signal, distinguishes between benign statistical anomalies and malignant novel anomalies, and outputs the updated local cognitive model to the regular anomaly inference unit, as well as outputs task priority evaluation.

[0030] A dynamic resource scheduling unit is configured to receive the network status signal, the resource load level, and the task priority assessment; when the resource load level is exhausted, the dynamic resource scheduling unit is activated, executes a differentiated resource reallocation strategy, and outputs resource regulation instructions to the edge status monitoring unit and the regular anomaly inference unit.

[0031] The cognitive data reporting unit is configured to be triggered when the network state signal recovers from the isolated state to the connected state, and to report the local cognitive model generated by the autonomous learning and cognitive unit as a high-value cognitive data package to the cloud.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. This invention solves the cognitive freeze problem during network isolation; by activating autonomous learning and cognitive units, when edge nodes are disconnected, it can perform online learning only for unknown data streams, autonomously distinguish between benign statistical anomalies and malignant novel anomalies, and update the learned new patterns to the local cognitive model, thus constructing a local discovery-learning-update cognitive closed loop, enabling edge nodes to have instant generalization recognition capabilities.

[0034] 2. This invention achieves elastic management and resilience under extreme constraints. Under the dual constraints of network isolation and resource depletion, the dynamic resource scheduling unit can execute differentiated resource reallocation strategies based on the task priority assessment output by the self-learning unit: actively discarding or downgrading benign data storms, and concentrating the released valuable resources for high-priority malicious anomaly analysis, thus ensuring the ability to analyze core threats.

[0035] 3. This invention provides multi-scenario, differentiated resource self-protection strategies; it not only solves the scenario of concurrent disconnection and resource depletion, but also provides an emergency resource protection strategy for the scenario of network connectivity but resource depletion; at this time, the system will prioritize the basic operation of edge connection perception and normal anomaly inference, while actively limiting the resource investment in unknown data streams that it cannot process, thus ensuring the survivability and basic protection capabilities of the system under different failure modes.

[0036] 4. This invention improves the overall efficiency of cloud-edge collaboration. When the network connection is restored, the edge node only reports the local cognitive model, which is generated by autonomous learning and contains new abnormal patterns, as a high-value cognitive data packet to the cloud, instead of uploading the massive amount of raw logs accumulated during the isolation period. This greatly reduces the network transmission burden and provides high-value input for the iteration of the global model in the cloud, thereby improving the iteration efficiency of the global model. Attached Figure Description

[0037] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0038] Figure 1 This is a flowchart of the method of the present invention.

[0039] Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0040] 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.

[0041] Example 1:

[0042] Please see Figure 1 A method for printer anomaly detection based on cloud-edge collaboration includes the following steps:

[0043] Step 1, State Awareness: Obtain network state signals in connected or isolated states through the edge connection sensing unit; obtain the resource load level of idle, busy or depleted resources through the local resource sensing unit;

[0044] Step 2, Data Monitoring and Conventional Reasoning: The edge state monitoring unit collects multi-dimensional data from the printer to form a real-time state data stream; the conventional anomaly reasoning unit receives the real-time state data stream, uses a local cognitive model containing feature vector centers and pattern signatures of known normal tasks to perform reasoning, and outputs known anomaly alarms or unknown data stream signals.

[0045] Step 3, Autonomous Learning: When the network state signal is in an isolated state, the autonomous learning and cognition unit is activated. The autonomous learning and cognition unit performs online learning on the unknown data stream signal, distinguishes between benign statistical anomalies and malignant novel anomalies, and outputs an updated local cognitive model and task priority evaluation.

[0046] Step 4, Dynamic Scheduling: When the resource load level is exhausted, the dynamic resource scheduling unit is activated; the dynamic resource scheduling unit executes a differentiated resource reallocation strategy based on the network status signal and the task priority assessment, and outputs resource regulation instructions to adjust the data acquisition frequency and analysis depth;

[0047] Step 5, Cognitive Reporting: When the network status signal returns to the connected state, the cognitive data reporting unit reports the local cognitive model as a high-value cognitive data packet to the cloud.

[0048] This embodiment describes a complete method and process deployed on the edge node of an enterprise office network to monitor and determine the abnormal status of a printer cluster.

[0049] The implementation of the method described in this invention begins with the dual-state perception in step 1; the edge connection perception unit is configured to periodically send heartbeat probe packets to the cloud center and detect responses. When responses are lost continuously, it outputs an isolated network status signal; at the same time, the local resource perception unit collects local CPU and memory usage and cache queue length in real time. When the load continues to be higher than a dynamically set resource safety baseline, it outputs a depleted resource load level.

[0050] After obtaining the baseline of the environment and its own state, the method proceeds to the data monitoring and routine inference stage in step 2; the edge state monitoring unit collects sensor readings and task logs from the printer cluster to form a real-time state data stream; this data stream is sent to the routine anomaly inference unit; this unit uses a local cognitive model to match the data stream; if the match is successful, a known anomaly alarm is output; if the data stream cannot be understood or matched by the current model, it is marked and output as an unknown data stream signal;

[0051] The core design concept of this invention lies in addressing the dual extremes of network isolation and resource exhaustion; this dual constraint is a fatal weakness of existing cloud master-edge-slave architectures, and this invention precisely activates its autonomous capabilities in this scenario, specifically manifested in the collaboration of steps 3 and 4:

[0052] Suppose that at the end of a certain month, the network fails and employees rush to print reports, causing a data storm.

[0053] To address this extreme scenario, in the autonomous learning phase of step 3, the autonomous learning and cognitive unit activated by the isolated state signal begins to assume the cognitive functions that should traditionally be handled by the cloud. The key responsibility of this unit is to autonomously understand new situations occurring locally without cloud guidance. It receives the unknown data stream signal output from step 2 and performs online learning on it. Through analysis, it distinguishes data storms into benign statistical anomalies and distinguishes the mixed threats into malignant new anomalies. Based on this distinction, the unit outputs two things: an updated local cognitive model containing the newly identified benign and malignant pattern signatures, and a task priority evaluation.

[0054] In parallel with autonomous learning, during the dynamic scheduling phase in step 4, the dynamic resource scheduling unit activated by the exhaustion signal begins to perform computing power self-preservation. The key responsibility of this unit is to ensure the continuation of core functions when resources are exhausted. Since it is currently in an isolated state, the decision-making logic of this unit relies entirely on the task priority assessment generated in step 3 as the basis for decision-making. It executes a differentiated discard-preservation strategy: for data streams assessed as low-priority - benign storms, it actively discards them; at the same time, it centrally preserves and allocates the valuable computing and cache resources released after discarding them to data streams assessed as high-priority - suspected malignant data streams. This decision is encapsulated as a resource regulation instruction and output.

[0055] When the external constraints are removed, the method enters the cognitive reporting stage in step 5; the network state signal is restored to the connected state, and the cognitive data reporting unit is triggered accordingly; at this time, it does not upload the massive amount of raw logs accumulated during the isolation period, but only packages and reports the local cognitive model generated by the autonomous learning and cognitive unit in step 3 and verified locally as high-value cognition to the cloud.

[0056] Through the above steps, this method constructs a transformation from passive execution to autonomous intelligence. It solves the defects of rigid division of labor between cloud master and edge slave in the existing technology, enabling edge nodes to have real-time generalized recognition and elastic management capabilities under the dual constraints of disconnection and resource depletion. This adaptive closed-loop information flow, which uses cognition to guide regulation and uses regulation to protect cognition, is the core innovation of this invention. It ensures the system's resilience in extreme situations: it can not only distinguish between benign storms and malign anomalies, avoiding false alarms and missed alarms, but also provide high-value cognition to the cloud after network recovery, greatly improving the iteration efficiency of the global model.

[0057] Example 2:

[0058] The local resource awareness unit collects the CPU utilization, memory usage, and cache queue length of the local node; the determination of the resource load level is obtained by comparing the CPU utilization, memory usage, and cache queue length with the resource security baseline; the resource security baseline is dynamically set through statistical analysis based on the hardware configuration and historical normal operation data of the edge node.

[0059] In this particular configuration of step 1, the method by which the local resource awareness unit determines the resource load level is specifically explained; the responsibility of this unit is to provide an accurate and reliable signal about whether the edge node itself is overloaded.

[0060] In this embodiment, the unit continuously collects three key indicators: CPU utilization, memory usage, and cache queue length. The core of its judgment lies in the introduction of a resource security baseline. In this invention, the technical meaning of this resource security baseline is a dynamic threshold used to ensure that the core monitoring functions of the edge node can operate to a minimum. Its key role is to serve as a basis for judging the exhaustion state, but it avoids using a fixed, one-size-fits-all threshold such as 90% CPU, because a fixed threshold cannot adapt to the normal fluctuations of different hardware configurations and services.

[0061] The baseline is determined by dynamically setting and adjusting it based on statistical analysis, taking into account common knowledge in the field and the current hardware configuration and historical normal operation data of the edge nodes.

[0062] When any two of the three metrics collected by the local resource awareness unit—CPU utilization, memory usage, and cache queue length—continuously exceed the dynamically set resource safety baseline, or when the cache queue length alone reaches 95% of its maximum capacity, the unit determines that the system can no longer handle more new tasks and then switches the resource load level output signal to exhaustion.

[0063] By using a resource security baseline dynamically set based on statistical analysis and hardware configuration, and supplemented by a clear, multi-indicator combination judgment logic, this specific configuration of the present invention greatly improves the accuracy of resource load level judgment. This avoids the erroneous triggering of exhaustion state due to short-term, normal business bursts, and also ensures that when the system is truly facing the risk of resource exhaustion, the dynamic resource scheduling unit can be activated in a timely and accurate manner, so that subsequent elastic management strategies can be launched at the most appropriate time.

[0064] Example 3:

[0065] The conventional anomaly reasoning unit loads the local cognitive model; the unknown data stream signal refers to the real-time state data stream that cannot be matched or understood by the current local cognitive model.

[0066] In step 3, the autonomous learning and cognition unit only receives the data stream corresponding to the unknown data stream signal as input for online learning; the updated local cognitive model output by the autonomous learning and cognition unit is sent to the regular anomaly reasoning unit to update its reasoning rule base.

[0067] When the autonomous learning and cognitive unit performs the online learning, it distinguishes between benign statistical anomalies and malignant novel anomalies by comprehensively evaluating the suddenness, periodicity, feature distribution, and correlation anomalies of the unknown data stream's internal features.

[0068] The benign statistical anomalies were identified as having high suddenness and strong periodicity, and their characteristic distribution was consistent with that of known normal tasks.

[0069] The malignant novel anomaly is determined to be that the feature distribution is inconsistent with or deviates from the known normal task; or, although its overall feature distribution is similar to the normal task, the combination of multiple features or temporal relationships within its data show an abnormal association that does not conform to the known normal pattern. For example, a task that is claimed to be a normal report contains an abnormal instruction sequence.

[0070] This embodiment describes in detail how the conventional anomaly reasoning unit and the autonomous learning and cognitive unit work together to construct a local, autonomous cognitive-evolutionary closed loop;

[0071] In this closed loop, the regular anomaly inference unit plays the role of a regular discriminator of the data stream; it loads a local cognitive model; its responsibility is to process all real-time state data streams input from the edge state monitoring unit; when a data stream cannot be matched or understood by any known pattern in its model library, the data stream is defined as an unknown data stream signal.

[0072] A key information flow design of this invention is that the conventional anomaly reasoning unit does not attempt to process or discard this unknown data stream signal on its own; instead, it sends this unknown data stream as a key input only to the autonomous learning and cognition unit.

[0073] In contrast, the autonomous learning and cognitive unit plays the role of a local cognizer; it is activated only in the isolated state, and its input comes only from the unknown data stream sent by the regular abnormal reasoning unit; this on-demand learning design, that is, only processing unknown data, greatly saves the already valuable computing resources of the edge nodes.

[0074] When performing online learning, such as through lightweight streaming clustering or density estimation, the purpose of autonomous learning and cognitive units is to qualitatively characterize the unknown; it diagnoses the nature of the unknown data stream by comprehensively evaluating three dimensions:

[0075] Burstness: This dimension is assessed by calculating the instantaneous increase in the data stream per unit of time;

[0076] Periodicity: This dimension is evaluated by comparing the timestamp of the current data stream with the historical business calendar stored at the edge nodes;

[0077] Feature distribution: This dimension is the decisive basis for distinguishing between good and bad; it is implemented by vectorizing the unknown data stream and comparing its cosine similarity with the feature vector centers of known normal tasks stored in the model.

[0078] Based on this assessment, the units are differentiated:

[0079] Determined as a benign statistical anomaly: If the data stream exhibits high burstiness and strong periodicity, but its feature distribution vector has a cosine similarity to that of a known normal task that is higher than a preset threshold, it is determined to be feature consistent.

[0080] Determined as a malignant novel anomaly: If the cosine similarity of the feature distribution vector of a data stream is lower than the threshold, it is determined to be a feature deviation; When a data stream is determined to be a malignant novel anomaly, the autonomous learning and cognition unit will trigger a signature extraction sub-process: For example, the unit performs online clustering on the data stream determined to be malignant and extracts its new feature vector center or key features as the pattern signature of the malignant novel anomaly.

[0081] After completing the above differentiation and signature extraction, the autonomous learning and cognition unit integrates the signatures of the newly identified benign patterns and the newly learned malignant patterns into an updated local cognitive model and sends it back to the regular anomaly reasoning unit. Upon receiving this updated model, the regular anomaly reasoning unit immediately loads and updates its reasoning rule base.

[0082] At the same time, the autonomous learning and cognition unit maps the qualitative results of the data flow mentioned above, namely benign statistical anomalies or malignant new anomalies, to specific scheduling priorities, for example: benign = low, malignant = high, and outputs them as task priority evaluation to the dynamic resource scheduling unit.

[0083] This embodiment constructs an efficient local learning loop through mechanisms such as learning from unknown data, multi-dimensional qualitative analysis, and real-time model feedback. The creativity of this information flow lies in the fact that it constructs a complete autonomous cognitive loop at the edge, which involves discovering the unknown, learning the unknown, and updating the known. This is fundamentally different from the passive mode of existing technologies, which involves discovering the unknown, reporting the unknown, and waiting for the cloud. This enables edge nodes to digest data storms when they are disconnected and learn to distinguish between benign storms and malignant anomalies, thus perfectly solving the defect of cognitive freeze.

[0084] Example 4:

[0085] When the network status signal is in an isolated state and the resource load level is exhausted, the dynamic resource scheduling unit executes the differentiated resource reallocation strategy, including:

[0086] Based on the task priority assessment, data streams assessed as low-priority - benign storms are proactively discarded or downgraded; and the released computing and caching resources are centrally preserved and allocated to data streams assessed as high-priority - suspected malicious.

[0087] This embodiment describes in detail the core survival strategy of the present invention in the most severe failure scenario, namely, when isolation and exhaustion occur simultaneously.

[0088] In this scenario, the dynamic resource scheduling unit is activated; at the same time, due to its isolated state, the autonomous learning and cognitive unit is also activated and working; at this moment, the decision-making logic of the dynamic resource scheduling unit depends entirely on the cognitive output of the autonomous learning and cognitive unit.

[0089] The dynamic resource scheduling unit receives a task priority assessment from the autonomous learning and cognitive unit; this assessment has categorized the data stream into low-priority - benign storm and high-priority - suspected malignant.

[0090] The differentiated resource reallocation strategy executed by the dynamic resource scheduling unit is a proactive and asymmetric trade-off:

[0091] Proactive discarding: For all data streams assessed as low-priority - benign storms, the dynamic resource scheduling unit proactively performs degradation processing; it outputs resource adjustment instructions, forcing the edge state monitoring unit to significantly reduce the sampling rate of these data streams, and forcing the regular anomaly inference unit to suspend in-depth analysis of these streams;

[0092] Centralized protection: By discarding the valuable CPU cycles and cache queue space freed up by low-priority tasks, the dynamic resource scheduling unit immediately and centrally allocates them to those data streams that are assessed as high-priority and suspected of being malicious; this ensures that the analysis of real threats has a minimum computing power guarantee, so that the analysis process will not be interrupted or data lost due to resource constraints.

[0093] This cognition-based regulation is one of the core innovations of this invention; it achieves resilient survivability under extreme resource constraints; by using cognition to guide regulation to make sacrifices, the system avoids wasting valuable computing power on benign storms; more importantly, the resources released by the sacrifices made by regulation, in turn, provide computing power to ensure the survival of cognition, enabling it to continuously learn online about subsequent high-priority threats.

[0094] Example 5:

[0095] When the network status signal is in a connected state and the resource load level is depleted, the dynamic resource scheduling unit executes the differentiated resource reallocation strategy, which is an emergency resource protection strategy, including:

[0096] Prioritize the basic operation of the edge connection sensing unit and the regular anomaly inference unit; and proactively limit the resources used to process the unknown data stream signals.

[0097] This embodiment describes a strategy for another failure scenario; in this scenario, the edge nodes are not disconnected from the cloud, but are also exhausted due to a data storm.

[0098] Under this specific configuration, since the network state signal is not in an isolated state, the autonomous learning and cognitive units will not be activated; this means that the edge nodes do not have the cognitive ability to autonomously distinguish between benign storms and malign anomalies at this time.

[0099] Given this situation, the goal of the dynamic resource scheduling unit is very clear: not autonomous cognition, but to maintain survival and ensure basic functions; it instead implements an emergency resource protection strategy:

[0100] Ensuring Core Functionality: This strategy allocates resources to prioritize the basic operation of the two most critical units:

[0101] Edge connectivity sensing unit: It must be kept running to maintain heartbeat and connectivity awareness with the cloud;

[0102] Conventional anomaly reasoning unit: It must be guaranteed to operate in order to at least be able to handle and identify known anomaly patterns;

[0103] Limiting unknown resource consumption: Since nodes are exhausted and unable to learn locally, the massive amount of unknown data stream signals becomes the biggest burden. Accordingly, this strategy actively limits the resources used to process these unknown data stream signals, such as limiting the bandwidth for reporting to the cloud or limiting the size of the local cache queue, to prevent the massive amount of unknown data from overwhelming the local cache or network.

[0104] This emergency resource protection strategy is a safety valve to ensure that the system does not collapse when connectivity is exhausted. It prioritizes the absolute stability of the two core functions of connectivity awareness and known anomaly handling by limiting resource allocation to unknown data that cannot be processed. This ensures that edge nodes can survive at least during resource storms and maintain their most basic protection capabilities.

[0105] Example 6:

[0106] The resource regulation command output in step 4 is sent to the edge state monitoring unit and the regular anomaly inference unit;

[0107] The edge status monitoring unit dynamically adjusts the collection frequency of specific data sources according to the resource regulation instructions;

[0108] The conventional anomaly reasoning unit dynamically adjusts its analysis depth based on the resource control instructions.

[0109] This embodiment further clarifies how the resource control instructions output by the dynamic resource scheduling unit are executed; the purpose of these instructions is to transform the scheduling decisions of the dynamic resource scheduling unit into specific actions of downstream units.

[0110] In this configuration, resource regulation commands are sent simultaneously to two units: the edge state monitoring unit and the regular anomaly inference unit;

[0111] Control of the edge status monitoring unit: When the edge status monitoring unit receives a degradation instruction for a specific data stream, it will dynamically adjust the acquisition frequency of that specific data source; for example, it will reduce its data acquisition frequency from real time to low frequency, thereby reducing the input load of the system from the data source.

[0112] Control of the regular anomaly inference unit: When the regular anomaly inference unit receives the same demotion instruction, it will dynamically adjust its analysis depth; for example, for data streams marked as low priority, the regular anomaly inference unit may be instructed to skip deep packet inspection or complex multidimensional pattern matching, and instead perform only the most basic log format check, thereby significantly reducing the CPU consumption of the data processing stage.

[0113] By using this resource control command to dynamically regulate both the acquisition frequency and analysis depth, this invention achieves a refined resource management system. It does not simply turn a function on or off, but rather reduces the frequency and depth of data stream processing based on the results of the analysis. This allows the discard and preservation strategies to be executed specifically and effectively, thus achieving true flexible management.

[0114] Example 7:

[0115] Please see Figure 2 A printer anomaly detection system based on cloud-edge collaboration includes:

[0116] Edge connectivity sensing units are configured to detect connections to the cloud center and output network status signals in either a connected or isolated state.

[0117] The local resource sensing unit is configured to collect the load of local nodes and output the resource load level as idle, busy, or depleted.

[0118] An edge status monitoring unit is configured to collect printer data and form a real-time status data stream.

[0119] A conventional anomaly inference unit is configured to receive the real-time state data stream, perform inference using a local cognitive model that includes feature vector centers and pattern signatures of known normal tasks, and output known anomaly alarms or unknown data stream signals.

[0120] The autonomous learning and cognition unit is configured to receive the network status signal and the unknown data stream signal; when the network status signal is in an isolated state, the autonomous learning and cognition unit is activated, performs online learning on the unknown data stream signal, distinguishes between benign statistical anomalies and malignant novel anomalies, and outputs the updated local cognitive model to the regular anomaly inference unit, as well as outputs task priority evaluation.

[0121] A dynamic resource scheduling unit is configured to receive the network status signal, the resource load level, and the task priority assessment; when the resource load level is exhausted, the dynamic resource scheduling unit is activated, executes a differentiated resource reallocation strategy, and outputs resource regulation instructions to the edge status monitoring unit and the regular anomaly inference unit.

[0122] The cognitive data reporting unit is configured to be triggered when the network state signal recovers from the isolated state to the connected state, and to report the local cognitive model generated by the autonomous learning and cognitive unit as a high-value cognitive data package to the cloud.

[0123] This embodiment provides a system deployed on an edge node that implements the above-described method; the system includes multiple unit modules configured to perform specific functions.

[0124] The edge connection sensing unit and the local resource sensing unit together constitute the system's state sensing layer, which respectively provide two key triggering bases: network state signals and resource load levels.

[0125] The edge state monitoring unit is responsible for the input of data, and its output real-time state data stream is sent to the regular anomaly inference unit; the inference unit is regular and is configured to process the data using a local cognitive model and sort out unknown data stream signals that it cannot understand.

[0126] The core of this system's architecture lies in the configuration and coupling of the autonomous learning and cognitive unit and the dynamic resource scheduling unit;

[0127] The autonomous learning and cognition unit is configured to activate only in the isolated state and to process only unknown data stream signals; it is configured to execute online learning algorithms to distinguish between benign and malign anomalies; its configured output has two directions: feeding back the updated local cognitive model to the regular anomaly reasoning unit and feeding forward the task priority evaluation to the dynamic resource scheduling unit.

[0128] The dynamic resource scheduling unit is configured to activate only when exhausted; its key configuration is that it can simultaneously receive network status signals, resource load levels, and task priority assessments, enabling it to execute the differentiated resource reallocation strategy as described in the previous embodiments based on these three inputs; its output resource regulation instructions are configured to be sent to the edge status monitoring unit and the regular anomaly inference unit simultaneously to achieve closed-loop control of the data link and the processing link.

[0129] The cognitive data reporting unit is configured to be triggered when the network recovers and is configured to report only the local cognitive model, rather than massive amounts of raw data, in order to achieve efficient cloud iteration;

[0130] This system constructs a cognitive-regulatory bidirectional coupling adaptive mechanism on edge nodes through its specific unit configuration and interconnection relationships. This architecture transforms the edge node from a passive executor into an autonomous edge. The non-obviousness of this architecture lies in the fact that it changes the behavior of the edge node under extreme constraints by reconstructing the information flow, rather than relying solely on hardware stacking or model optimization, thus transforming it from a rigid executor into a resilient adaptive system.

[0131] 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 cloud-edge collaboration-based printer anomaly determination method, characterized by, The method is deployed at an edge node, and the method comprises the following steps: Step 1, state awareness: acquiring a network state signal in a connected state or an isolated state through an edge connection awareness unit; and acquiring a resource load level in an idle state, a busy state or an exhausted state through a local resource awareness unit; Step 2, data monitoring and regular inference: collecting multi-dimensional data of a printer by an edge state monitoring unit to form a real-time state data stream; a regular abnormality inference unit receives the real-time state data stream, performs inference by using a local cognitive model containing a feature vector center and a mode signature of a known normal task, and outputs a known abnormality alarm or an unknown data stream signal; Step 3, autonomous learning: when the network state signal is in the isolated state, an autonomous learning and cognition unit is activated, the autonomous learning and cognition unit performs online learning on the unknown data stream signal, distinguishes between benign statistical abnormalities and malignant new abnormalities, and outputs an updated local cognitive model and a task priority evaluation; Step 4, dynamic scheduling: when the resource load level is in the exhausted state, a dynamic resource scheduling unit is activated; the dynamic resource scheduling unit performs a differentiated resource reallocation strategy according to the network state signal and the task priority evaluation, and outputs a resource regulation instruction to adjust the data collection frequency and the analysis depth; Step 5, cognitive reporting: when the network state signal returns to the connected state, a cognitive data reporting unit reports the local cognitive model as a high-value cognitive data packet to the cloud.

2. The cloud edge collaboration-based printer abnormality determination method according to claim 1, characterized by, In the step 1, the local resource awareness unit collects the CPU usage, the memory occupancy and the cache queue length of the local node; the determination of the resource load level is obtained by comparing the CPU usage, the memory occupancy and the cache queue length with a resource safety baseline; the resource safety baseline is dynamically set according to the hardware configuration and the historical normal operation data of the edge node through statistical analysis.

3. The cloud edge collaboration-based printer abnormality determination method according to claim 1, characterized by, In the step 2, the regular abnormality inference unit loads the local cognitive model; the unknown data stream signal refers to the real-time state data stream that cannot be matched or understood by the current local cognitive model; In the step 3, the autonomous learning and cognition unit only receives the data stream corresponding to the unknown data stream signal as the input of online learning; the updated local cognitive model output by the autonomous learning and cognition unit is sent to the regular abnormality inference unit to update the inference rule base thereof.

4. The cloud edge collaboration-based printer abnormality determination method according to claim 3, characterized by, When the autonomous learning and cognition unit performs the online learning, the benign statistical abnormalities and the malignant new abnormalities are distinguished by comprehensively evaluating the burstiness, the periodicity and the feature distribution of the unknown data stream; The benign statistical abnormalities are determined to have high burstiness, strong periodicity, and consistent feature distribution with the known normal task; The malignant new abnormalities are determined to have inconsistent or deviated feature distribution from the known normal task.

5. The cloud edge collaboration-based printer abnormality determination method according to claim 1, characterized by, In the step 4, when the network state signal is in the isolated state and the resource load level is in the exhausted state, the dynamic resource scheduling unit performs the differentiated resource reallocation strategy, which comprises: According to the task priority evaluation, the data stream evaluated as a low-priority benign storm is actively subjected to discard or downgrade processing; and The released computing and cache resources are centrally secured and allocated to the data stream evaluated as a high-priority suspected malicious storm.

6. The cloud edge collaboration-based printer abnormality determination method according to claim 1, characterized by, In step 4, when the network state signal is in the connected state and the resource load level is exhausted, the dynamic resource scheduling unit executes the differentiated resource reallocation strategy, which is an emergency resource protection strategy, including: Prioritize the basic operation of the edge connection awareness unit and the regular anomaly reasoning unit; and Active restriction of resources for processing the unknown data stream signal.

7. The cloud edge collaboration-based printer abnormality determination method according to claim 1, characterized by, The resource control instruction output in step 4 is sent to the edge state monitoring unit and the regular anomaly reasoning unit; The edge state monitoring unit dynamically adjusts the collection frequency of specific data sources according to the resource control instruction; The regular anomaly reasoning unit dynamically adjusts its analysis depth according to the resource control instruction.

8. A cloud-edge collaboration based printer abnormality judging system configured to execute the cloud-edge collaboration based printer abnormality judging method according to any one of claims 1 to 7. The system is deployed on an edge node, and the system includes: An edge connection awareness unit configured to detect a connection with a cloud center and output a network state signal in a connected state or an isolated state; A local resource awareness unit configured to collect local node load and output a resource load level in an idle, busy, or exhausted state; An edge state monitoring unit configured to collect printer data and form a real-time state data stream; A regular anomaly reasoning unit configured to receive the real-time state data stream, use a local cognitive model containing a feature vector center and a pattern signature of a known normal task for reasoning, and output a known anomaly alarm or an unknown data stream signal; An autonomous learning and cognition unit configured to receive the network state signal and the unknown data stream signal; when the network state signal is in the isolated state, the autonomous learning and cognition unit is activated, performs online learning on the unknown data stream signal, distinguishes between benign statistical anomalies and malignant new anomalies, and outputs an updated local cognitive model to the regular anomaly reasoning unit and a task priority evaluation; A dynamic resource scheduling unit configured to receive the network state signal, the resource load level, and the task priority evaluation; when the resource load level is exhausted, the dynamic resource scheduling unit is activated to execute a differentiated resource reallocation strategy and output a resource control instruction to the edge state monitoring unit and the regular anomaly reasoning unit; A cognitive data reporting unit configured to be triggered when the network state signal is restored from the isolated state to the connected state, and report the local cognitive model generated by the autonomous learning and cognition unit as high-value cognitive data packets to the cloud.

Images