Edge computing and decision-based incident response method, system, device, and medium

Abstract

The application provides an event response method, system, device and medium based on edge computing and decision, the method comprising: obtaining running data and environment data of a target system; performing double-channel feature fusion on the running data and the environment data, filtering out background interference through state space segmentation to separate out an event abnormal high-risk area; inputting a feature matrix corresponding to the event abnormal high-risk area into a dynamic decision network to generate a dynamic constraint boundary of a response candidate strategy; calculating a multi-dimensional trend difference of the running data using a time series difference operator, filtering the multi-dimensional trend difference using a double threshold to separate out strong correlation points and weak correlation points, and identifying linear topological structures of the strong correlation points and the weak correlation points using a sequence topological analysis algorithm to extract time series direction features; drawing a decision evolution reference line in a state space according to the time series direction features; and solving two space-time intersection points of the decision evolution reference line and the dynamic constraint boundary as a starting trigger point and a termination cutoff point of event response execution, respectively.

Description

Technical Field

[0001] This application relates to the field of event response technology, and in particular to an event response method, system, device and medium based on edge computing and decision-making. Background Technology

[0002] With the widespread application of edge computing and distributed intelligent decision-making technologies, an increasing number of IoT terminals, industrial control systems, and intelligent network devices require localized real-time event response at edge nodes. By directly collecting and processing real-time operational data and multi-dimensional environmental data at the edge, close to the data source, the transmission latency caused by data uploading to the cloud can be significantly reduced, and the system's response time to sudden abnormal events can be significantly improved. To achieve highly reliable closed-loop control at the edge with limited computing resources, how to quickly fuse multi-dimensional heterogeneous data using lightweight algorithms, accurately extract core abnormal regions, and delineate adaptive decision-control boundaries has become a research hotspot and core evolution direction in the field of edge computing and event response.

[0003] In related technologies, existing edge event response methods mostly rely on fixed single-index threshold triggering or static hard-coded rule matching. When faced with complex, variable, and noisy multidimensional time-series data streams, these methods are prone to false alarms or missed alarms due to their inability to effectively filter out environmental interference, making them ill-suited for handling complex and concealed sudden events. Furthermore, existing solutions typically lack topological analysis of the internal correlations of multidimensional data and the dynamic evolution trends of event timelines, failing to accurately define the effective rule boundaries of response strategies in highly dynamic operating conditions. This directly leads to edge decision engines being unable to accurately determine the optimal execution timing and safe exit mechanism for event responses, frequently resulting in delayed response actions, premature termination, or over-response leading to resource waste, failing to meet the requirements for high-precision, deterministic, and rapid closed-loop control. Summary of the Invention

[0004] In view of this, it is necessary to provide an event response method, system, device and medium based on edge computing and decision-making that can at least overcome one of the above-mentioned defects.

[0005] In a first aspect, embodiments of this application provide an event response method based on edge computing and decision-making, applied to edge computing nodes, the method comprising:

[0006] Acquire the target system's operational and environmental data;

[0007] The operational data and the environmental data are fused using dual-channel features, and background interference is filtered out by state space segmentation to separate high-risk areas of event anomalies.

[0008] The feature matrix corresponding to the high-risk area of ​​the event is input into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy.

[0009] The multidimensional trend difference of the running data is calculated using a time-series difference operator. The multidimensional trend difference is filtered using a dual threshold to separate strong and weak correlation points. The linear topological structure of the strong and weak correlation points is identified by a sequence topology analysis algorithm to extract time-series directional features.

[0010] Draw a decision evolution reference line in the state space based on the temporal direction characteristics;

[0011] The two spatiotemporal intersections of the decision evolution reference line and the dynamic constraint boundary are calculated and used as the starting trigger point and the ending cutoff point for the event response execution, respectively.

[0012] In one embodiment, the step of inputting the feature matrix corresponding to the high-risk area of ​​the event abnormality into a dynamic decision network to generate a dynamic constraint boundary for the response candidate strategy includes:

[0013] The feature matrix corresponding to the high-risk area of ​​the event is input into a dynamic decision network based on a temporal deep learning network;

[0014] Construct a multi-scale policy response map using a multi-scale temporal domain feature network;

[0015] Key state features are selected by applying multi-dimensional state alignment technology combined with a preset strategy intersection-union ratio threshold.

[0016] The dynamic constraint boundary is output through high-dimensional deconvolution mapping and decoding.

[0017] In one embodiment, after extracting the start trigger point and the end cutoff point as the start trigger point and end cutoff point for event response execution, the method further includes:

[0018] The target state features of the start trigger point and the end cutoff point within the initial time domain window are obtained by applying a time domain filtering algorithm.

[0019] In the subsequent time series of the data stream, frequency domain correlation filtering is used for real-time retrieval and relocation of the target state features;

[0020] By combining the Kalman filter algorithm, the trajectory of the evolution trend of the sudden event under dynamic noise interference is predicted, and the spatiotemporal offset between the starting trigger point and the ending cutoff point is calculated.

[0021] In one embodiment, the method further includes:

[0022] The system load change and communication network jitter amplitude of the edge computing node are obtained respectively;

[0023] The system load change, the communication network jitter amplitude and the spatiotemporal offset are fused in multiple dimensions to calculate the timing decision deviation between the starting trigger point or the termination cutoff point and the actual closed-loop operation point.

[0024] Based on the instantaneous timing decision deviation, an adaptive correction command is output to drive the parameter adjustment engine to make dynamic adjustments, so that the event decision engine is continuously aligned with the starting trigger point or the ending cutoff point.

[0025] In one embodiment, the method further includes:

[0026] The event decision engine's feedback execution delay and dynamic tracking error during the emergency response process are obtained.

[0027] When the feedback execution delay or the dynamic tracking error exceeds a preset safety threshold, a degradation protection command is output;

[0028] The degradation protection command controls the event decision engine to stop the current response job and reset the running status of the edge computing node.

[0029] In one embodiment, the step of inputting the feature matrix corresponding to the high-risk area of ​​the event abnormality into a dynamic decision network to generate a dynamic constraint boundary for the response candidate strategy includes:

[0030] The edge operating parameters during feature acquisition are obtained, including data noise intensity, system throughput, and network packet loss rate.

[0031] The environmental correction coefficient is determined based on the edge condition parameters, and the environmental correction coefficient is applied to dynamically correct the decision confidence of the dynamic decision network output.

[0032] In one embodiment, the step of dynamically correcting the decision confidence level output by the dynamic decision network using the environmental correction coefficient includes:

[0033] Multiply the decision confidence level by the environmental correction coefficient to obtain the corrected event decision score;

[0034] The response candidate strategies are re-evaluated and filtered based on the corrected event decision score.

[0035] Secondly, embodiments of this application provide an event response system based on edge computing and decision-making, applied to implement the method described in the first aspect, the system comprising:

[0036] The data acquisition module acquires the target system's operational and environmental data.

[0037] The feature fusion module performs dual-channel feature fusion of the operational data and the environmental data, and filters out background interference through state space segmentation to separate high-risk areas of event anomalies.

[0038] The decision boundary generation module inputs the feature matrix corresponding to the high-risk area of ​​the event into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy.

[0039] The time-series trend analysis module uses a time-series difference operator to calculate the multidimensional trend difference of the running data, uses a dual threshold to filter the multidimensional trend difference to separate strong correlation points and weak correlation points, and uses a sequence topology analysis algorithm to identify the linear topological structure of the strong correlation points and the weak correlation points to extract time-series direction features.

[0040] The decision reference line drawing module draws decision evolution reference lines in the state space according to the temporal direction characteristics.

[0041] The spatiotemporal intersection calculation module calculates the two spatiotemporal intersections of the decision evolution reference line and the dynamic constraint boundary, which serve as the starting trigger point and the ending cutoff point for the event response execution, respectively.

[0042] Thirdly, embodiments of this application provide an electronic device, including:

[0043] Processor; and

[0044] A memory having computer-readable instructions stored thereon for controlling the processor to perform the method as described in the first aspect.

[0045] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the method described in the first aspect at runtime.

[0046] This application provides an event response method, system, device, and medium based on edge computing and decision-making. By fusing dual-channel features of operational data and environmental data and combining them with state space segmentation, it can effectively filter out background noise interference in the high-dimensional and complex data streams of ports and ocean shipping, accurately separate high-risk areas of event anomalies, and fundamentally solve the problems of false alarms and emergency delays caused by data lag and information silos in traditional shipping management systems. At the same time, by utilizing dynamic decision networks, event priority management, and knowledge graph linkage, it breaks through the limitations of traditional static hard-coded rule matching, and realizes flexible adaptive adjustment of control boundaries and support for complex semantic reasoning in the resource environment of edge computing. Attached Figure Description

[0047] Figure 1This is a flowchart illustrating an event response method based on edge computing and decision-making provided in an embodiment of this application.

[0048] Figure 2 This is a schematic diagram of the modules of an event response system provided in an embodiment of this application.

[0049] Figure 3 This is a schematic diagram of the modules of an electronic device provided in an embodiment of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0051] It should be noted that, in the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.

[0052] It should be noted that in the embodiments of this application, the terms "first," "second," etc., are used only for descriptive purposes and should not be construed as indicating or implying relative importance, nor as indicating or implying order. Features specified as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0053] Based on the embodiments described in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] With the widespread application of edge computing and distributed intelligent decision-making technologies, an increasing number of IoT terminals, industrial control systems, and intelligent network devices require localized real-time event response at edge nodes. By directly collecting and processing real-time operational data and multi-dimensional environmental data at the edge, close to the data source, the transmission latency caused by data uploading to the cloud can be significantly reduced, and the system's response time to sudden abnormal events can be significantly improved. To achieve highly reliable closed-loop control at the edge with limited computing resources, how to quickly fuse multi-dimensional heterogeneous data using lightweight algorithms, accurately extract core abnormal regions, and delineate adaptive decision-control boundaries has become a research hotspot and core evolution direction in the field of edge computing and event response.

[0055] In related technologies, existing edge event response methods mostly rely on fixed single-index threshold triggering or static hard-coded rule matching. When faced with complex, variable, and noisy multidimensional time-series data streams, these methods are prone to false alarms or missed alarms due to their inability to effectively filter out environmental interference, making them ill-suited for handling complex and concealed sudden events. Furthermore, existing solutions typically lack topological analysis of the internal correlations of multidimensional data and the dynamic evolution trends of event timelines, failing to accurately define the effective rule boundaries of response strategies in highly dynamic operating conditions. This directly leads to edge decision engines being unable to accurately determine the optimal execution timing and safe exit mechanism for event responses, frequently resulting in delayed response actions, premature termination, or over-response leading to resource waste, failing to meet the requirements for high-precision, deterministic, and rapid closed-loop control.

[0056] In view of this, the event response method, system, device and medium based on edge computing and decision-making provided in this application can effectively filter out background noise interference in the high-dimensional and complex data streams of ports and ocean shipping by fusing dual-channel features of operational data and environmental data and combining them with state space segmentation. This accurately separates high-risk areas of event anomalies, fundamentally solving the problems of false alarms and emergency delays caused by data lag and information silos in traditional shipping management systems. At the same time, by utilizing dynamic decision networks, event priority management and knowledge graph linkage, the limitations of traditional static hard-coded rule matching are broken. In the resource environment of edge computing, flexible adaptive adjustment of control boundaries and support for complex semantic reasoning are realized.

[0057] Building upon the aforementioned core precise triggering and boundary constraint mechanisms, the entire solution achieves a beneficial extension from local core algorithms to global systematic applications. By introducing joint time-frequency domain filtering and Kalman filtering algorithms, the solution predicts the trajectory and calculates the spatiotemporal offset of sudden events under dynamic noise interference conditions. Furthermore, it combines system load and network jitter of edge computing nodes for multi-dimensional data fusion correction, significantly enhancing the edge's anti-disturbance capability and decision robustness under extreme conditions. In addition, the solution comprehensively considers the dynamic impact of edge operating parameters such as data noise, system throughput, and packet loss rate on decision confidence, supplemented by over-limit degradation protection mechanisms for feedback execution latency and tracking errors. This ensures that the entire system possesses extremely high intelligent response depth while also considering the safety and fault tolerance boundaries at the hardware and software levels, fundamentally achieving an organic unity of "ultra-low latency," "high accuracy," and "strong engineering robustness" in edge-side event response.

[0058] Figure 1 This is a flowchart illustrating an event response method based on edge computing and decision-making provided in an embodiment of this application. Figure 1 The event response method based on edge computing and decision-making, as shown, includes at least the following steps: S100: Acquire the operational data and environmental data of the target system; S200: Perform dual-channel feature fusion on the operational data and environmental data, and filter out background interference through state space segmentation to separate the high-risk area of ​​event anomalies; S300: Input the feature matrix corresponding to the high-risk area of ​​event anomalies into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy; S400: Calculate the multidimensional trend difference of the operational data using a temporal difference operator, filter the multidimensional trend difference using a dual threshold to separate strong and weak correlation points, and identify the linear topology of strong and weak correlation points through a sequence topology analysis algorithm to extract temporal direction features; S500: Draw a decision evolution reference line in the state space based on the temporal direction features; S600: Solve the two spatiotemporal intersections of the decision evolution reference line and the dynamic constraint boundary, which are respectively used as the starting trigger point and the ending cutoff point for the event response execution.

[0059] S100: Acquire the target system's operating data and environmental data.

[0060] In this embodiment of the application, the event response method based on edge computing and decision-making includes, in step S100, acquiring the operating data and environmental data of the target system.

[0061] Specifically, the target system can be an industrial control node, smart gateway, IoT terminal, communication server, or power automation equipment deployed at the edge. The operational data includes, but is not limited to, the target system's CPU utilization, memory usage, interface throughput, service message latency, internal voltage amplitude, current frequency, or real-time status parameters of the actuators. The environmental data includes the physical and network parameters of the external environment in which the target system operates, such as ambient temperature, humidity, electromagnetic interference intensity, network signal-to-noise ratio, channel packet loss rate, and network jitter. The above data can be collected in real-time periodically through sensors, network management protocol interfaces, distributed message queues, or standard API interfaces built into the edge computing node, or asynchronously triggered by abnormal events.

[0062] Understandably, acquiring both operational and environmental data simultaneously provides multi-dimensional, composite feature support for subsequent accurate decision-making. Since edge devices typically operate in complex and dynamic open physical environments and network conditions, single system operational data is highly susceptible to distortion due to random fluctuations in the external environment. Using environmental data as a collaborative reference input allows for a deep binding between the system's internal operating state and the external mechanistic environment, laying a solid data foundation for eliminating sporadic background noise and accurately assessing the system's true abnormal state.

[0063] S200: Performs dual-channel feature fusion of operational data and environmental data, and filters out background interference through state space segmentation to separate high-risk areas of event anomalies.

[0064] In this embodiment of the application, the event response method based on edge computing and decision-making includes, in step S200, performing dual-channel feature fusion of running data and environmental data, and filtering out background interference through state space segmentation to separate high-risk areas of event anomalies.

[0065] Specifically, the dual-channel feature fusion employs a parallel feature extraction architecture, constructing independent running data feature channels and environmental data feature channels respectively. After extracting deep features from both channels using a deep learning network, high-dimensional feature mapping is performed using feature concatenation, weighted fusion, or a multi-head cross-attention mechanism to obtain a composite feature matrix. The state space segmentation maps this composite feature matrix to a multi-dimensional state geometric space composed of system state variables. Using a preset clustering algorithm, support vector machine boundary, or lightweight space partitioning hyperplane, baseline noise in the "environment-state" under normal operating conditions is filtered out, thereby accurately identifying and delineating high-risk areas of events in the multi-dimensional state space where data indicators have significantly shifted and potential safety risks exist.

[0066] Understandably, by combining dual-channel feature fusion and state space segmentation, the traditional "point-triggered" anomaly alarm mode based on a single scalar threshold can be upgraded to a "surface separation" recognition mode based on high-dimensional state geometric topology. This segmentation method allows normal, environmentally reasonable fluctuations in operational data (such as high-temperature environments causing adaptive frequency reduction in equipment, or weak network conditions causing a normal decrease in network throughput) to be identified and classified as normal background noise. Only abnormal offsets that truly deviate from the range of environmental adaptive compensation are classified as high-risk areas, greatly improving the accuracy of the system in identifying hidden events at the edge.

[0067] S300: Input the feature matrix corresponding to the high-risk area of ​​the event into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy.

[0068] In this embodiment of the application, the event response method based on edge computing and decision-making includes, in step S300, inputting the feature matrix corresponding to the high-risk area of ​​the event into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy.

[0069] Specifically, the dynamic decision-making network can be constructed using a lightweight Long Short-Term Memory (LSTM) network, a Gated Recurrent Unit (GRU), or a state-space model. After the feature matrix extracted from the high-risk area of ​​the event is input into the dynamic decision-making network, the network dynamically outputs one or more dynamic constraint boundaries for the candidate response strategies based on the current risk severity, the available computing resources of the edge nodes, battery power, and historical evolution patterns. Mathematically, these dynamic constraint boundaries are represented as a set of multidimensional inequality control envelopes that dynamically adjust with time and spatial state, used to strictly limit the maximum allowable action range, energy consumption threshold, bandwidth quota, and safe exit threshold for the response action.

[0070] Understandably, the generation of dynamic constraint boundaries defines dynamic "safety red lines" and resource quotas for event response at the edge. Traditional event response strategies often employ hard-coded fixed safety boundaries, which can easily lead to over-response or resource exhaustion when edge resources are scarce or multiple events occur concurrently. This application uses a dynamic decision network to dynamically scale and deform the constraint boundary based on the actual evolution characteristics of high-risk areas and real-time edge operating conditions, ensuring that the subsequently generated response strategy can effectively suppress event escalation without exceeding the local load-bearing limit of the edge nodes.

[0071] S400: It uses a time-series difference operator to calculate the multidimensional trend difference of the running data, uses a dual threshold to filter the multidimensional trend difference to separate strong and weak correlation points, and uses a sequence topology analysis algorithm to identify the linear topological structure of strong and weak correlation points to extract time-series directional features.

[0072] In this embodiment of the application, the event response method based on edge computing and decision-making includes, in step S400, using a temporal difference operator to calculate the multidimensional trend difference of the running data, using a dual threshold to filter the multidimensional trend difference to separate strong correlation points and weak correlation points, and using a sequence topology analysis algorithm to identify the linear topological structure of strong correlation points and weak correlation points to extract temporal direction features.

[0073] Specifically, a time-series difference operator is used to perform sliding window differential calculations on the running data stream to obtain multidimensional trend difference values ​​reflecting the rate of change of various indicators. Then, a high threshold Thigh and a low threshold Tlow are set to filter the multidimensional trend difference values ​​within a range: when the trend difference value exceeds Thigh, it is determined to be a strongly correlated point where the system state undergoes a drastic change; when the trend difference value is between Tlow and Thigh, it is determined to be a weakly correlated point with a gradual or potential evolutionary trend. Next, using a sequence topology analysis algorithm, these strongly correlated points and weakly correlated points are concatenated according to chronological order and the strength of causal association, constructing a directional topological directed graph, i.e., the linear topology structure; finally, using a feature mapping function, the temporal directional features representing the event propagation speed and state transition trend are vectorized and extracted from this linear topology structure.

[0074] Understandably, this step achieves the "micro-differentiation, discretization, and structuring" of raw time-series data into higher-order topological features. Time-series differential data collected by edge devices is often filled with instantaneous spikes and random noise caused by device jitter, making direct use highly susceptible to misjudgment. By discretizing continuous differential data into strong and weak key points using dual thresholds, and then reconstructing the scattered points into a continuous evolutionary skeleton using sequence topology analysis, instantaneous random disturbances can be effectively eliminated, thereby accurately capturing the true trajectory trend of events in the state space over time.

[0075] S500: Draw a decision evolution reference line in the state space based on the temporal direction characteristics.

[0076] In this embodiment of the application, the event response method based on edge computing and decision-making includes, in step S500, drawing a decision evolution reference line in the state space according to the temporal direction characteristics.

[0077] Specifically, based on the temporal direction features extracted in step S400, these features include the tangent slope, acceleration, and state transition vector of the event evolution. In the geometric state space constructed from the system's multidimensional operational data, using the latest state point in the current high-risk area of ​​the event as the physical starting point, forward prediction or linear extrapolation is performed along the vector direction indicated by the temporal direction features. This allows for the fitting or drawing of a geometric ray or curve characterizing the state transition trajectory of the target system on the future time axis, i.e., the decision evolution reference line.

[0078] Understandably, drawing the decision evolution reference line transforms abstract temporal trend characteristics into a concrete "virtual behavioral trajectory" in geometric state space. This reference line represents the true future development trend of the system state without any control or intervention in the current unexpected event. It provides an intuitive and calculable geometric reference for the next step of mathematically solving for the intersection of the "event evolution trend" and the "strategy safety range".

[0079] S600: The two spatiotemporal intersections of the solution decision evolution reference line and the dynamic constraint boundary serve as the starting trigger point and the ending cutoff point for event response execution, respectively.

[0080] In this embodiment of the application, the event response method based on edge computing and decision-making includes, in step S600, solving for two spatiotemporal intersections of the decision evolution reference line and the dynamic constraint boundary, which are respectively used as the starting trigger point and the ending cutoff point for the event response execution.

[0081] Specifically, in the multidimensional state space, the decision evolution reference line equation, representing the future state trajectory, and the dynamic constraint boundary equation, representing the control range of the security policy, are solved simultaneously to obtain two intersection points in the spatiotemporal dimension. The intersection point where the decision evolution reference line first crosses the dynamic constraint boundary and enters the abnormal envelope region is defined as the starting trigger point, used to instruct the event decision engine to immediately intervene and initiate the response control program. The intersection point where the decision evolution reference line again crosses the dynamic constraint boundary, leaves the abnormal envelope region, and returns to the normal security baseline is defined as the termination cutoff point, used to instruct the event decision engine to safely exit and shut down the current response action.

[0082] Understandably, by calculating the spatiotemporal intersection of the decision evolution reference line and the dynamic constraint boundary, precise entry and smooth exit of event response closed-loop control on the time axis are achieved. This mechanism avoids the common problem in traditional methods where the response action terminates prematurely or lags behind due to reliance on a single static hard threshold. It enables edge computing nodes to apply control precisely within the most critical time window, ensuring absolute system security while maximizing the conservation of valuable computing resources, communication bandwidth, and execution energy consumption on the edge side.

[0083] In this embodiment, the feature matrix corresponding to the high-risk area of ​​an event anomaly is input into a dynamic decision network to generate a dynamic constraint boundary for response candidate strategies. This includes: inputting the feature matrix corresponding to the high-risk area of ​​the event anomaly into a dynamic decision network based on a temporal deep learning network; constructing a multi-scale policy response map through a multi-scale temporal feature network; applying multi-dimensional state alignment technology combined with a preset policy intersection-union threshold for key state feature selection; and outputting the dynamic constraint boundary through high-dimensional deconvolution mapping and decoding.

[0084] Specifically, the temporal deep learning network can employ Long Short-Term Memory (LSTM), Bidirectional Gated Recurrent Unit (Bi-GRU), or Temporal Convolutional Network (TCN). The multi-scale temporal feature network captures both the instantaneous anomaly features of sudden events at the micro-level and the long-term evolution trend features at the macro-level by setting time sliding windows of different lengths or introducing dilated convolution operators. These two sets of temporal features are then concatenated and dimensionally mapped in parallel to construct the multi-scale policy response map. The multi-dimensional state alignment technique utilizes Mahalanobis distance or Euclidean distance to measure the topological similarity between the real-time multi-dimensional state vector of the target system and the preset standard response state. Combined with a preset policy intersection-over-union (IoU) threshold, key state features with the highest overlap and meeting safety boundary requirements are selected. Finally, high-dimensional deconvolution layers are used for upsampling mapping and nonlinear decoding to restore the features and output them as dynamically evolving constraint boundaries over time.

[0085] Understandably, constructing multi-scale policy response maps and using state alignment and intersection-over-union (IoU) thresholding for key feature filtering enables the system to define policy boundaries with extremely high granularity and environmental adaptability at the edge. Precise feature filtering using IoU thresholding for both positive and negative samples effectively eliminates ineffective or inefficient candidate control policies, while high-dimensional deconvolution decoding ensures the continuity and smoothness of the final output constraint boundaries in the spatiotemporal dimensions, avoiding severe and frequent oscillations in the constraint boundaries due to data spikes, thus guaranteeing the stability of decision control.

[0086] In this embodiment, after extracting the start trigger point and end cutoff point as the start trigger point and end cutoff point for event response execution, the method further includes: applying a time-domain filtering algorithm to obtain the target state features of the start trigger point and end cutoff point within an initial time-domain window. Real-time retrieval and target state feature relocation are performed in the subsequent time series of the data stream using frequency-domain correlation filtering. The Kalman filter algorithm is combined to predict the trajectory of the sudden event under dynamic noise interference conditions, and the spatiotemporal offset of the start trigger point and end cutoff point is calculated.

[0087] Specifically, the time-domain filtering algorithm can be an adaptive Wiener filter or a median filter, used to eliminate instantaneous high-frequency white noise near the start trigger point and the end cutoff point within the initial time-domain window, thereby locking onto the target state features reflecting its core anomaly trend. The frequency-domain correlation filter adopts a discriminant correlation filter (DCF) architecture, transforming the target state features into the frequency domain for fast sliding window cross-correlation matrix calculation, thereby retrieving and accurately repositioning the spatiotemporal locations of these two points in real time and at high speed in subsequent time series of the data stream. Based on this, the Kalman filter algorithm establishes an extended linear equation about the event state transition, weighted iteratively fusing the current real-time observations and historical predictions to predict the true evolution trajectory of the sudden event under dynamic noise interference conditions, and then accurately calculates the spatiotemporal offset of the start trigger point and the end cutoff point on the spatiotemporal axis caused by environmental dynamic disturbances.

[0088] Understandably, the combination of time-domain filtering and frequency-domain correlation filtering significantly improves the real-time performance and accuracy of the system in tracking "critical trigger nodes" in continuous dynamic data streams, effectively preventing nodes from being lost midway or drifting due to high-frequency noise. Further introducing Kalman filtering for trajectory prediction of evolution trends endows edge nodes with powerful dynamic feedforward prediction capabilities. Even under harsh operating conditions with sudden and severe noise interference, it can accurately calculate the spatiotemporal offset of the start and end points, providing high-precision data support for subsequent adaptive deviation correction.

[0089] In this embodiment, the method further includes: acquiring the system load change and communication network jitter amplitude of the edge computing node respectively; fusing the system load change, communication network jitter amplitude, and spatiotemporal offset into multi-dimensional data to calculate the timing decision deviation between the starting trigger point or termination cutoff point and the actual closed-loop operation point; and outputting an adaptive correction command based on the instantaneous timing decision deviation to drive the parameter adjustment engine to dynamically adjust, ensuring that the event decision engine continuously aligns with the starting trigger point or termination cutoff point.

[0090] Specifically, the system load change includes the CPU utilization increment of the edge computing nodes, the memory bandwidth consumption change rate, and the remaining computing resource quota; the communication network jitter amplitude refers to the variance or root mean square error of the local communication channel data packet transmission delay. The multi-dimensional data fusion can employ an extended Kalman filter (EKF) algorithm or a fuzzy neural network fusion device to perform multi-dimensional weighted fusion of the above indicators characterizing the degree of node performance degradation and the spatiotemporal offset characterizing external event disturbances, thereby accurately quantifying and calculating the instantaneous timing decision deviation between the theoretical start and end points and the actual hardware closed-loop operation points. After receiving the adaptive correction command generated based on this deviation, the parameter adjustment engine dynamically adjusts the hyperparameters or internal sliding window step size of the dynamic decision network, thereby dynamically compensating for the control lag caused by its own computational delay or communication bottlenecks.

[0091] Understandably, this step fully considers the impact of the physical and network health status of edge computing nodes on the accuracy of closed-loop control. Due to limitations in hardware resources, edge nodes experience significantly increased processing and transmission latency under high-load conditions with multiple concurrent events or when facing weak network jitter. By fusing multi-dimensional data of the node's load and network metrics with the spatiotemporal offset of events, it is possible to perceive and compensate for execution lag caused by performance degradation in real time, ensuring that the event decision engine does not miss or misalign critical start and stop control opportunities due to local hardware lag.

[0092] In this embodiment, the method further includes: acquiring the feedback execution delay and dynamic tracking error of the event decision engine during the emergency response process. When the feedback execution delay or dynamic tracking error exceeds a preset safety threshold, a degradation protection command is output. Based on the degradation protection command, the event decision engine is controlled to stop the current response operation and reset the running state of the edge computing node.

[0093] Specifically, the feedback execution delay refers to the actual closed-loop time interval from when the event decision engine issues a control command to when the external actuator fully completes its response action; the dynamic tracking error refers to the multidimensional Euclidean distance between the target system state value expected by the control command and the actual state value fed back by the external actuator. The preset safety thresholds include the maximum allowable execution delay upper limit and the maximum allowable tracking error absolute value threshold. If any of the above indicators exceeds the corresponding preset safety threshold due to actuator jamming, severe network disconnection, or sudden hardware / software failure, the system's internal circuit breaker will be immediately triggered, outputting a degradation protection command. Based on this command, the event decision engine immediately stops the high-load current response job, releases the occupied core computing and communication resources, and resets the edge computing node's operating state by performing a hardware / software reset, returning it to a safe normal mode.

[0094] Understandably, degradation protection mechanisms constitute the last line of defense for edge response systems. In complex industrial or distributed network operating conditions, external actuator failures or extreme interference are unavoidable. By monitoring and feeding back execution latency and dynamically tracking errors in real time, the system gains self-awareness and self-healing capabilities. When errors or latency exceed control limits, timely degradation protection, circuit breaking, and resetting the operating state effectively prevent edge computing nodes from getting stuck in deadlocks, hardware overload damage, or a wider system cascading failure due to blind over-response, significantly improving the system-level fault tolerance and robustness of the entire solution in engineering implementation.

[0095] In this embodiment, the feature matrix corresponding to the high-risk area of ​​the event is input into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy. This includes: acquiring edge condition parameters during feature acquisition, which include data noise intensity, system throughput, and network packet loss rate; determining environmental correction coefficients based on the edge condition parameters; and applying these environmental correction coefficients to dynamically correct the decision confidence level output by the dynamic decision network.

[0096] Specifically, the data noise intensity can be quantitatively represented by calculating the variance, standard deviation, or signal-to-noise ratio of the data stream running within the current sliding window; the system throughput can be obtained through the operating system kernel monitoring component of the edge node, manifested as the number of I / O requests processed per second, the CPU scheduling queue length, or the number of service packets forwarded per second; the network packet loss rate can be determined by statistically analyzing the TCP retransmission rate or periodic ICMP probe loss rate in the network protocol stack. When determining the environmental correction coefficient, the normalized data noise intensity, system throughput, and network packet loss rate are input into a preset multivariate negative feedback function for calculation. Typically, when the operating conditions on the edge side deteriorate (e.g., increased noise, system throughput approaching saturation, or a surge in network packet loss rate), the calculated environmental correction coefficient will decrease accordingly within the range of 0 to 1, characterizing the degree of degradation of the current environmental data.

[0097] Understandably, introducing edge condition parameters for correction based on the original confidence level of the model output can effectively solve the problem of "blind confidence" in deep learning networks at the edge. Traditional neural networks sometimes still output high confidence levels when faced with input features that are incomplete due to packet loss or distorted due to noise, thus leading to incorrect boundary judgments. This application constructs environmental correction coefficients by collecting underlying physical and network condition parameters in real time and actively intervenes in the decision confidence level of the upper-layer algorithm. This achieves cross-layer closed-loop linkage between the underlying physical environment constraints and the upper-layer intelligent algorithm, greatly improving the fault tolerance and true accuracy of the dynamic constraint boundary generation scheme under harsh and variable edge conditions.

[0098] In this embodiment, the decision confidence score output by the dynamic decision network is dynamically corrected using an environmental correction coefficient. This includes multiplying the decision confidence score by the environmental correction coefficient to obtain a corrected event decision score. Response candidate strategies are then re-evaluated and screened based on the corrected event decision score.

[0099] Specifically, let the initial decision confidence of the dynamic decision network for a given response candidate strategy be . The environmental correction factor obtained by the solution is (in, The corrected event decision score is then... Calculate according to the following formula:

[0100]

[0101] in, This represents the corrected event decision score; This represents the original decision confidence level output by the dynamic decision network. This represents the environmental correction coefficient calculated based on edge condition parameters. After obtaining the corrected event decision score for each candidate response strategy, the system compares this score with a preset security trust threshold. If the event decision score of a strategy is lower than the threshold, the strategy is directly removed from the candidate list. If multiple candidate response strategies have scores higher than the threshold, these strategies are sorted in descending order of score, and the strategy with the highest score is determined as the preferred event response strategy to be finally issued to the executing agency.

[0102] Understandably, using a multiplicative discount mechanism like multiplication to calculate the final event decision score offers significant mathematical advantages in terms of lightweight computation, making it highly suitable for edge computing nodes with limited computing resources. This mechanism ensures the control effect of "dynamic environment veto"—even if the original theoretical confidence level given by the dynamic decision network is extremely high, if the current edge network is on the verge of collapse or the local hardware system is severely overloaded (resulting in an extremely low environment correction coefficient), the final event decision score will be forcibly suppressed. Through this re-evaluation and screening based on operating condition scores, high-risk strategies that are theoretically optimal but are highly susceptible to execution failure due to hardware latency or communication loss under current physical conditions can be thoroughly filtered out, ensuring that the strategies ultimately retained and executed have absolute practical feasibility and security at the current edge.

[0103] The following section uses an intelligent shipping and end-to-end logistics collaboration system as an example implementation scenario to elaborate on the complete workflow of the event response method provided in this application:

[0104] Phase 1: Acquisition and Preprocessing of Multi-Source Heterogeneous Maritime Data

[0105] In actual operation, edge computing nodes are widely deployed in ship terminals, port smart gateways, and multimodal transport hubs. The system uses these edge nodes to collect operational and environmental data from the target system in real time and in parallel.

[0106] Operational data acquisition: The edge nodes of the ship terminal acquire real-time data on the ship's Automatic Identification System (AIS) location and trajectory, cargo hold container status (such as temperature and humidity of refrigerated containers, gas pressure of hazardous chemicals), quay crane lifting equipment operating parameters, ship main propulsion system load, main engine fuel consumption rate, and historical booking, release and other business operation status.

[0107] Environmental data acquisition: Simultaneously collect external meteorological and hydrological information (such as typhoon path evolution, wave height, tide table, wind speed), port yard resource saturation, Suez Canal or key waterway traffic efficiency index, and the system load change of edge nodes and the signal-to-noise ratio and packet loss rate of wireless satellite communication network.

[0108] Hot and cold data separation storage: The collected data is separated into hot and cold data based on the access frequency at the underlying level. High-frequency, low-latency access data, such as real-time AIS location, terminal berth occupancy rate, and instantaneous typhoon weather warnings, are stored directly as "hot data" on high-performance solid-state drives (SSDs) for instant retrieval and optimal solution planning. Low-frequency access data, such as fuel consumption curves over the past 5 years, 10-year long-term freight rate contracts, and historical accident investigation reports, are stored as "cold data" on low-cost hard disk drives (HDDs) for long-term operational benefit analysis.

[0109] Phase Two: Cross-channel Feature Fusion and Delineation of High-Risk Anomalies

[0110] After acquiring multidimensional time-series maritime data, the system sends it to a dual-channel feature fusion layer:

[0111] Feature fusion processing: The first feature channel extracts operational data such as ship speed, main engine fuel consumption, and cargo hold internal status using a fast sliding window; the second feature channel extracts macro-trend data such as external typhoon path changes, wave turbulence amplitude, and terminal efficiency. Through a multi-head cross-attention mechanism, external severe weather or port congestion parameters are deeply coupled to the ship's own status features.

[0112] State Space Segmentation: The fused high-dimensional composite feature matrix is ​​projected into a multi-dimensional state geometric space. Through the state space segmentation engine, the system automatically filters out baseline fluctuation noise caused by normal economic speed adjustments, normal seasonal tidal fluctuations, etc. Therefore, when a freight forwarder experiences a temperature control system malfunction in a 500-ton lithium-ion battery (dangerous goods) cargo hold, or when a cargo ship's main propulsion system experiences an abnormal power drop due to a clogged fuel filter, its feature vectors will undergo significant shifts in the multi-dimensional state space. The segmentation engine, through hyperplane partitioning, precisely delineates this cluster of deviation points from the normal baseline, defining them as "high-risk areas for event anomalies," thus achieving precise capture of hidden and complex shipping risks.

[0113] Phase 3: Dynamic Decision Network and Policy Constraint Boundary Generation

[0114] The feature matrix corresponding to the high-risk areas of the above-mentioned events is input into the dynamic decision network of the edge gateway or central cloud (such as a lightweight Bi-GRU or Mamba model).

[0115] Event Priority Management and Business Mapping: During the decision generation process, the event priority management module adopts a dynamic weight allocation model based on deep reinforcement learning, dynamically optimizing the event based on event type, user value attributes, and real-time resource status. For example, when a Norwegian salmon transport container belonging to a cold chain logistics company (with high Lifetime Value) triggers a cargo temperature control anomaly, the system automatically identifies its historical cooperation frequency and spoilage risk cost, elevates the event to P1 level, and prioritizes the allocation of available refrigerated resources. If a port container hoisting equipment experiences a sudden malfunction due to motor overheating or a freight forwarder payment anomaly, resulting in port congestion, the highest priority P0 level response is directly triggered.

[0116] Boundary rule generation: The dynamic decision-making network uses business objectives such as "reducing per-container transportation costs" or "minimizing rescue and claims risks" as optimization functions, and makes dynamic decisions by combining historical navigation behavior and the aforementioned real-time context data. After deconvolution mapping and decoding, the network outputs a set of dynamic constraint boundaries in the state space for the response candidate strategies. These boundaries dynamically limit the maximum allowable nautical miles around the typhoon, the maximum fuel consumption threshold, and the 48-hour availability of port dangerous goods storage yards, under the constraints of the current typhoon range, the remaining temperature-controlled cargo space on the ship, and the main engine load.

[0117] Phase Four: Trend Differentiation and Temporal Topology Reconstruction

[0118] In order to accurately pinpoint the development of events, the system synchronously uses the time-series difference operator to perform sliding window differential calculations on the running data stream to obtain multidimensional trend differences of cargo hold temperature rise gradient, ship stall slope, or port yard saturation change rate.

[0119] Dual-threshold key point separation: Multidimensional trend difference is filtered using a high threshold T-high and a low threshold T-low. Points with difference values ​​exceeding the high threshold are separated into "strongly correlated points" (such as complete engine shutdown or sudden equipment failure), while points between the two thresholds are separated into "weakly correlated points" (such as slow speed reduction due to increasing wind and waves or gradual depletion of yard resources).

[0120] Topology and Directional Feature Extraction: The sequence topology analysis algorithm connects these strongly and weakly correlated points with directed line segments according to the time stamp order to construct a linear topology that reflects the spread or evolution trend of the event. Through the feature mapping function, the geometric velocity and state transition vector of this topological skeleton are merged, and finally the "temporal directional features" that represent the spread speed and transition trend of the event on the future time axis are vectorized and extracted.

[0121] Phase 5: Drawing the Decision Evolution Reference Line in the State Space

[0122] The decision reference line plotting module takes over the aforementioned temporal directional features. The system uses the latest observed state of the current high-risk anomaly zone as the geometric starting point in the state space, and uses the development state transition vector in the temporal directional features as the directional derivative to perform linear extrapolation or fitting to the future time axis in multidimensional space.

[0123] Virtual evolution trajectory fitting: This step successfully plots a "decision evolution reference line" with predictive properties in the state space. This reference line, in a digital and computable geometric form, concretely predicts whether the target vessel will drift into a strong wind and wave area or whether the temperature of hazardous chemicals in the cargo hold will completely get out of control in the next few hours, without external intervention or changing course.

[0124] Phase Six: Intersection Calculation and End-to-End Fault-Tolerant Self-Healing Execution

[0125] Finally, the spatiotemporal intersection solution module solves the decision evolution reference line equation drawn in step S500 and the dynamic constraint boundary equation generated in step S300 simultaneously, accurately calculating the two key spatiotemporal intersections:

[0126] The initial trigger point and edge-local execution: The first intersection point, defined as the moment when the reference line first intersects the dynamic constraint boundary, is the "initial trigger point." During peak shipping seasons or when network bandwidth is limited, the adaptive execution engine, upon reaching this trigger point, prioritizes calling edge node resources to execute localized instructions (such as immediately triggering local cargo hold cooling or activating the auxiliary propulsion system on-site). In multi-system multimodal transport, the cross-chain event bridging module will automatically synchronize the event through blockchain smart contracts (such as the port system updating "loading completed," the railway system triggering "car wagon scheduling priority" within seconds, and the logistics platform adjusting the ETA synchronously), eliminating cross-platform collaboration trust costs and information lag.

[0127] Fault self-healing and alternative solution recommendations: If the adaptive execution engine detects a fault in the execution node during execution (such as a port quay crane jamming due to motor overheating or a TMS (Transportation Management System) outage), it will automatically activate the self-healing control logic. The system will immediately activate the backup hoisting equipment at the same berth or switch to the cloud-based disaster recovery maritime dedicated node; at the same time, the knowledge graph-enhanced decision-making module will integrate marketing and shipping knowledge graphs (user-product-resource-environment associations) to perform semantic reasoning and recommend alternative solutions (such as automatically adjusting the loading and unloading sequence according to the segregation requirements of the International Maritime Dangerous Goods Code, or coordinating with the customs of the destination port to prioritize handling or detour to an alternative safe harbor).

[0128] Termination Cutoff Point and Closed-Loop Exit: The second intersection point is defined as the point where the decision reference line, after control intervention, crosses the dynamic constraint boundary again from the inside out and returns to the normal safety baseline. When the timeline evolves to this cutoff point, it means that the system state (such as cargo hold temperature, ship navigation, and yard turnover rate) has been safely reset and restored to a safe and stable normal mode. The event decision engine issues instructions to control the external actuators to smoothly exit the response operation.

[0129] Phase Seven: Post-Evaluation and Causal Optimization of the Strategy (Ensuring Continuous Iteration of the Closed Loop)

[0130] After the entire event response process is completed, the system asynchronously archives the data to cold storage. The strategy effectiveness evaluation module integrates a causal inference model, calling upon these historically accumulated data (such as ship AIS tracks, port throughput records, and fuel consumption curves) to construct a sandbox counterfactual analysis framework, initiating closed-loop strategy optimization.

[0131] For strategies such as "prioritizing the loading and unloading of cold chain goods", the model uses propensity score matching to exclude weather-related variables such as typhoons, accurately quantifying the true net effect of the strategy on customer satisfaction.

[0132] For the "freight cost and oil price linkage mechanism" or "fast customs clearance package promotion", the model performs difference-in-differences analysis or instrumental variable method to handle endogeneity issues and quantifies the contribution of the strategy and customer churn rate indicators.

[0133] The evaluation results will be directly fed back into the optimization function of the reinforcement learning engine for hyperparameter fine-tuning. Furthermore, encrypted model gradient parameters will be exchanged among multiple shipping companies and ports through the federated learning module (the original data will be kept locally to protect cross-domain user privacy). This collaborative training will update the global fault prediction and route optimization models, thereby enabling cross-domain iteration and continuous evolution of the entire system's decision-making intelligence while ensuring data privacy.

[0134] Figure 2This is a schematic diagram of the modules of an event response system provided in an embodiment of this application. For example... Figure 2 The event response system 10 shown includes at least the following components: a data acquisition module 11, a feature fusion module 12, a decision boundary generation module 13, a time-series trend analysis module 14, a decision reference line drawing module 15, and a spatiotemporal intersection calculation module 16.

[0135] In this embodiment, the data acquisition module 11 acquires the operating data and environmental data of the target system; please refer to the following for details. Figure 1 The details and their corresponding descriptions are not repeated here.

[0136] In this embodiment, the feature fusion module 12 performs dual-channel feature fusion of runtime data and environmental data, and filters out background interference through state space segmentation to separate high-risk areas of event anomalies.

[0137] In this embodiment, the decision boundary generation module 13 inputs the feature matrix corresponding to the high-risk area of ​​the event anomaly into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy. Please refer to the following for details. Figure 1 The details and their corresponding descriptions are not repeated here.

[0138] In this embodiment, the time-series trend analysis module 14 uses a time-series difference operator to calculate the multidimensional trend difference of the running data, uses a dual-threshold filter to separate strong and weak correlation points, and uses a sequence topology analysis algorithm to identify the linear topological structure of strong and weak correlation points to extract time-series directional features. Please refer to the following for details. Figure 1 The details and their corresponding descriptions are not repeated here.

[0139] In this embodiment, the decision reference line drawing module 15 draws decision evolution reference lines in the state space according to the temporal direction characteristics. Please refer to the following for details. Figure 1 The details and their corresponding descriptions are not repeated here.

[0140] In this embodiment, the spatiotemporal intersection solution module 16 solves for two spatiotemporal intersections between the decision evolution reference line and the dynamic constraint boundary, which serve as the starting trigger point and the ending cutoff point for event response execution, respectively. Please refer to the following for details. Figure 1 The details and their corresponding descriptions are not repeated here.

[0141] Figure 3 This is an electronic device 20 provided in one embodiment of this application. For example... Figure 3 As shown, the electronic device 20 includes at least the following components: a processor 21 and a memory 22.

[0142] In this embodiment, the memory 22 is used to store executable instructions for the processor 21, which, when configured to execute instructions, implement... Figure 1 The method shown in the figure.

[0143] In one embodiment of this application, the program operating in the electronic device 20 may be a program that controls a central processing unit (CPU) or similar device to achieve the functions of the above-described embodiments of the present invention (a program that enables the computer to function). The information processed by these devices is then temporarily stored in random access memory (RAM) during processing, and subsequently stored in various ROMs such as read-only memory (Flash ROM) and hard disk drives (HDDs), and read, corrected, and written by the CPU as needed.

[0144] It should be noted that a portion of the electronic device 20 described above can also be implemented using a computer. In this case, the program for implementing the control function can be recorded on a computer-readable recording medium, and the program recorded on the recording medium can be read into the computer and executed.

[0145] It should be noted that the term "computer" as used here refers to a computer built into electronic device 20, employing hardware including an operating system and peripheral devices. Furthermore, "computer-readable recording media" refers to removable media such as floppy disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard drives built into the computer.

[0146] Furthermore, a "computer-readable recording medium" can include: a medium that dynamically stores a program for a short period of time, such as a communication line used when transmitting a program via a network such as the Internet or a communication line such as a telephone line; or a medium that stores a program for a fixed period of time, such as volatile memory inside a computer that serves as a server or client in this case. In addition, the aforementioned program can be a program used to implement the above-mentioned functions, or it can be a program that can implement the above-mentioned functions by combining with programs already recorded in the computer.

[0147] It is understood that the event response method, system, device and medium based on edge computing and decision-making provided in this application can effectively filter out background noise interference in the high-dimensional and complex data streams of ports and oceans by fusing dual-channel features of operational data and environmental data and combining them with state space segmentation. This can accurately separate high-risk areas of event anomalies and fundamentally solve the problems of false alarms and emergency delays caused by data lag and information silos in traditional shipping management systems. At the same time, by using dynamic decision networks, event priority management and knowledge graph linkage, the limitations of traditional static hard-coded rule matching are broken. In the resource environment of edge computing, flexible adaptive adjustment of control boundaries and support for complex semantic reasoning are realized.

[0148] Building upon this foundation, by filtering strong and weak correlation points and reconstructing the evolution trend topology through temporal differential operators, a decision evolution reference line can be intelligently drawn. Furthermore, a system of simultaneous mathematical equations precisely calculates the spatiotemporal intersection of this reference line and the dynamic constraint boundary. This perfectly locks the starting trigger point and ending cutoff point of event response execution, effectively overcoming the shortcomings of existing solutions, such as delayed response actions, premature termination, or over-response leading to ship fuel waste and terminal resource congestion. Combined with the reliable synchronization of cross-chain event bridging, the fault self-healing logic of the adaptive execution engine, and the closed-loop optimization of digital twins and causal inference, the entire system achieves a high degree of intelligent response while also considering the safety and fault tolerance boundaries of multi-party cross-modal transport collaboration. This truly realizes the organic unity of "ultra-low latency," "automatic reliable synchronization," and "strong engineering robustness" in shipping-side event response.

[0149] Those skilled in the art should recognize that the above embodiments are only used to illustrate this application and are not intended to limit this application. Any appropriate changes and variations made to the above embodiments within the essential spirit and scope of this application fall within the scope of protection claimed in this application.

Claims

1. An event response method based on edge computing and decision-making, applied to edge computing nodes, characterized in that, The method includes: Acquire the target system's operational and environmental data; The operational data and the environmental data are fused using dual-channel features, and background interference is filtered out by state space segmentation to separate high-risk areas of event anomalies. The feature matrix corresponding to the high-risk area of ​​the event is input into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy. The multidimensional trend difference of the running data is calculated using a time-series difference operator. The multidimensional trend difference is filtered using a dual threshold to separate strong and weak correlation points. The linear topological structure of the strong and weak correlation points is identified by a sequence topology analysis algorithm to extract time-series directional features. Draw a decision evolution reference line in the state space based on the temporal direction characteristics; The two spatiotemporal intersections of the decision evolution reference line and the dynamic constraint boundary are calculated and used as the starting trigger point and the ending cutoff point for the event response execution, respectively.

2. The event response method based on edge computing and decision-making according to claim 1, characterized in that, The step of inputting the feature matrix corresponding to the high-risk area of ​​the event into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy includes: The feature matrix corresponding to the high-risk area of ​​the event is input into a dynamic decision network based on a temporal deep learning network; Construct a multi-scale policy response map using a multi-scale temporal domain feature network; Key state features are selected by applying multi-dimensional state alignment technology combined with a preset strategy intersection-union ratio threshold. The dynamic constraint boundary is output through high-dimensional deconvolution mapping and decoding.

3. The event response method based on edge computing and decision-making according to claim 2, characterized in that, After extracting the start trigger point and the end cutoff point as the start trigger point and end cutoff point for event response execution, the method further includes: The target state features of the start trigger point and the end cutoff point within the initial time domain window are obtained by applying a time domain filtering algorithm. In the subsequent time series of the data stream, frequency domain correlation filtering is used for real-time retrieval and relocation of the target state features; By combining the Kalman filter algorithm, the trajectory of the evolution trend of the sudden event under dynamic noise interference is predicted, and the spatiotemporal offset between the starting trigger point and the ending cutoff point is calculated.

4. The event response method based on edge computing and decision-making according to claim 3, characterized in that, The method further includes: The system load change and communication network jitter amplitude of the edge computing node are obtained respectively; The system load change, the communication network jitter amplitude and the spatiotemporal offset are fused in multiple dimensions to calculate the timing decision deviation between the starting trigger point or the termination cutoff point and the actual closed-loop operation point. Based on the instantaneous timing decision deviation, an adaptive correction command is output to drive the parameter adjustment engine to make dynamic adjustments, so that the event decision engine is continuously aligned with the starting trigger point or the ending cutoff point.

5. The event response method based on edge computing and decision-making according to claim 4, characterized in that, The method further includes: The event decision engine's feedback execution delay and dynamic tracking error during the emergency response process are obtained. When the feedback execution delay or the dynamic tracking error exceeds a preset safety threshold, a degradation protection command is output; The degradation protection command controls the event decision engine to stop the current response job and reset the running status of the edge computing node.

6. The event response method based on edge computing and decision-making according to claim 1, characterized in that, The step of inputting the feature matrix corresponding to the high-risk area of ​​the event into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy includes: The edge operating parameters during feature acquisition are obtained, including data noise intensity, system throughput, and network packet loss rate. The environmental correction coefficient is determined based on the edge condition parameters, and the environmental correction coefficient is applied to dynamically correct the decision confidence of the dynamic decision network output.

7. The event response method based on edge computing and decision-making according to claim 6, characterized in that, The application of the environmental correction coefficient to dynamically correct the decision confidence of the dynamic decision network output includes: Multiply the decision confidence level by the environmental correction coefficient to obtain the corrected event decision score; The response candidate strategies are re-evaluated and filtered based on the corrected event decision score.

8. An event response system based on edge computing and decision-making, characterized in that, The system is adapted to implement the method as described in any one of claims 1 to 7, wherein the system comprises: The data acquisition module acquires the target system's operational and environmental data. The feature fusion module performs dual-channel feature fusion of the operational data and the environmental data, and filters out background interference through state space segmentation to separate high-risk areas of event anomalies. The decision boundary generation module inputs the feature matrix corresponding to the high-risk area of ​​the event into the dynamic decision network to generate the dynamic constraint boundary of the response candidate strategy. The time-series trend analysis module uses a time-series difference operator to calculate the multidimensional trend difference of the running data, uses a dual threshold to filter the multidimensional trend difference to separate strong correlation points and weak correlation points, and uses a sequence topology analysis algorithm to identify the linear topological structure of the strong correlation points and the weak correlation points to extract time-series direction features. The decision reference line drawing module draws decision evolution reference lines in the state space according to the temporal direction characteristics. The spatiotemporal intersection calculation module calculates the two spatiotemporal intersections of the decision evolution reference line and the dynamic constraint boundary, which serve as the starting trigger point and the ending cutoff point for the event response execution, respectively.

9. An electronic device, characterized in that, include: processor; as well as A memory having computer-readable instructions stored thereon for controlling the processor to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program is configured to perform the method described in any one of claims 1 to 7 when executed.

Images