Security Management Method for Distributed Monitoring Nodes in Outdoor Monitoring Systems
By constructing a multi-source influencing factor correlation model and a hierarchical arbitration mechanism, the problem of insufficient cross-domain correlation cognition in distributed outdoor monitoring systems was solved, achieving optimized control of security isolation and resource collaboration, and improving the system's adaptability and accuracy.
Patent Information
- Application Number
- CN202611474681.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-21
- Publication Date
- 2026-11-17
AI Technical Summary
In distributed outdoor monitoring systems, the lack of cross-domain correlation cognition and execution feedback correction mechanisms leads to conflicts between node state adjustment commands and resource mutual assistance scheduling commands at the physical topology level. Furthermore, the decision-making model gradually deviates from the true physical laws during long-term operation, affecting the accuracy of collaborative control.
A multi-source influencing factor correlation model is constructed, and the correlation between influencing factors is represented by a directed weighted graph. Node state deduction is performed, and a hierarchical arbitration mechanism is introduced to generate collaborative control instructions. The model is then corrected using execution feedback and environmental verification data to form a closed-loop control system.
It achieves global optimality of secure isolation and resource coordination under complex operating conditions, avoids topology conflicts and energy waste, and ensures the adaptive evolution and long-term stability of monitoring nodes.
Smart Images

Figure CN123113603A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of intelligent security and Internet of Things technology, and in particular to a method for security management of distributed monitoring nodes based on outdoor monitoring systems. Background Technology
[0002] Distributed outdoor monitoring systems are critical infrastructure deployed in scenarios such as urban transportation, forest fire prevention, and industrial parks. They typically consist of a large number of monitoring nodes distributed outdoors. Each node is connected to the central management platform via wired or wireless networks and is equipped with surge protectors and independent photovoltaic energy storage units to achieve video acquisition, transmission, and basic protection functions.
[0003] However, in practical applications, the physical environment in which outdoor monitoring nodes are located is highly dynamic and characterized by multi-factor coupling. Various heterogeneous data, such as atmospheric electric fields, radar echoes, animal behavior, battery state of charge, and load power consumption, exhibit complex nonlinear correlations across time and space. When the system triggers protection actions or energy dispatch strategies based on a single-dimensional threshold, the generated node state adjustment commands and resource mutual aid dispatch commands often conflict directly at the physical topology level because the implicit causal relationships between cross-domain influencing factors are not analyzed. For example, safety isolation requires disconnecting interconnecting lines to prevent surge propagation, while energy mutual aid requires maintaining line connectivity to sustain power supply. Furthermore, the lack of a closed-loop mechanism to feed back the actual physical response and environmental verification data after command execution to the decision model prevents the preset correlation rules and prediction parameters from adaptively correcting with environmental evolution. This causes the decision model to gradually deviate from true physical laws over long-term operation, leading to a continuous decline in the accuracy of collaborative control commands. Summary of the Invention
[0004] To address the problems of conflicting security management decisions and model drift in existing technologies due to the lack of cross-domain correlation cognition and execution feedback correction mechanisms, this application proposes a distributed monitoring node security management method based on outdoor monitoring systems. This method enables dynamic correction and security isolation of the multi-source influencing factor correlation model and adaptive hierarchical arbitration of resource coordination conflicts.
[0005] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for security management of distributed monitoring nodes in an outdoor monitoring system includes: acquiring multi-source state influencing factor data of distributed monitoring nodes, and updating the multi-source influencing factor correlation model based on the multi-source state influencing factor data. The multi-source influencing factor correlation model is a directed weighted graph model representing the correlation relationship, correlation strength, and confidence level between influencing factor nodes. Node state is simulated using the multi-source influencing factor correlation model to determine the current environmental risk level, and node state adjustment instructions and resource mutual aid scheduling instructions are generated. When a node state adjustment instruction requires disconnecting the physical interconnection between distributed monitoring nodes, while a resource mutual aid scheduling instruction requires maintaining the physical interconnection between distributed monitoring nodes, hierarchical arbitration is performed based on the current environmental risk level to generate a collaborative control instruction, which is then sent to the corresponding distributed monitoring nodes. Execution feedback data and environmental verification data after the distributed monitoring nodes execute the collaborative control instructions are acquired, and the multi-source influencing factor correlation model is corrected based on the execution feedback data and environmental verification data. The corrected multi-source influencing factor correlation model is used for node state simulation in the next cycle.
[0006] The above scheme constructs a directed weighted graph model that represents the relationship between influencing factors, and uses execution feedback and environmental verification data to perform closed-loop correction on the model, so that the decision basis can be adaptively updated as the physical environment evolves. At the same time, when topological conflicts occur between security isolation and resource coordination, graded arbitration is carried out according to the risk level to ensure the global optimality and security of control instructions under complex working conditions.
[0007] In conjunction with the first aspect mentioned above, in one possible implementation, the process of updating the multi-source influence factor association model based on multi-source state influence factor data includes: during operation, when an abnormal change in the target influence factor in the multi-source state influence factor data is detected, a local subgraph update is triggered, and the association strength and confidence of the association edges between the neighboring nodes corresponding to the target influence factor are recalculated. According to a preset period, batch updates are performed on the historically accumulated multi-source state influence factor data, and the association strength and confidence of all association edges in the multi-source influence factor association model are recalculated.
[0008] In conjunction with the first aspect mentioned above, in one possible implementation, when a node state adjustment instruction requires disconnecting the physical interconnection between distributed monitoring nodes, while a resource mutual aid scheduling instruction requires maintaining the physical interconnection between distributed monitoring nodes, the process of performing hierarchical arbitration based on the current environmental risk level and generating a collaborative control instruction includes: when the current environmental risk level is the first risk level, generating a collaborative control instruction containing the node state adjustment instruction to disconnect the physical interconnection between distributed monitoring nodes; when the current environmental risk level is the second risk level, generating a collaborative control instruction containing compromise control to maintain the physical interconnection between distributed monitoring nodes, reducing the transmission power of the interconnection lines and increasing transient voltage suppression protection; when the current environmental risk level is the third risk level, generating a collaborative control instruction containing the resource mutual aid scheduling instruction to maintain the full-power physical interconnection between distributed monitoring nodes. The risk level of the first risk level is higher than that of the second risk level, and the risk level of the second risk level is higher than that of the third risk level.
[0009] In conjunction with the first aspect mentioned above, one possible implementation involves revising the multi-source influence factor correlation model based on execution feedback data and environmental validation data. This process includes: defining a confidence interval for the predicted state quantity output by the energy prediction model; and triggering an online correction process when the actual measured value falls outside the confidence interval. In the online correction process, the weight parameters of the energy prediction model are updated using execution feedback data. The energy prediction model is a time-series prediction model trained based on historical operational data to predict the future energy state of distributed monitoring nodes. Furthermore, the conditional probabilities of the corresponding correlation relationships in the multi-source influence factor correlation model are updated using environmental validation data to improve the accuracy of subsequent node state projections.
[0010] In conjunction with the first aspect mentioned above, in one possible implementation, after recalculating the association strength and confidence of all association edges in the multi-source impact factor association model, the method further includes: performing a chi-square test on the association rules corresponding to the association edges in the multi-source impact factor association model, and removing association edges that fail the significance test; performing time consistency verification by determining the time lag by calculating the peak position of the cross-correlation function of the time series of two impact factors, requiring that the antecedent of the association rule occurs earlier than the consequent; if the time lag is negative, it is judged as a spurious association and removed; and performing continuous effectiveness monitoring. Within a preset sliding time window, if the number of windows where the confidence of an association edge is continuously lower than a preset confidence threshold reaches a first preset number, the association edge is marked as pending verification. If a second preset number is reached, the association edge is deleted from the multi-source impact factor association model, wherein the second preset number is greater than the first preset number.
[0011] In conjunction with the first aspect mentioned above, in one possible implementation, the generation process of node state adjustment instructions includes: calculating the importance score of the distributed monitoring node based on its topological centrality, task criticality, and real-time requirements. Combining the current environmental risk level, the importance level corresponding to the importance score, and the current energy state of the distributed monitoring node, a corresponding protection mode instruction is generated through a three-dimensional decision matrix. The protection mode instruction is the specific execution mode of the node state adjustment instruction. Here, topological centrality represents the connectivity of the distributed monitoring node in the network topology, task criticality represents the preset importance of the distributed monitoring node's location, real-time requirements represent whether the distributed monitoring node is currently performing a monitoring task, and the importance level is determined based on the comparison between the importance score and a preset score threshold. The mapping rule of the three-dimensional decision matrix is as follows: when the current environmental risk level is the highest risk level, an emergency protection mode instruction is generated. When the current environmental risk level is an intermediate risk level, an early warning protection mode instruction or an information push instruction is generated based on the combination of the importance level and the current energy state.
[0012] In conjunction with the first aspect mentioned above, in one possible implementation, the process of updating the weight parameters of the energy prediction model using execution feedback data includes: employing an elastic weight integration algorithm to update the weight parameters of the energy prediction model to mitigate catastrophic forgetting during incremental learning; mixing newly collected execution feedback data with the historical training data of the energy prediction model according to a preset ratio to update the parameters of the energy prediction model; and performing Markov chain Monte Carlo process verification on the parameter-updated energy prediction model. In the Markov chain Monte Carlo process verification, when the convergence diagnostic index of the parameter posterior distribution reaches a preset convergence threshold, the verification is deemed successful, and the updated energy prediction model is deployed.
[0013] In conjunction with the first aspect mentioned above, one possible implementation further includes: acquiring the leakage current and varistor voltage of the intelligent surge protector configured in the distributed monitoring node, and predicting the remaining lifespan of the intelligent surge protector based on an exponential decay model; triggering a protector replacement reminder when the remaining lifespan is lower than a preset lifespan threshold. When the distributed monitoring node enters emergency protection mode, it controls the electronic circuit breaker to disconnect the interconnection port and triggers the anti-reverse current diode for reverse current isolation. The emergency protection mode is the protection mode corresponding to the node state adjustment command executed when the current environmental risk level is the first risk level.
[0014] In conjunction with the first aspect mentioned above, in one possible implementation, the generation of resource mutual aid scheduling instructions is based on energy evolution prediction results. The process of obtaining energy evolution prediction results includes: taking historical solar intensity, wind speed, temperature, battery state of charge, load power consumption, and future weather forecast data as input, and outputting the time-by-time photovoltaic power generation prediction value and battery state of charge prediction value for each distributed monitoring node within a preset future time period through a long short-term memory network. The photovoltaic power generation prediction value and battery state of charge prediction value constitute the energy evolution prediction result. The long short-term memory network is configured using a hybrid training architecture combining offline training and online incremental learning. Offline training completes the initial parameter configuration based on historical data, while online incremental learning fine-tunes the network weights using newly collected data at preset intervals.
[0015] In conjunction with the first aspect mentioned above, one possible implementation further includes: when a communication link interruption with the central management platform is detected, the distributed monitoring node executes localized protection actions based on the most recently received decision instructions stored locally and the data collected in real time by local sensors. These localized protection actions include: determining the lightning risk level based on local atmospheric electric field strength measurements, and automatically switching to the corresponding protection mode when the lightning risk level exceeds a preset risk threshold.
[0016] This application provides a distributed monitoring node security management method based on an outdoor monitoring system. By constructing a multi-source influencing factor association model that characterizes the correlation, strength, and confidence of influencing factor nodes, discrete meteorological, environmental, and equipment status data can be transformed into a structured causal knowledge graph. This allows node status inference to move beyond relying on a single threshold judgment and instead rely on a comprehensive assessment based on the joint probability distribution of cross-domain factors, effectively improving the foresight of environmental risk level determination. Furthermore, addressing the inherent contradiction between node status adjustment commands and resource mutual assistance scheduling commands in the physical interconnection topology, a hierarchical arbitration mechanism based on the current environmental risk level is introduced. This allows for a dynamic balance of the priorities of security isolation and energy coordination at different risk levels, avoiding secondary failures or energy waste caused by rigid strategies. Finally, by acquiring the execution feedback data and environmental verification data after the collaborative control command is executed, and using them as correction signals to feed back to the multi-source influence factor correlation model, a complete data closed loop from perception, inference, arbitration, execution to correction can be formed. This allows the model to continuously calibrate the correlation strength and conditional probability based on the actual physical response, effectively overcoming the decision failure problem caused by environmental drift in the long-term operation of static models, and realizing the adaptive evolution and global optimization of the security management strategy of distributed monitoring nodes.
[0017] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description
[0018] Figure 1 A flowchart illustrating the security management method for distributed monitoring nodes based on an outdoor monitoring system provided in this application embodiment; Figure 2 The overall architecture and module interaction diagram of the distributed monitoring node security management method based on an outdoor monitoring system provided in the embodiments of this application; Figure 3 This is a flowchart of the dual-mode update and triple verification process of the multi-source influence factor correlation model in the distributed monitoring node security management method based on an outdoor monitoring system provided in the embodiments of this application. Figure 4 This application provides a method for security management of distributed monitoring nodes in an outdoor monitoring system, which includes a node importance calculation and a three-dimensional decision matrix mapping diagram. Figure 5 This is a diagram illustrating the hierarchical arbitration execution path in the distributed monitoring node security management method based on an outdoor monitoring system provided in this application embodiment. Figure 6 The flowchart of online closed-loop correction triggering and verification in the distributed monitoring node security management method based on an outdoor monitoring system provided in the embodiments of this application; Figure 7 This is a diagram of the hybrid training architecture of the energy evolution prediction model in the distributed monitoring node security management method based on an outdoor monitoring system provided in the embodiments of this application. Figure 8 Diagram of hardware collaborative protection and local autonomy mechanism in the distributed monitoring node security management method based on outdoor monitoring system provided in the embodiments of this application. Detailed Implementation
[0019] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.
[0020] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.
[0021] Example 1: like Figure 1-2 As shown, this embodiment provides a method for the security management of distributed monitoring nodes based on an outdoor monitoring system. This method constructs a complete data flow backbone from environmental perception, risk simulation, conflict arbitration to closed-loop correction, aiming to solve the topology conflicts and model drift problems of security isolation and resource coordination under complex operating conditions. The method of this embodiment includes the following steps: Step S100: Obtain multi-source state influence factor data of distributed monitoring nodes, and update the multi-source influence factor association model based on the multi-source state influence factor data. The multi-source influence factor association model is a directed weighted graph model that characterizes the association relationship, association strength and confidence between influence factor nodes.
[0022] Among them, multi-source state influence factor data refers to a heterogeneous data set that comes from distributed monitoring nodes and their surrounding environment and can characterize the system's operating status or external environmental risks, including but not limited to atmospheric electric field intensity, radar echo, lightning location information, battery state of charge, load power consumption, equipment temperature and vibration amplitude, etc.
[0023] The multi-source influence factor association model uses a directed weighted graph data structure. The vertices in the graph represent various influence factors, the directed edges represent the causal or spatiotemporal relationships between factors, the weights of the edges represent the strength of the association, and the attributes of the edges include the confidence level of the association with the current data support.
[0024] In some implementations, the system collects data on the aforementioned multi-source state influencing factors in real time through sensor units deployed at each monitoring node and access to external meteorological data interfaces. Then, using pre-defined correlation mining algorithms or probabilistic inference models, it analyzes the synchronous change patterns or causal dependencies of each factor over time, dynamically constructing or updating a directed weighted graph.
[0025] For example, when it is detected that the increase in atmospheric electric field intensity and subsequent lightning strike events co-occur frequently in historical data and satisfy the temporal sequence constraint, a directed edge from the atmospheric electric field node to the lightning strike risk node can be established in the graph, and the correlation strength and confidence of the edge can be calculated based on statistical significance.
[0026] The model can be updated either by a full recalculation or by incremental updates of a local subgraph for changes in specific factors, as long as it can reflect the true correlation between factors at the current moment.
[0027] It should be noted that this step transforms discrete and heterogeneous sensor data into a structured directed weighted graph model, thereby enabling a shift from single threshold judgment to cross-domain correlation cognition. This allows the system to no longer view the numerical anomaly of a single sensor in isolation, but to comprehensively analyze the joint evolution trend of multiple factors based on the correlation paths in the graph. This provides a data foundation with causal explanatory power for subsequent risk extrapolation and improves the dimensionality and accuracy of environmental perception.
[0028] Step S200: Perform node status simulation through multi-source influence factor correlation model, determine the current environmental risk level, and generate node status adjustment instructions and resource mutual assistance scheduling instructions.
[0029] Among them, node state inference refers to the process of predicting the risk state or energy state that distributed monitoring nodes may face in a future preset period based on the correlation structure and parameters in the current multi-source influence factor correlation model and combined with real-time input influence factor data.
[0030] The current environmental risk level is a quantitative classification of the simulation results.
[0031] Node state adjustment commands are used to change the node's own operating mode to deal with security risks, such as disconnecting physical interconnects or reducing power consumption.
[0032] Resource sharing and dispatching commands are used to regulate energy transfer between nodes to ensure power supply continuity, such as closing interconnection switches and adjusting transmission power.
[0033] In some implementations, the system maps the real-time collected impact factor data to the corresponding nodes of the multi-source impact factor association model. Through the probability propagation algorithm or path reasoning mechanism on the graph, it calculates the probability that the target node (such as a key monitoring point) will suffer from risks such as lightning strikes, overvoltage or undervoltage in the future period, and determines the current environmental risk level based on the preset risk classification threshold.
[0034] Based on the energy evolution prediction results, the remaining endurance and mutual assistance needs of each node are assessed. If the simulation results show that the lightning risk in a certain area is high, a node state adjustment instruction is generated requiring the nodes in that area to disconnect their interconnection lines; if the simulation results show that a node's power is critical and adjacent nodes have surplus power, a resource mutual assistance scheduling instruction is generated requiring the maintenance or establishment of interconnection lines.
[0035] It should be noted that this step can utilize the correlation model as an inference engine, so that the determination of risk level is no longer a simple matching of static rules, but a real-time calculation based on dynamic correlation knowledge. This allows it to generate two types of instructions in parallel, serving security protection and energy security respectively. This provides decision input for subsequent handling of potential conflicts between the two at the physical topology level, ensuring the comprehensiveness and foresight of the control strategy.
[0036] Step S300: When the node status adjustment instruction requires disconnecting the physical interconnection between distributed monitoring nodes, while the resource mutual assistance scheduling instruction requires maintaining the physical interconnection between distributed monitoring nodes, a graded arbitration is performed based on the current environmental risk level to generate a collaborative control instruction, which is then sent to the corresponding distributed monitoring node.
[0037] Among them, graded arbitration refers to the logical process of dynamically weighing the execution weights of the two types of instructions according to a preset priority strategy when a mutual contradiction is detected between security isolation requirements and energy coordination requirements in the physical connection state, based on the severity of the current environmental risk level, and generating the final executable coordinated control instructions.
[0038] The coordinated control command is the sole execution command after arbitration, including executing disconnection, executing hold, or executing a compromise control mode that balances safety and power supply.
[0039] In some implementations, the system has a built-in hierarchical arbitration rule base. When a conflict is detected between disconnect and hold commands for the same physical interconnect port, the current environmental risk level determined in step S200 can be read first. If the risk level is the highest, then safety takes priority. The arbitration result will enforce the node state adjustment command and generate a cooperative control command to disconnect the physical interconnection in order to prevent the surge from spreading along the line. If the risk level is intermediate, a compromise arbitration is executed to generate a coordinated control command that includes reducing transmission power and enabling transient voltage suppression protection, thereby limiting the risk of surge intrusion while maintaining a minimum level of energy exchange. If the risk level is the lowest, then energy supply takes priority, and the arbitration result will enforce the resource mutual assistance scheduling instruction to maintain full-power physical interconnection.
[0040] The generated collaborative control commands can be sent to the execution controller of the target monitoring node via wired or wireless communication links.
[0041] It should be noted that by introducing a risk-level-based hierarchical arbitration mechanism, this step can avoid unnecessary service interruptions caused by power outage protection in traditional solutions, and also prevent the risk of equipment damage caused by blindly pursuing power supply continuity. Through a dynamic trade-off strategy, the system can automatically find the optimal balance point under different risk levels, resolve the inherent contradiction between security and efficiency in distributed systems, and improve the adaptability and robustness of the control methods.
[0042] Step S400: Obtain the execution feedback data and environment verification data after the distributed monitoring node executes the collaborative control command.
[0043] Among them, execution feedback data refers to the measured data of the distributed monitoring nodes on the execution status of the command and their own internal physical parameters after executing the collaborative control command, such as the actual opening and closing status of the circuit breaker, port voltage and current values, and battery SOC change rate.
[0044] Environmental verification data refers to the true external environmental data collected by local or nearby sensors during and after the execution of instructions, used to verify the correctness of instruction decisions, such as the actual location and intensity of lightning strikes, actual rainfall, and actual light intensity.
[0045] In some implementations, after receiving a coordinated control command and completing the corresponding action, the monitoring node can immediately collect its own electrical and equipment status and upload it as execution feedback data.
[0046] The system continuously receives environmental monitoring data from meteorological sensors or third-party data sources, using it as environmental verification data.
[0047] By precisely aligning the timestamps of these two types of data with the issuance times of the collaborative control commands, a complete data sample can be formed, encompassing decision-making, execution, and verification. The execution feedback data reflects whether the commands were executed correctly and the immediate physical response after execution, while the environmental verification data reflects whether the external assumptions upon which the decisions were based are consistent with the real world.
[0048] It should be noted that this step establishes a feedback channel from the physical world to the digital decision-making space. By distinguishing between internal execution feedback and external environment verification, the system can not only monitor the effectiveness of instruction execution but also obtain an objective benchmark for evaluating the accuracy of the decision-making model. This provides an indispensable data foundation for subsequent model correction, thus overcoming the problem that traditional open-loop control systems cannot perceive the consequences of decisions.
[0049] Step S500: Based on the execution feedback data and environmental verification data, revise the multi-source influence factor correlation model. The revised multi-source influence factor correlation model is used for the node state deduction of the next cycle.
[0050] Among them, "correction" refers to the process of adjusting and optimizing the correlation strength, confidence level, or structural parameters in the multi-source influence factor correlation model by using newly acquired execution feedback data and environmental verification data as training samples or constraints.
[0051] The next cycle refers to any subsequent extrapolation period immediately following the execution of the current decision.
[0052] In some implementations, the system can compare the environmental verification data with the deduction results in step S200: If actual environmental data shows that the projection results are accurate (e.g., a high-risk prediction is followed by an actual lightning strike), then the confidence level of the relevant associated edges is increased. If the projection results do not match the actual situation (e.g., high risk is predicted but no lightning strike occurs, or low risk is predicted but equipment is damaged), the confidence level of the relevant related edges will be reduced or the correlation strength will be adjusted according to the degree of deviation.
[0053] Execution feedback data is used to verify the accuracy of energy prediction or equipment response models, and the model parameters are fine-tuned accordingly. The corrected model can then immediately replace the old model for the next round of state simulation and decision generation.
[0054] The correction can be made by adjusting the conditional probability based on Bayesian updates or by optimizing the parameters based on gradient descent, as long as the model parameters can adaptively evolve with the accumulation of data.
[0055] It should be noted that this step, by feeding back the actual execution results and environmental truth values to the model, enables the multi-source influencing factor correlation model to have the ability to learn online and self-correct, thus forming a cognitive evolution closed loop of the safety management method. This not only solves the industry problem of static models gradually becoming ineffective due to environmental drift, but also enables the system to continuously accumulate experience and knowledge as the running time increases, thereby achieving continuous optimization and adaptive evolution of management strategies and improving the long-term survival capability of distributed monitoring nodes in complex and ever-changing outdoor environments.
[0056] Based on the above technical solution, this embodiment constructs a directed weighted graph model representing cross-domain associations. When topological conflicts occur between security isolation and resource collaboration, it performs hierarchical arbitration based on risk levels. At the same time, it uses execution feedback and environmental verification data to perform closed-loop correction of the model, forming a complete data flow backbone from perception, inference, arbitration, execution to correction. This not only enables the system to improve from passive threshold response to proactive association inference, but also resolves inherent contradictions at the physical level through a dynamic arbitration mechanism. Furthermore, the online correction mechanism enables the system to have the ability to adaptively evolve, thereby maximizing energy utilization efficiency and business continuity while ensuring equipment safety.
[0057] Example 2: like Figure 3 As shown in Example 1, this embodiment further details the dynamic update mechanism and association rule verification mechanism of the multi-source influencing factor association model. During the long-term operation of a distributed outdoor monitoring system, the relationship between environmental factors and equipment status will dynamically evolve with seasonal changes, equipment aging, and adjustments to maintenance strategies. To ensure the continuous accuracy of model inference, this application constructs a dual-mode update strategy that includes local real-time updates and batch periodic updates, and introduces a rigorous triple verification mechanism to eliminate false associations, thereby preventing erroneous associations from leading to misjudgments in decision-making.
[0058] Step A100: During operation, when an abnormal change is detected in the target influence factor in the multi-source state influence factor data, a local subgraph update is triggered to recalculate the association strength and confidence of the association edges between the neighboring nodes corresponding to the target influence factor.
[0059] Among them, abnormal changes in target impact factors refer to the real-time observed values of one or a group of impact factors deviating from their historical steady-state distribution or preset safety boundary, and the degree of deviation is sufficient to characterize that the system has entered a new operating mode or risk state.
[0060] Neighborhood nodes refer to the set of nodes in a multi-source influence factor association model that have direct directed edges connected to the target influence factor node, including predecessor nodes and successor nodes.
[0061] Local subgraph updates are an incremental model maintenance strategy that allows parameter recalculation only for local topologies directly affected by abnormal events.
[0062] In some implementations, the system monitors the data streams from various sensors in real time. When it detects a step change in a certain influencing factor (such as atmospheric electric field strength) within a short period of time (e.g., an increase from 0.8 kV / m to 2.5 kV / m within 5 minutes), or when its statistical characteristics (such as variance and skewness) exceed a preset steady-state threshold, the system can mark this factor as a target influencing factor. The system then automatically identifies all neighborhood nodes of this target influencing factor in the directed weighted graph, thereby extracting a local subgraph composed of these nodes.
[0063] For each associated edge within this local subgraph, the algorithm for calculating association strength and confidence is re-executed using measured data from the most recent time window. For example, a Bayesian update method is employed, using newly observed data as likelihood evidence to update the parameters in the conditional probability table for that edge. Since the number of neighboring nodes is typically much smaller than the total number of nodes in the entire network, the computational complexity of this local update process is reduced, thus achieving a real-time response at the millisecond or second level.
[0064] It should be noted that this event-triggered local update mechanism can effectively solve the problem of delayed response of traditional full updates when faced with sudden environmental changes. By focusing computing resources on the local areas where changes occur, the system can quickly capture and adapt to new correlation patterns while ensuring the overall stability of the model. This ensures that in emergency situations such as approaching thunderstorms, the model can reflect the latest risk transmission paths in real time, providing timely assurance for subsequent graded arbitration.
[0065] Step A200: According to the preset cycle, perform batch updates on the historically accumulated multi-source state influence factor data, and recalculate the association strength and confidence of all association edges in the multi-source influence factor association model.
[0066] The preset period refers to a time interval set in advance based on the data accumulation rate and model drift characteristics of the business scenario, in order to balance model accuracy and computing resource consumption.
[0067] Batch updates are a global model calibration strategy that can utilize long-term historical data to eliminate short-term noise interference that may be introduced by local updates, and to uncover potential long-range or weak correlations from a macro perspective.
[0068] In some implementations, the system sets 2:00 AM daily as the execution window for batch updates, which is usually a low-peak period for monitoring business and when data integrity is relatively high.
[0069] When an update is triggered, the system extracts all the multi-source state influence factor data accumulated over the past 24 hours and combines it with a longer-term historical dataset (such as the past 30 days) to rerun the association rule mining algorithm (such as the Apriori algorithm) and the probabilistic inference model (such as Bayesian network learning).
[0070] The above process will perform a comprehensive parameter recalculation and structural evaluation of all existing and potential new related edges in the model. For each related edge, based on the complete historical dataset, its conditional probability will be recalculated as the association strength, and its confidence level will be updated according to the statistical test results. At the same time, the batch update will also scan for node pairs that were not touched in the local update to discover new associations that are not apparent in the short term due to data sparsity but are statistically significant over a long period.
[0071] It should be noted that batch updates and partial updates can form a complementary dual-mode driven system. Partial updates ensure the model's agile response to sudden events, while batch updates act as a benchmark calibrator, preventing model parameters from gradually deviating from true physical laws due to frequent local corrections. This dual-mode strategy enables multi-source influencing factor correlation models to possess both real-time capabilities to address instantaneous risks and long-term robustness and global consistency.
[0072] Step A300: After recalculating the association strength and confidence of all association edges in the multi-source influence factor association model, a triple verification mechanism is performed on the association edges to eliminate spurious associations and ensure the accuracy of the model.
[0073] The triple verification mechanism refers to a filtering process that, after the model is updated, sequentially reviews the quality of related edges from three dimensions: statistical significance, temporal causality, and persistent validity.
[0074] False associations refer to relationships that appear to co-occur at the data mining level but lack a real causal relationship or are merely coincidental. This mechanism effectively prevents data correlation from being mistaken for physical causality, thus avoiding erroneous decisions.
[0075] In some implementations, the first layer of verification is a statistical significance test: the system performs a chi-square test on the association rule corresponding to each associated edge, calculating the significance of the difference between the observed frequency and the expected frequency. Only when the p-value is less than the preset significance level (e.g., 0.05) is the association considered statistically significant; otherwise, the edge is directly determined to be a spurious association generated by random noise and is removed.
[0076] The second verification is the time consistency verification: considering that causal relationships in the physical world necessarily follow a chronological order, the system can determine the time lag τ corresponding to the peak position by calculating the cross-correlation function of the time series of two influencing factors. If the antecedent of the association rule occurs later than the consequent (i.e., τ is negative), it violates the causal law, and regardless of its statistical significance, it is judged as a spurious association and eliminated.
[0077] The third layer of verification is continuous effectiveness monitoring: the system sets a preset sliding time window (e.g., 30 days) to continuously track the confidence changes of each associated edge. If the confidence of the edge is lower than the preset confidence threshold (e.g., 60%) for a consecutive first preset number of windows (e.g., 3), it is marked as a state to be verified, and its weight contribution in decision inference is suspended; if this state continues to reach the second preset number of windows (e.g., 6), it is confirmed that the association has failed and is permanently deleted from the model.
[0078] During its specific operation, a second preset quantity can be set to be greater than the first preset quantity, so that the two-level buffer architecture can effectively avoid accidental deletion caused by short-term data fluctuations.
[0079] It should be noted that the triple verification mechanism can construct a three-dimensional protection network for model accuracy from different dimensions. Its chi-square test eliminates accidental co-occurrence, its time consistency verification eliminates logical fallacies such as reversing cause and effect, and its sliding window monitoring eliminates outdated and invalid historical experience. This combined strategy effectively ensures that every relational edge remaining in the model has a statistical basis, a clear physical causal direction, and current real-world validity. This significantly reduces the risk of false positives and false negatives caused by model "illusions" or "outdated memories," thereby improving the credibility of security control decisions.
[0080] Based on the above technical solution, this embodiment constructs a dual-mode update strategy combining local real-time and batch periodic updates, supplemented by a triple verification mechanism of statistics, time series, and persistence. This enables the multi-source influencing factor association model to achieve adaptive evolution and quality control in dynamic environments. This mechanism not only ensures the model can quickly respond to sudden operating conditions but also effectively eliminates spurious and invalid associations generated during data mining, keeping the model in a high-confidence state that reflects real physical laws. This provides reliable support for the precise and secure management of distributed monitoring nodes.
[0081] Example 3: like Figure 4-5 As shown, this embodiment, based on Embodiment 1, further details the generation basis of node state adjustment commands and the specific execution strategy of hierarchical arbitration. In a distributed outdoor monitoring system, the conflict between security isolation and resource coordination is not a simple binary opposition, but requires a refined trade-off based on the real-time value of nodes and environmental risks. This embodiment transforms abstract arbitration logic into executable physical control parameters by dynamically calculating node importance and constructing a three-dimensional decision matrix, ensuring the optimality of collaborative control commands under complex operating conditions.
[0082] Step B100: Calculate the importance score of the distributed monitoring nodes based on their topological centrality, task criticality, and real-time requirements.
[0083] Among them, importance score It is a comprehensive indicator used to quantitatively represent the contribution of distributed monitoring nodes to the overall function of the system at the current moment. Its value is dynamically updated as the network topology changes, business load fluctuates and preset attributes change.
[0084] Topological centrality Characterize the degree of connectivity of a distributed monitoring node in the network topology. Specifically, it can be calculated using degree centrality or betweenness centrality algorithms, reflecting the physical status of the node as an energy or data relay point. Mission criticality The preset importance of the location of the distributed monitoring node is determined by the business attributes at the time of deployment. For example, key locations such as border crossings and fire lookout towers are given a higher baseline value. Real-time demand This indicates whether the distributed monitoring node is currently performing a monitoring task. The value is 1 when it is performing high-priority tasks such as target tracking, event recording, or AI inference, and 0 when it is idle or only in a standby polling state.
[0085] In some implementations, the system uses a weighted summation formula to calculate the importance score:
[0086] in, , , These are the weighting coefficients, configured by default. =0.4、 =0.4、 =0.2, which can be adjusted by maintenance personnel according to the actual scenario.
[0087] The calculated importance score The nodes are compared with preset score thresholds to classify them into different importance levels. For example, a score greater than or equal to 0.7 is considered a Level 1 importance node, a score between 0.4 and 0.7 is considered a Level 2 importance node, and a score less than 0.4 is considered a Level 3 importance node.
[0088] The above calculation process needs to be executed periodically at a preset frequency (such as every 10 minutes), or triggered immediately when events such as topology changes or task start / stop are detected.
[0089] It should be noted that this dynamic calculation mechanism can overcome the resource misallocation problem caused by the static fixation of node priorities in traditional solutions. It allows the system to take into account real-time business needs and network topology, enabling it to identify temporarily important nodes that, while not having a high preset priority, are currently undertaking critical relay or forensic tasks. This allows the system to assign them appropriate protection weights in subsequent arbitration decisions, avoiding critical business interruptions or network disruptions caused by mechanically applying preset priorities.
[0090] Step B200: Combining the current environmental risk level, the importance level of the importance score, and the current energy state of the distributed monitoring nodes, a corresponding protection mode instruction is generated through a three-dimensional decision matrix. The protection mode instruction is the specific execution mode of the node state adjustment instruction.
[0091] The three-dimensional decision matrix is a predefined discrete mapping lookup table. Its three dimensions are the current environmental risk level, importance level, and current energy state, respectively, and the output is a specific protection mode command. Furthermore, this matrix can transform continuously changing multi-dimensional situational awareness data into deterministic control strategies, serving as a decision translator connecting upper-level deduction and lower-level execution.
[0092] The protection mode instructions include, but are not limited to, emergency protection mode, early warning protection mode, information push mode and normal operation mode, and each mode corresponds to a set of explicit hardware control parameters and software behavior constraints.
[0093] In some implementations, the mapping rules of the three-dimensional decision matrix follow the principle of risk-driven approach while also considering efficiency: When the current environmental risk level is the highest risk level (i.e., the first risk level), regardless of the node's importance level and energy status, an emergency protection mode command is generated to ensure that the physical safety of the equipment is the absolute priority. When the current environmental risk level is intermediate risk level (i.e., second risk level), differentiated mapping can be performed based on the combination of importance level and current energy status: if the node is of first-level importance and has sufficient energy, the mapping generates an early warning protection mode instruction to enhance protection while maintaining basic functions; if the node is of second- or third-level importance, or has a tight energy status, the mapping generates an information push instruction or a downgrade operation instruction to avoid excessive consumption of resources. When the current environmental risk level is the lowest risk level (i.e., the third risk level), active protection is usually not triggered, and full-power interconnection is maintained to ensure the efficiency of resource sharing.
[0094] It should be noted that by introducing energy state as a third dimension, this step effectively addresses the coarseness of decisions based on a single risk dimension. This allows the decision matrix to prioritize nodes with sufficient power (e.g., activating additional monitoring) under moderate risk conditions, while prioritizing the preservation of remaining battery life for nodes with critical power levels. This multi-dimensional coupled mapping mechanism ensures that the generated protection mode commands are both risk-specific and consider the node's own survivability, providing refined decision input for subsequent tiered arbitration.
[0095] Step B300: When the node status adjustment instruction requires disconnecting the physical interconnection between distributed monitoring nodes, while the resource mutual aid scheduling instruction requires maintaining the physical interconnection between distributed monitoring nodes, a graded arbitration is performed based on the current environmental risk level to generate a collaborative control instruction.
[0096] Among them, graded arbitration refers to the logical process of dynamically selecting the execution path according to a preset priority strategy when a physical topological conflict is detected between the security isolation requirement and the energy coordination requirement.
[0097] The collaborative control instruction is the final execution instruction after arbitration, and its content strictly corresponds to the physical actions specified in the protection mode instruction generated in step B200.
[0098] The risk level of the first risk level is higher than that of the second risk level, and the risk level of the second risk level is higher than that of the third risk level. The higher the risk level, the higher the priority of security isolation.
[0099] In some implementations, when the current environmental risk level is the highest risk level, the system determines that the safety risk outweighs the power supply demand, generates a collaborative control command that includes node status adjustment instructions, and forcibly executes the action of disconnecting the physical interconnection between distributed monitoring nodes. Specifically, this may include controlling the electronic circuit breaker to trip and the anti-reverse current diode to reverse cut off, so as to completely cut off the surge propagation path. When the current environmental risk level is Level 2, the system determines that safety and power supply need to be dynamically balanced. It generates coordinated control commands that include trade-off controls to maintain the physical interconnection between distributed monitoring nodes, while simultaneously reducing the transmission power of interconnect lines and adding transient voltage suppression protection. The reduction in transmission power can be achieved by adjusting the PWM duty cycle or current limiting threshold of the DC microgrid controller, limiting the current in the interconnect lines to below 50% of the rated value, thereby reducing the energy carrier scale during surge intrusion. The addition of transient voltage suppression protection can be achieved by activating the TVS diode array connected in parallel at the interconnect ports, clamping the line overvoltage within a safe level within nanoseconds. When the current environmental risk level is level three, the system prioritizes energy supply demand and generates collaborative control instructions that include resource mutual assistance scheduling instructions, thereby maintaining full-power physical interconnection between distributed monitoring nodes to maximize energy complementarity efficiency.
[0100] It should be noted that this step, through a combination of current limiting and clamping, constructs a controlled energy transmission channel at the physical level. Reducing power decreases the potential surge energy injection, while TVS protection absorbs residual transient overvoltage. The synergistic effect of these two measures allows nodes to maintain a minimum level of energy exchange and data transmission capacity while withstanding a certain level of lightning-induced risk. This avoids unnecessary service interruptions caused by over-protection and prevents equipment damage caused by blind exchange, maximizing performance within the safety boundary and thus improving the resilience and availability of the distributed monitoring system under complex weather conditions.
[0101] Based on the above technical solution, this embodiment dynamically calculates node importance scores and constructs a three-dimensional decision matrix, which accurately maps multi-dimensional situational awareness data into differentiated protection mode commands. This allows for the execution of a tiered arbitration strategy, including compromise controls, based on risk levels when conflicts arise between security isolation and resource coordination. This mechanism not only frees node importance assessment from the limitations of static configuration and enables real-time response to service and topology changes, but also effectively balances safety and power supply under medium-risk conditions through a combination of physical-level power limiting and voltage suppression. This solves the problem of secondary failures or energy waste caused by traditional binary decision-making, providing solid decision support for the global optimal management of distributed monitoring nodes.
[0102] Example 4: like Figure 6 As shown, this embodiment, based on step S500 of embodiment 1, further details the triggering mechanism of closed-loop correction and the model stability guarantee during online learning. In the long-term operation of a distributed outdoor monitoring system, the evolution of environmental patterns is non-stationary. If the model continues to perform undifferentiated online learning, it is highly susceptible to "catastrophic forgetting" due to short-term noise interference or sample distribution shifts; that is, the model forgets historically accumulated patterns while learning new knowledge. Therefore, this application constructs a precise triggering mechanism based on cognitive thresholds and introduces elastic weight integration and Bayesian verification strategies, forming a robust sub-loop from triggering to correction and then to verification, ensuring that the model maintains stable performance throughout the adaptive evolution process.
[0103] Step C100: Define the confidence interval of the predicted state quantity output by the energy prediction model. When the actual measured value falls outside the confidence interval, trigger the online correction process.
[0104] The confidence interval refers to the probability coverage range given by the energy prediction model under the current parameter state for the predicted state quantity (such as battery state of charge, photovoltaic power generation) within a preset future time period. It characterizes the model's cognitive uncertainty about its own prediction results. The interval can be an adaptive boundary that adapts to the volatility of the input data and the dynamic changes of the model's internal state.
[0105] The online correction process refers to the calculation process of incrementally updating model parameters using real-time data streams during normal system operation without interrupting business services.
[0106] In some implementations, the system constructs dynamic confidence intervals for each output variable of the energy prediction model. ,in To predict the lower quantile of the distribution, The upper quantile is the range between the two values, which is considered a reasonable prediction range by the model.
[0107] When the actual measured value collected by the sensor If the value falls outside this range, it indicates that the current environmental conditions have exceeded the model's existing cognitive scope, or that the model has experienced significant performance degradation. In this case, the system uses the Heaviside step function... Construct trigger criteria In the formula, This is the trigger criterion value. When If the prediction deviation is determined to exceed the tolerance limit, the online correction process is immediately activated.
[0108] The aforementioned triggering mechanism is an event-driven, on-demand learning strategy. For example, during periods of stable weather, the actual values remain within the confidence interval, and the system does not trigger corrections, effectively saving computing power and maintaining model stability. However, when sudden thunderstorms or drastic changes in cloud cover cause abrupt changes in illumination, the actual values quickly exceed the limits, and the system responds immediately and initiates learning.
[0109] It should be noted that this step, by transforming abstract model errors into mathematically precise confidence interval out-of-bounds events, achieves objectivity and automation in correction triggering. This avoids the subjectivity of manually setting fixed error thresholds and effectively prevents model oscillations caused by frequent invalid updates. Simultaneously, it endows the model with metacognitive capabilities, enabling the system to distinguish between normal environmental fluctuations and genuine cognitive blind spots, consuming resources only for learning when necessary. This ensures model timeliness while maximizing the system's operational reliability under steady-state conditions.
[0110] Step C200: In the online correction process, the weight parameters of the energy prediction model are updated using execution feedback data, and the conditional probabilities of the corresponding correlations in the multi-source influence factor correlation model are updated using environmental validation data.
[0111] Updating the weight parameters of the energy prediction model refers to adjusting the connection weights and bias terms within the Long Short-Term Memory network so that its output better matches the latest measured data.
[0112] The Elastic Weight Integration algorithm is a regularization method used to mitigate catastrophic forgetting in neural networks during continuous learning tasks. It effectively identifies parameters that are crucial to the historical task and imposes greater constraints on them during updates.
[0113] Conditional probability update refers to adjusting the strength of the causal or correlation relationships represented by the directed edges in the multi-source influence factor association model based on the latest environmental verification results, so that the graph knowledge evolves with real-world feedback.
[0114] In some implementations, for updating the energy prediction model, the system first calculates the Fisher information matrix to evaluate the contribution of each weight parameter to the historical prediction accuracy, and then constructs an elastic weight penalty term.
[0115] In the gradient descent optimization process, the loss function includes the current task loss brought about by the newly collected execution feedback data, as well as the penalty loss for parameters deviating from the old optimal value.
[0116] For parameters that are important to historical knowledge, the penalty coefficient is larger to limit their range of change; for parameters that are not sensitive, they are allowed to adjust freely to adapt to new data.
[0117] The system mixes newly collected execution feedback data with historical training data in a preset ratio (e.g., 1:9) to form a fine-tuning dataset, thus ensuring the injection of new knowledge while consolidating old memories by replaying historical data.
[0118] For updating the multi-source influencing factor association model, the system can use environmental validation data (such as actual lightning strike records and actual meteorological data) as new evidence nodes, input them into the Bayesian network, and use Bayesian update rules to recalculate the posterior probabilities of relevant association edges. If the actual validation results support the original inference, the confidence of that edge is increased; otherwise, the confidence is decreased or the association strength is adjusted.
[0119] It should be noted that this step, through the flexible weighting integration algorithm, can protect the model's historical memory, ensuring that the system does not forget the patterns of low temperature and low light in winter when learning new summer thunderstorm patterns. Simultaneously, its data mixing strategy provides double protection at the sample level. The synchronous correction of the correlation map ensures that the data used in decision-making simulations remain consistent with the numerical model for energy prediction. This dual-model co-evolution mechanism allows the safety management system to continuously accumulate experience over long operating cycles, effectively overcoming the technical problem of traditional static models being unable to adapt to long-term environmental drift.
[0120] Step C300: Perform Markov chain Monte Carlo process verification on the updated energy prediction model. In the Markov chain Monte Carlo process verification, when the convergence diagnostic index of the parameter posterior distribution reaches the preset convergence threshold, the verification is deemed successful, and the updated energy prediction model is deployed.
[0121] Among them, the Markov chain Monte Carlo process verification is a sampling-based Bayesian inference method used to evaluate whether the posterior distribution after the model parameter update has reached a stable state.
[0122] Convergence diagnostic metrics are statistical measures that quantify whether multiple parallel Markov chains are sufficiently mixed and whether the sampling represents the true posterior distribution. Commonly used metrics include the Gelman-Rubin statistic R-hat or the effective sample size ESS.
[0123] The preset convergence threshold is a quality threshold for determining whether a model update is safe and usable. Only when all indicators meet this threshold is the new model considered qualified to replace the old model.
[0124] In some implementations, after the parameter fine-tuning in step C200 is completed, a background verification process is started directly, running multiple independent Markov chains to sample the updated parameter space and calculate the R-hat value.
[0125] If the R-hat values of all key parameters are less than the preset threshold (e.g., 1.05) and the effective sample size exceeds the minimum requirement, it is determined that the posterior distribution of the parameters has converged, the model update is effective and stable, and the system will atomize and replace the new model in the online inference engine.
[0126] If the verification fails, it indicates that the update may have gotten stuck in a local minimum or been contaminated by abnormal data. The system will automatically roll back to the previous version of the model and mark the update as failed, waiting for the next trigger opportunity.
[0127] It is worth noting that introducing the MCMC verification mechanism elevates online learning from blind trial and error to reliable evolution, ensuring that each model iteration is a statistically validated and reliable improvement, rather than random fluctuations. Furthermore, combining the cognitive threshold triggering in step C100 with the anti-forgetting update in step C200 allows for the construction of a complete adaptive learning loop with self-correction and quality assurance capabilities. This enables the distributed monitoring node security management method to maintain both keen adaptability and adherence to safety standards when facing complex and ever-changing outdoor environments.
[0128] Based on the above technical solution, this embodiment can achieve on-demand initiation and precise response of model correction through a predefined cognitive threshold triggering mechanism. Simultaneously, by introducing an elastic weight integration algorithm and a new-old data mixing strategy, the catastrophic forgetting problem in online incremental learning is effectively mitigated, ensuring the continuity of historical knowledge. Furthermore, verification through a Markov chain Monte Carlo process sets a strict statistical safety threshold for model updates, effectively preventing model degradation due to data noise. Its overall closed-loop process from triggering to correction to verification also enables the security management system to continuously, stably, and reliably evolve in an open and dynamic environment, thereby effectively improving the management efficiency and security resilience of distributed monitoring nodes throughout their entire lifecycle.
[0129] Example 5: like Figure 7 As shown in Example 1, this embodiment further details the construction and training architecture of the energy evolution prediction model upon which the resource mutual aid scheduling command generation depends. The effectiveness of the resource mutual aid scheduling command highly depends on the accurate prediction of future energy supply and demand trends. The non-stationarity of the outdoor environment requires the prediction model to possess both a stable understanding of long-term physical laws and the ability to quickly adapt to short-term environmental drift. Therefore, this application constructs a hybrid training architecture based on a Long Short-Term Memory network, providing data support for scheduling decisions that is both forward-looking and adaptive through the synergy of offline and online learning.
[0130] Step D100: Taking historical light intensity, wind speed, temperature, battery state of charge, load power consumption, and future weather forecast data as input, and using a long short-term memory network, output the time-by-time photovoltaic power generation prediction value and battery state of charge prediction value for each distributed monitoring node within the preset time period. The photovoltaic power generation prediction value and battery state of charge prediction value constitute the energy evolution prediction result.
[0131] Among them, the energy evolution prediction result is a quantitative representation of the energy supply capacity and storage status of distributed monitoring nodes within a specific time window in the future, and is the direct data basis for generating resource mutual assistance scheduling instructions.
[0132] Long Short-Term Memory (LSTM) networks are a variant of recurrent neural networks suitable for processing time-series data. Their internal gating mechanisms can effectively capture the nonlinear dependencies and dynamic evolution patterns of multi-source heterogeneous data over long time spans.
[0133] Historical light intensity, wind speed, and temperature in the input features can directly determine the photoelectric conversion efficiency of photovoltaic modules. The state of charge of the battery reflects the current energy storage capacity, the load power consumption reflects the energy consumption rate, and future weather forecast data provides the constraints of the external environment.
[0134] In some implementations, the system first normalizes and aligns the multi-source input data to construct a standardized feature vector sequence, which is then input into a pre-configured long short-term memory network. Within this network, historical information is filtered through a forgetting gate, current observations are integrated through an input gate, and a predicted state is generated through an output gate. After temporal transmission and nonlinear transformation through multiple hidden units, the system finally decodes and outputs the hourly photovoltaic power generation prediction and battery state of charge prediction for each node within a preset future time period (e.g., the next 72 hours).
[0135] In other implementations, the granularity of time-period prediction can be flexibly configured according to scheduling needs. For example, it can be encrypted to the 15-minute level during periods of energy stress and relaxed to the hour level during periods of stability. In addition to Long Short-Term Memory networks, Transformer, GRU, or other deep learning models with time-series modeling capabilities can also be used to achieve the same function, as long as a mapping relationship between multidimensional inputs and future energy states can be established.
[0136] It should be noted that this step, by fusing multidimensional environmental and equipment status data into the time-series prediction model, achieves a shift from single-factor extrapolation to multi-factor coupled prediction. This enables the model to analyze the comprehensive energy response under complex conditions such as cloud cover causing a sudden drop in sunlight but increased wind speed improving heat dissipation, thereby improving the accuracy of prediction results under variable weather conditions. Simultaneously, its refined time-period output provides a precise time scale for resource sharing and scheduling, allowing scheduling commands to proactively avoid impending energy shortages rather than reacting passively after power depletion, thus enhancing the foresight and initiative of scheduling strategies.
[0137] Step D200: The Long Short-Term Memory network is configured with a hybrid training architecture that combines offline training and online incremental learning. The offline training completes the initial parameter configuration based on historical data.
[0138] Offline training refers to the complete training process of the long short-term memory network using a large-scale historical dataset before the system is officially deployed and run. This allows the model to learn the general physical laws and seasonal periodic characteristics of energy evolution in outdoor monitoring scenarios.
[0139] The initial parameter configuration includes setting hyperparameters such as the number of network layers, the number of hidden units, the learning rate, and the optimizer type, as well as the initial values of the weight matrix and bias vector obtained from training based on historical data.
[0140] In some implementations, historical data covering at least one complete annual cycle is acquired to ensure that the samples contain typical meteorological patterns and energy fluctuation characteristics of the four seasons, thus enabling offline training. During training, the system employs a sliding window sampling strategy to construct a training sample set, using mean squared error or mean absolute error as the loss function, and iteratively optimizes the network parameters through backpropagation until the model's performance on the validation set converges to a preset threshold.
[0141] In other implementations, strategies such as Dropout regularization and early stopping are introduced during training to prevent overfitting.
[0142] It should be noted that this step, through offline training, enables the hybrid architecture to perform effectively. This allows the model to gain a deep understanding of steady-state physical laws such as the law of conservation of energy and photoelectric conversion characteristics. It also allows subsequent online learning to be fine-tuned on a reliable benchmark instead of starting from scratch. This effectively solves the problem that pure online learning methods are prone to divergence or getting stuck in local optima when facing sparse events (such as extreme weather), ensuring the basic prediction accuracy and operational stability of the safety management system throughout its entire lifecycle.
[0143] Step D300: Online incremental learning fine-tunes network weights using newly collected data at preset intervals.
[0144] Online incremental learning refers to the automatic update process of model parameters during system operation at fixed time intervals, which can use the latest accumulated measured data to correct model biases caused by seasonal changes, equipment aging, or changes in the microenvironment.
[0145] The preset cycle is a time interval set in advance based on the data accumulation rate and model drift characteristics of the business scenario, such as once every 24 hours.
[0146] Fine-tuning network weights refers to making small adjustments to some or all connection weights while keeping the network structure unchanged, in order to adapt to new data distributions and avoid destroying the general knowledge learned during the offline training phase.
[0147] In some implementations, at the end of each preset period, the system automatically extracts the newly collected multi-source state influence factor data and the corresponding measured energy true values within that period to construct an incremental training sample set. This allows the current online model to be used as a starting point, and the sample set to be updated with gradient descent for several rounds using a relatively small learning rate.
[0148] In other implementations, in order to balance the weight of new and old knowledge, regularization techniques such as flexible weight integration can be combined during the fine-tuning process to limit the range of change of key parameters and prevent catastrophic forgetting.
[0149] The preset periodicity triggering mechanism in this step complements the cognitive threshold triggering mechanism described in Example 4: the preset periodicity fine-tuning is responsible for capturing slow, gradual environmental drift (such as the gradual shortening of daylight in autumn) and maintaining the baseline accuracy of the model; while the cognitive threshold triggering is responsible for dealing with sudden, drastic abnormal conditions (such as sudden changes in light caused by thunderstorms) and realizing emergency correction of the model. The two work together to form a multi-level, full-spectrum model adaptation system.
[0150] It should be noted that the online incremental learning mechanism, executed according to a preset cycle, enables the energy evolution prediction model to continuously absorb new knowledge and adapt to new environments as the operating time increases, making it a dynamic intelligent agent. This continuous self-calibration capability ensures that resource coordination and scheduling commands are always generated based on the latest energy situation, maintaining efficient energy coordination even in the event of equipment performance degradation or long-term changes in local climate, thereby extending the overall endurance and service life of the distributed monitoring system.
[0151] Based on the above technical solution, this embodiment constructs a long short-term memory network with clearly defined inputs and outputs. It employs a hybrid training architecture—offline training to lay the foundation for understanding physical laws, and online incremental learning for periodic fine-tuning to adapt to environmental drift—to provide high-precision and highly adaptive energy evolution prediction results for generating resource-assistance scheduling instructions. This architecture not only solves the problem of a single offline model being unable to cope with time-varying environments, but also, through complementary cooperation with an event-triggered correction mechanism, achieves dual adaptation of the model under both steady-state and transient conditions, improving the foresight and robustness of distributed monitoring node energy collaborative scheduling.
[0152] Example 6: like Figure 8 As shown, this embodiment, based on Embodiment 1, further details the hardware collaborative protection mechanism and local autonomous mechanism of the security management method. The aforementioned embodiments mainly focus on software decision-making and closed-loop correction based on a multi-source influencing factor correlation model. However, in actual outdoor monitoring scenarios, boundary conditions such as communication link interruptions, sensor failures, or actuator response delays frequently occur. To compensate for the execution uncertainty of pure software solutions under extreme conditions, this application constructs a multi-layered boundary security defense line, including intelligent surge protector lifespan prediction, hardware-level physical isolation, and autonomous decision-making in the event of network outages. This ensures that the security management strategy has a feasible physical execution basis and minimum protection capabilities under any operating condition.
[0153] Step E100: Obtain the leakage current and varistor voltage of the intelligent surge protector configured in the distributed monitoring node, and predict the remaining life of the intelligent surge protector based on the exponential decay model. When the remaining life is lower than the preset life threshold, trigger the protector replacement reminder.
[0154] Among them, intelligent surge protectors refer to intelligent protection devices that integrate microcontrollers and communication interfaces on the basis of traditional surge protection devices, and can collect their own electrical parameters and report status data in real time.
[0155] Leakage current refers to the tiny current flowing through the internal varistor or gas discharge tube of a surge protector under normal operating voltage. Its magnitude can directly reflect the aging degree and thermal stability of the internal insulating material of the device.
[0156] Varistor voltage refers to the critical voltage value at which a surge protector transitions from a high-resistance state to a low-resistance state. As the number of surge impacts increases, this voltage value will gradually drift or decrease, characterizing the degradation of the device's clamping capability.
[0157] The exponential decay model is a mathematical model that describes the nonlinear aging process and can be used to quantitatively characterize the degradation trend of surge protector performance parameters over time or with the accumulation of impact.
[0158] The preset lifespan threshold is a warning line for the remaining lifespan that is set in advance according to the equipment's safe operation specifications, such as 30 days or a specific performance percentage.
[0159] In some implementations, the system polls the real-time monitoring data of the intelligent surge protectors at each monitoring node via an RS485 bus or wireless communication module at a preset frequency (e.g., once per hour). After obtaining the leakage current and varistor voltage, these values can be substituted into a pre-calibrated exponential decay model for calculation. The initial parameters of this model can be obtained by fitting accelerated aging test data of the same model of device, and in actual operation, combined with the closed-loop correction mechanism of step S500, the decay coefficient is dynamically calibrated using actual failure cases.
[0160] The model uses the increment of the current leakage current relative to the initial value, the deviation of the varistor voltage relative to the nominal value, and the historical cumulative surge impact energy as input variables to calculate a normalized index characterizing the health of the device, which is then mapped to the predicted value of the remaining service life.
[0161] When the predicted value is lower than the preset lifespan threshold, the system automatically generates a replacement reminder work order containing the node location, device number, and predicted remaining days, and pushes it to the operation and maintenance management platform.
[0162] In other implementations, besides the exponential decay model, the Weibull distribution model, neural network regression model, or other algorithms that can characterize the degradation law of the device can also be used to predict the lifetime, as long as a quantitative mapping relationship between electrical parameters and remaining lifetime can be established.
[0163] It should be noted that this step, by upgrading the passive protection of surge protectors to active self-sensing, solves the safety blind spot problem caused by the unpredictable failure of traditional lightning protection devices. Simultaneously, predictive maintenance based on real-time status monitoring avoids resource waste caused by premature replacement and prevents the risk of protection failure due to faulty devices. This mechanism provides accurate boundary information of the underlying protection capabilities to the upper-level decision-making model, enabling the tiered arbitration strategy to be formulated based on the actual health status of the equipment rather than assumed ideal conditions, thereby improving the overall reliability of safety management decisions.
[0164] Step E200: When the distributed monitoring node enters the emergency protection mode, the control electronic circuit breaker disconnects the interconnection port and triggers the anti-reverse current diode to isolate the reverse current.
[0165] Among them, the emergency protection mode is the specific protection mode corresponding to the node status adjustment command executed when the environmental risk level is the first risk level, as described in Example 3. It can cut off all possible risk propagation paths to protect the equipment itself.
[0166] Electronic circuit breakers are solid-state or hybrid switching devices with millisecond-level fast breaking capability. They are connected in series at the DC interconnect port of the distributed monitoring node and controlled by the node's main control unit.
[0167] Anti-reverse current diodes are unidirectional conductive semiconductor devices connected in parallel or series in interconnect lines. They are used to provide nanosecond-level reverse current blocking capability after an electronic circuit breaker is tripped, preventing reverse surge current caused by line induced electromotive force or faults in adjacent nodes from entering the device.
[0168] In some implementations, when the node master control unit receives a coordinated control command and determines that it is in emergency protection mode, it immediately sends a trip signal to the electronic circuit breaker drive circuit. At this time, the electronic circuit breaker can complete the separation action of mechanical or solid contacts within 1 millisecond after receiving the signal, physically cutting off the electrical connection of the interconnecting lines. Alternatively, within microseconds before the electronic circuit breaker operates, the anti-reverse current diode, with its inherent unidirectional conductivity, automatically forms a high-impedance isolation barrier against reverse transient voltages or currents on the line.
[0169] The aforementioned timing coordination ensures that reverse surge energy is effectively blocked during the brief window period when the electronic circuit breaker has not yet fully disconnected or an electric arc is generated between the contacts.
[0170] Synchronously disable the enable signal of the DC-DC converter at the interconnect port to eliminate potential coupling paths.
[0171] The specific selection parameters of the electronic circuit breaker and anti-reverse diode (such as rated current 10A, reverse withstand voltage 100V, response time, etc.) can be adapted and adjusted according to the voltage level and surge protection level requirements of the actual system.
[0172] It should be noted that this step, through a hardware-software co-operational physical execution mechanism, transforms the upper-level decision-making logic into deterministic electrical isolation actions. Compared to isolation methods that rely solely on software protocols or relay control, the combination of the rapid breaking of electronic circuit breakers and the intrinsic isolation characteristics of anti-reverse current diodes can effectively cope with high-frequency, high-energy transient interference such as lightning surges, preventing isolation failures caused by software crashes, communication delays, or contact adhesion. This provides a solid physical safety net for safety management methods, ensuring that emergency protection commands can be reliably executed even in the most severe electromagnetic environments, thereby enhancing the survivability of distributed monitoring systems under extreme conditions.
[0173] Step E300: When a communication link interruption with the central management platform is detected, the distributed monitoring node executes localized protection actions based on the most recently received decision instructions stored locally and the data collected in real time by local sensors. The localized protection actions include: determining the lightning risk level based on the local atmospheric electric field strength measurement value, and automatically switching to the corresponding protection mode when the lightning risk level exceeds the preset risk threshold.
[0174] Among them, communication link interruption refers to a continuous failure of the wired or wireless data transmission channel between the distributed monitoring node and the central management platform, resulting in the node being unable to receive new collaborative control commands or report status data.
[0175] Localized protection actions refer to the degraded security protection strategies that nodes autonomously execute based on their own computing power and sensing resources when central control is lost.
[0176] The most recently received decision instruction stored locally refers to the last valid instruction and its associated policy parameters cached by the node when communication is normal, serving as a baseline behavior reference during network outages.
[0177] Local atmospheric electric field intensity measurements are physical quantities that characterize the intensity of local lightning activity and are collected in real time by atmospheric electric field meters deployed on or near the node. They are the direct basis for judging the environmental risk level under network outage conditions.
[0178] In some implementations, the communication heartbeat monitoring mechanism built into the node master control unit determines that the communication link is interrupted when no heartbeat response or instruction is received from the central platform for several consecutive cycles, and then automatically switches to local autonomous mode.
[0179] In this mode, the node first loads the most recent decision instruction cached in its local non-volatile memory and parses the protection strategy parameters and risk threshold configurations within it. This allows the node to continuously collect local atmospheric electric field strength data and compare it with the various risk thresholds defined in the cached strategy. If the measured atmospheric electric field strength exceeds the preset first risk threshold (corresponding to high risk), it immediately enforces an emergency protection mode, disconnecting all interconnect ports and reducing its own power consumption, regardless of other conditions. If the measured value is between the first and second risk thresholds (corresponding to medium risk), it performs power reduction operation and TVS protection enhancement operations, based on the compromise control parameters in the cached instruction. If the measured value is below the second risk threshold, it maintains the normal operating state or low-power standby state indicated by the cached instruction.
[0180] Throughout the autonomous process, the node will still attempt to reconnect the communication link periodically. Once the connection is restored, it will immediately exit the autonomous mode and synchronize the event logs recorded during the network outage.
[0181] In some implementation methods, in addition to atmospheric electric field strength, multi-dimensional sensor data such as local illumination, wind speed or vibration can be combined to conduct comprehensive risk assessment and improve the accuracy of autonomous decision-making. However, atmospheric electric field strength should be the primary criterion due to its direct causal relationship with lightning risk.
[0182] It should be noted that the security management system built in this step can ensure that nodes will not fall into a vulnerable state due to network outages by combining the edge deployment of central decision-making logic with local real-time perception and autonomous decision-making. This local autonomous mechanism not only maintains a minimum level of security protection capabilities but also ensures the continuity of behavioral policies before and after network outages through the inheritance of cached instructions, avoiding policy abrupt changes or service instability caused by sudden loss of connection. At the same time, it also endows the distributed monitoring system with decentralized resilience, ensuring that the communication failure of a single node will not cause the collapse of the entire network protection system, thus meeting the application requirements of outdoor monitoring systems in high-reliability scenarios.
[0183] Based on the above technical solutions, this embodiment constructs a triple-boundary security defense line of intelligent surge protector lifespan prediction, hardware-level dual isolation, and local autonomy during network outages. This not only upgrades the operation and maintenance mode of lightning protection devices from passive replacement to proactive prediction, eliminating protection blind spots, but also ensures the reliable execution of emergency protection commands under transient interference through rapid isolation actions coordinated by software and hardware. Furthermore, by leveraging local autonomy strategies, it guarantees minimum security for the system under extreme conditions of communication interruption. This effectively compensates for the limitations in the face of uncertainties in the physical world and the vulnerability of infrastructure, providing comprehensive and highly resilient technical support for the full lifecycle security of distributed monitoring nodes.
[0184] Example 7: This embodiment provides a full-process application verification scenario under the combined conditions of summer thunderstorms and continuous overcast and rainy weather. The aim is to place the dispersed technical features from the aforementioned embodiments into a real business environment for cascade verification, demonstrating the synergistic amplification effect of the perception, arbitration, and correction closed loop under complex and extreme conditions. In this scenario, a distributed monitoring network in a mountainous area encounters severe convective weather. The system detects that the atmospheric electric field intensity jumps to over 2.5kV / m in a short period. Simultaneously, due to five consecutive days of overcast and rainy weather, the state of charge of batteries across the entire network nodes is generally below 30%, creating a complex and critical situation with both high lightning risk and low energy reserves.
[0185] During the perception and risk simulation phase, the system executes step S100 to obtain multi-source state influence factor data and triggers the local subgraph update mechanism described in step A100.
[0186] Among them, the identification of abnormal changes under complex working conditions refers to the system's ability to analyze the coupling state of other factors while detecting a single factor exceeding its limit, thereby accurately defining the comprehensive risk attributes of the current environment.
[0187] In this scenario, when the atmospheric electric field sensor detects a sudden increase in value from 0.8 kV / m to 2.5 kV / m, the system determines that the factor has undergone an abnormal change and immediately activates the local subgraph update process to recalculate the correlation strength and confidence of the edges between the atmospheric electric field node and its neighboring nodes (such as humidity, wind speed, and battery SOC). At this time, the system can use the measured data from the most recent 10 minutes to quickly update the parameters of the two critical paths: the probability of anomalies in the atmospheric electric field pointing to lightning strikes and the vulnerability of low SOC to sustained power. This allows the multi-source influencing factor correlation model to reflect the current combined situation of high-risk lightning and energy shortage within seconds.
[0188] Therefore, the decision-making model can be directly updated based on the event-triggered local update mechanism to ensure that the decision-making model can capture the qualitative change of environmental risk in a timely manner within the tight time window when thunderstorms are approaching, providing timely and coupled data support for subsequent accurate arbitration.
[0189] In the decision generation and hierarchical arbitration phase, the system executes step B100 to calculate the importance score of the nodes, and executes hierarchical arbitration in step S300 to generate differentiated collaborative control instructions.
[0190] Among them, differentiated collaborative control refers to applying non-uniform protection strategies to different types of nodes based on the dynamic matching results of the real-time value and risk level of nodes under the contradictory state of limited resources and high risks.
[0191] In this scenario, the system first calculates the importance score of each node based on the formula in step B100, combined with topological centrality, task criticality, and real-time requirements. For example, a relay node located at the intersection of mountain ridges is identified as a first-level important node because of its high topological centrality and the fact that it is currently performing a video transmission task; while a regular acquisition node located at the edge of the forest is identified as a second-level important node because of its low connectivity and the fact that it is in an idle polling state.
[0192] Subsequently, for the primary node, the system generates a collaborative control command that includes disconnecting the physical interconnection according to the arbitration rules of the first risk level in step S300, and enforces the emergency protection mode to prevent surges from damaging the core hub equipment along the interconnection cables. As for the secondary nodes, considering that their own power is below 30% and there will be no sunlight to supplement them in the next 2 hours, if the interconnection is completely disconnected, it will cause them to lose power and connection before the thunderstorm ends. Based on the arbitration rules of the second risk level, the system generates a cooperative control command that includes compromise control, that is, to maintain the physical interconnection but limit the transmission power to 50% of the rated value, and simultaneously activate the port TVS protection array.
[0193] The above design creates a dynamic balance between security and survival at the physical level, which avoids both business paralysis caused by a rigid power outage across the entire network under high risk and chain damage caused by blind mutual assistance.
[0194] During the feedback and model correction phase, the system executes the closed-loop correction process described in steps S500 and C200 after the thunderstorm process ends.
[0195] Among them, cognitive evolution based on actual combat results refers to the process of using the consequences of actual physical events as truth labels to calibrate and optimize the associated knowledge and prediction parameters of the decision-making model.
[0196] In this scenario, after a thunderstorm passes, the system collects execution feedback data and environmental verification data from each node. If the data shows that the primary nodes performing emergency protection are all intact, while some secondary nodes performing compromise control are not damaged, they may experience slight port voltage fluctuations.
[0197] At this point, the system inputs these measured results into the correction process: for the intact records of the first-level nodes, the confidence level corresponding to the association rules between high risk and emergency protection ensuring equipment safety is enhanced; for the voltage fluctuation records of the second-level nodes, the safety boundary parameters corresponding to the trade-off control and TVS protection are fine-tuned, and the weights of power generation efficiency during the post-rain recovery period in the energy prediction model are updated using an elastic weight integration algorithm. Simultaneously, the updated model parameters are verified through a Markov chain Monte Carlo process to ensure convergence and stability. This correction mechanism based on real battle damage data enables the system not only to complete the current safety management task but also to transform the experience gained in dealing with this complex operating condition into long-term knowledge within the model, effectively improving the decision-making accuracy and robustness when facing similar scenarios in the future.
[0198] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for security management of distributed monitoring nodes based on an outdoor monitoring system, characterized in that, include: Acquire multi-source state influence factor data of distributed monitoring nodes, and update the multi-source influence factor association model based on the multi-source state influence factor data. The multi-source influence factor association model is a directed weighted graph model that characterizes the association relationship, association strength and confidence between influence factor nodes. The node status is simulated using the multi-source influence factor correlation model to determine the current environmental risk level and generate node status adjustment instructions and resource mutual assistance scheduling instructions. When the node status adjustment instruction requires disconnecting the physical interconnection between distributed monitoring nodes, while the resource mutual assistance scheduling instruction requires maintaining the physical interconnection between distributed monitoring nodes, a graded arbitration is performed based on the current environmental risk level to generate a collaborative control instruction, which is then sent to the corresponding distributed monitoring node. The execution feedback data and environmental verification data of the distributed monitoring node after executing the collaborative control command are obtained, and the multi-source influence factor correlation model is corrected based on the execution feedback data and the environmental verification data. The corrected multi-source influence factor correlation model is used for node state inference in the next cycle.
2. The method for security management of distributed monitoring nodes based on an outdoor monitoring system according to claim 1, characterized in that, The process of updating the multi-source influence factor correlation model based on the multi-source state influence factor data includes: During operation, when an abnormal change is detected in the target influence factor in the multi-source state influence factor data, a local subgraph update is triggered to recalculate the association strength and confidence of the associated edges between the neighboring nodes corresponding to the target influence factor. According to a preset cycle, the historically accumulated multi-source state influence factor data is updated in batches, and the association strength and confidence of all association edges in the multi-source influence factor association model are recalculated.
3. The method for security management of distributed monitoring nodes based on an outdoor monitoring system according to claim 1, characterized in that, The process of generating a collaborative control instruction by performing graded arbitration based on the current environmental risk level when the node status adjustment instruction requires disconnecting the physical interconnection between distributed monitoring nodes, while the resource mutual assistance scheduling instruction requires maintaining the physical interconnection between distributed monitoring nodes, includes: When the current environmental risk level is the first risk level, a collaborative control instruction containing the node status adjustment instruction is generated to disconnect the physical interconnection between the distributed monitoring nodes; When the current environmental risk level is the second risk level, a collaborative control command containing compromise control is generated to maintain the physical interconnection between distributed monitoring nodes, reduce the transmission power of interconnection lines, and increase transient voltage suppression protection. When the current environmental risk level is the third risk level, a collaborative control instruction containing the resource mutual assistance scheduling instruction is generated to maintain full-power physical interconnection between distributed monitoring nodes; The risk level of the first risk level is higher than that of the second risk level, and the risk level of the second risk level is higher than that of the third risk level.
4. The method for security management of distributed monitoring nodes based on an outdoor monitoring system according to claim 1, characterized in that, The process of correcting the multi-source influence factor correlation model based on the execution feedback data and the environmental verification data includes: Define a confidence interval for the predicted state quantity. When the actual measured value falls outside the confidence interval, trigger an online correction process. In the online correction process, the weight parameters of the energy prediction model are updated using the execution feedback data, and the conditional probabilities of the corresponding correlations in the multi-source influence factor correlation model are updated using the environmental validation data. Define a confidence interval for the predicted state quantity output by the energy prediction model. When the actual measured value falls outside the confidence interval, trigger an online correction process. In the online correction process, the weight parameters of the energy prediction model are updated using the execution feedback data. The energy prediction model is a time-series prediction model trained based on historical operating data and used to predict the future energy state of distributed monitoring nodes. Using the environmental validation data, the conditional probabilities of the corresponding associations in the multi-source influence factor association model are updated to improve the accuracy of subsequent node state inferences.
5. The method for security management of distributed monitoring nodes based on an outdoor monitoring system according to claim 2, characterized in that, After recalculating the association strength and confidence of all association edges in the multi-source influence factor association model, the method further includes: Perform a chi-square test on the association rules corresponding to the association edges in the multi-source influence factor association model, and remove association edges that fail the significance test; Time consistency verification is performed by determining the time lag by calculating the peak position of the cross-correlation function of the time series of two influencing factors. It is required that the antecedent of the association rule occurs earlier than the consequent. If the time lag is negative, it is judged as a false association and removed. Continuous effectiveness monitoring is performed. Within a preset sliding time window, if the number of windows in which the confidence of the associated edge is continuously lower than a preset confidence threshold reaches a first preset number, the associated edge is marked as pending verification. If the number reaches a second preset number, the associated edge is deleted from the multi-source influence factor association model. The second preset number is greater than the first preset number.
6. The method for security management of distributed monitoring nodes based on an outdoor monitoring system according to claim 3, characterized in that, The process of generating the node state adjustment instruction includes: The importance score of the distributed monitoring nodes is calculated based on the topological centrality, task criticality, and real-time requirements of the distributed monitoring nodes. Combining the current environmental risk level, the importance level to which the importance score belongs, and the current energy state of the distributed monitoring node, a corresponding protection mode instruction is generated by mapping through a three-dimensional decision matrix. The protection mode instruction is the specific execution mode of the node state adjustment instruction. Wherein, the topology centrality characterizes the degree of connection hub of the distributed monitoring node in the network topology, the task criticality characterizes the preset importance of the location of the distributed monitoring node, the real-time demand characterizes whether the distributed monitoring node is currently performing a monitoring task, and the importance level is determined based on the comparison result of the importance score and the preset score threshold.
7. The method for security management of distributed monitoring nodes based on an outdoor monitoring system according to claim 4, characterized in that, The process of updating the weight parameters of the energy prediction model using the execution feedback data includes: An elastic weight integration algorithm is used to update the weight parameters of the energy prediction model in order to mitigate catastrophic forgetting during the incremental learning process. The newly collected execution feedback data is mixed with the historical training data of the energy prediction model according to a preset ratio to update the parameters of the energy prediction model; The updated energy prediction model is validated using a Markov chain Monte Carlo process. In the Markov chain Monte Carlo process validation, when the convergence diagnostic index of the parameter posterior distribution reaches a preset convergence threshold, the validation is deemed successful, and the updated energy prediction model is deployed.
8. The method for security management of distributed monitoring nodes based on an outdoor monitoring system according to claim 1, characterized in that, Also includes: The system obtains the leakage current and varistor voltage of the intelligent surge protector configured in the distributed monitoring node, and predicts the remaining lifespan of the intelligent surge protector based on the exponential decay model. When the remaining lifespan is lower than the preset lifespan threshold, a protector replacement reminder is triggered. When the distributed monitoring node enters emergency protection mode, it controls the electronic circuit breaker to disconnect the interconnection port and triggers the anti-reverse current diode to isolate the reverse current. The emergency protection mode is the protection mode corresponding to the node status adjustment command executed when the current environmental risk level is the first risk level.
9. The method for security management of distributed monitoring nodes based on an outdoor monitoring system according to claim 1, characterized in that, The generation basis of the resource mutual assistance scheduling command includes energy evolution prediction results, and the process of obtaining the energy evolution prediction results includes: Using historical light intensity, wind speed, temperature, battery state of charge, load power consumption, and future weather forecast data as input, and through a long short-term memory network, the system outputs the time-by-time photovoltaic power generation prediction value and battery state of charge prediction value for each distributed monitoring node within a preset future time period. The photovoltaic power generation prediction value and the battery state of charge prediction value constitute the energy evolution prediction result. The Long Short-Term Memory (LSTM) network is configured with a hybrid training architecture that combines offline training and online incremental learning. The offline training completes the initial parameter configuration based on historical data, while the online incremental learning fine-tunes the network weights using newly collected data at preset intervals.
10. The method for security management of distributed monitoring nodes based on an outdoor monitoring system according to claim 1, characterized in that, Also includes: When a communication link with the central management platform is interrupted, the distributed monitoring node performs localized protection actions based on the most recently received decision instructions stored locally and the data collected in real time by local sensors. The localized protection actions include: determining the lightning risk level based on the local atmospheric electric field strength measurement value, and automatically switching to the corresponding protection mode when the lightning risk level exceeds a preset risk threshold.
