Energy Sector Disaster Prediction and Assessment Methods Based on Multi-Source Data

Abstract

This invention relates to the field of disaster prediction technology in the energy industry, and discloses a method for disaster prediction and assessment in the energy industry based on multi-source data. The method includes: generating binary status words through distributed edge nodes based on fixed rules; calculating the Hamming distance of the status word sequence on the central platform to obtain the consensus volatility and its first-order difference as the collapse acceleration; and triggering an early warning when the acceleration continuously exceeds a threshold. This invention transforms traditional probabilistic early warning into deterministic action instructions through the collaborative monitoring of binary status and collapse acceleration. It also combines topology propagation coherence verification to eliminate false alarms and outputs diagnostic-level composite early warnings based on micro-rule classification, thus realizing a closed loop from disaster identification to precise handling. This method and its simplified architecture enable real-time stability monitoring of large-scale energy networks, significantly improving the reliability of early warnings and response efficiency.

Description

Technical Field

[0001] This invention relates to a method for disaster prediction and assessment in the energy industry based on multi-source data, belonging to the field of disaster prediction technology in the energy industry. Background Technology

[0002] The disaster prediction system in the energy industry has long relied on the fusion of multi-source data and the construction of complex models. Its mainstream technical path focuses on approximating the physical nature of disasters by improving the accuracy of the models. Existing methods usually deploy highly complex AI algorithms or physical models to perform real-time fusion analysis of distributed sensor data and output probabilistic risk assessment results such as 70% icing risk.

[0003] However, this approach has fundamental technical contradictions: 1. When the system is coupled with a sudden change in wind speed before freezing rain and a sudden drop in equipment temperature in a complex disaster scenario, the multiple sets of probabilistic alarms it outputs often conflict with each other. Maintenance personnel need to subjectively judge the authenticity of the risk amidst information overload, leading to delays in critical response windows, which is the core bottleneck of decision-making efficiency; 2. To ensure full coverage of potential risks, the system needs to maintain continuous high sensitivity, causing frequent low- and medium-risk alarms to be triggered by daily fluctuations. Frequent non-emergency alarms reduce decision-makers' trust in the system, resulting in a failure mode where real high-risk signals are drowned out by background noise; 3. Existing systems lack the ability to perceive the coupling effect of new complex disasters not covered by training data, such as unknown corrosion and mechanical vibration. Their architecture is inherently difficult to cope with sudden chain failures that exceed historical experience.

[0004] In recent years, the industry has attempted to alleviate the above problems through model lightweighting or dynamic threshold adjustment, but has encountered new dilemmas: simplifying the model sacrifices the ability to capture key features, while dynamic thresholds, due to the lack of a systematic state assessment benchmark, introduce new risks of misjudgment in equipment aging or environmental drift. Therefore, how to construct a system that can avoid probabilistic decision ambiguity and alleviate the problem of reduced trust caused by frequent alarms has become the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a disaster prediction and assessment method for the energy industry based on multi-source data. Its main purpose is to solve the problems of decision delay caused by the ambiguity of probability output in existing prediction systems, the crisis of trust caused by high sensitivity in alarms, and the blind spots of unknown risks caused by model dependence.

[0006] To achieve the above objectives, this invention provides a method for disaster prediction and assessment in the energy industry based on multi-source data, comprising the following steps:

[0007] Step a: Deploy multiple distributed edge computing nodes within the energy system and pre-store their respective spatial location information. Based on their respective local data sources and a set of logically fixed micro-heuristic rules stored within the edge computing nodes, generate binary state words representing local states independently within each predetermined time step.

[0008] Step b: On the central platform, binary status words generated by multiple distributed edge computing nodes within a predetermined time step are aggregated to form a global status vector sequence.

[0009] Step c: Based on the global state vector sequence, calculate the Hamming distance between global state vectors of adjacent time steps to obtain a consensus volatility sequence, and calculate its first-order difference in time based on the consensus volatility sequence to obtain the consensus collapse acceleration.

[0010] Step d: When the consensus collapse acceleration exceeds a first threshold within a predetermined duration, the following steps are executed concurrently: Step d1: Obtain the global state vector sequence within a backtracking time window that caused the consensus collapse acceleration to exceed the first threshold, and combine it with the spatial location information of the edge computing nodes to calculate the propagation coherence of the edge computing nodes whose states changed within the backtracking time window in the spatial topology, and obtain the topology propagation coherence index; Step d2: Based on the physical attributes of the local data source associated with the micro-heuristic logic rules, classify the micro-heuristic logic rules into domains in advance, and count the domain classifications of the binary bits that changed during the consensus collapse acceleration exceeding the first threshold, to determine a dominant collapse mode;

[0011] Step e: When the topology propagation coherence index exceeds the second threshold, the event is confirmed as an endogenous system collapse, and a disaster warning containing the dominant collapse mode as diagnostic information is triggered.

[0012] Preferably, the micro-heuristic logic rule in step a is a fixed logic judgment set based on the safety operation limits of the design specifications of equipment in the energy system, or based on the empirical limits obtained from the statistical analysis of historical safety operation data.

[0013] Preferably, the method further includes an adaptive calibration step for the judgment threshold of the micro-heuristic logic rules. The steps include: on the central platform, statistical analysis of the consensus volatility sequence within a preset statistical period during the operation period when no disaster warning is triggered, to obtain the statistical distribution characteristics of the background consensus volatility; comparing the statistical distribution characteristics of the background consensus volatility with a stored ideal distribution interval that characterizes the health status of the system; when the statistical distribution characteristics of the background consensus volatility exceed the ideal distribution interval, generating a global threshold adjustment factor and applying it to the micro-heuristic logic rules of all edge computing nodes to incrementally adjust their judgment thresholds.

[0014] Preferably, the calculation of the topology propagation coherence index in step d1 specifically includes: identifying edge computing nodes that undergo state changes within the backtracking time window; calculating the average proportion of the nodes that undergo state changes within the backtracking time window whose direct neighbor nodes in the preset physical space also undergo state changes in adjacent time steps, and using this average proportion as the topology propagation coherence index.

[0015] Preferably, the domain classification of the micro-heuristic logic rules in step d2 includes classifying rules related to wind speed and air pressure into the meteorological stress domain, rules related to equipment vibration and structural deformation into the mechanical stress domain, and rules related to current, voltage, and insulation parameters into the electrical anomaly domain.

[0016] Preferably, the method further includes: after confirming that the event is an endogenous system crash, recording the dominant crash mode of the event and the peak value of the consensus volatility during the event; associating and matching the dominant crash mode and the peak value of the consensus volatility with a historical event knowledge base that stores the mapping relationship between parameters of historical endogenous system crash events and the actual loss level determined after the fact; and determining and outputting the estimated risk level of the current endogenous system crash event based on the association matching result.

[0017] Preferred risk level The determination of is based on the following calculation rules: ,in, This represents the peak value of consensus volatility in the current endogenous system crash event. The dominant crash pattern for the current event. To extract data from a historical event knowledge base based on the dominant crash pattern Query and extract the risk weight coefficients.

[0018] Preferably, the method further includes: when the consensus collapse acceleration exceeds a first threshold within a predetermined duration, but the topology propagation coherence index does not exceed a second threshold, the event is judged as a synchronous external disturbance event; when judged as a synchronous external disturbance event, a non-disaster alert indicating an abnormal external environment is issued, and any adaptive calibration action on the micro-heuristic logic rules is suppressed.

[0019] Preferably, the central platform in step b is a central server or cloud computing platform that has the computational capability to process global state vector sequences and stores the first threshold and the second threshold.

[0020] Preferably, the method further includes combining disaster warning, dominant collapse mode and estimated risk level into a composite decision instruction, and matching and outputting an emergency response plan from the emergency response plan library based on the composite decision instruction.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] 1. Edge nodes generate simplified binary status words based on fixed rules. The central platform calculates the consensus volatility and its first-order difference using Hamming distance. When the acceleration continuously exceeds the threshold, the system directly determines it as a critical point of stability collapse, rather than outputting a probabilistic risk value. This mechanism transforms the traditional warning from a vague indication of potential risk into a deterministic action instruction that the system is about to become unstable, reducing the risk of decision delay. Before the consensus collapse acceleration triggers the warning, the system backtracks and analyzes the propagation coherence of nodes with state changes in physical space, such as the correlation ratio of state changes of adjacent nodes. Only when the spatial propagation coherence synchronously meets the threshold is it confirmed as an intrinsic collapse. This mechanism, through spatiotemporal coupling verification, eliminates false alarms caused by external synchronization interference, while suppressing the malfunction of the adaptive calibration module under abnormal operating conditions, thus solving the problem of alarm fatigue.

[0023] 2. The micro-heuristic rules of edge nodes are pre-classified according to physical attributes, such as mechanical / electrical / meteorological fields, and the dominant field label is statistically analyzed in the crash event. The consensus volatility peak is further reused and the loss level mapping relationship of similar labels in historical events is combined to output the dominant crash mode + estimated risk level in real time. This enables a single alarm to carry disaster location and intensity diagnosis information simultaneously, and shifts the emergency response from comprehensive investigation to precise handling.

[0024] 3. The adaptive calibration module generates a global threshold adjustment factor by statistically analyzing the background consensus volatility over a long period and applies it incrementally to edge nodes; the crash diagnosis module continuously accumulates the feature mapping relationship of historical events; and the topology analysis module ensures the purity of calibration data. The three modules form a closed loop with consensus crash acceleration as the hub, enabling the system to have the ability to autonomously adapt to equipment aging and environmental drift, and to optimize diagnostic accuracy during continuous operation.

[0025] 4. Edge nodes only need to perform Boolean logic judgments, and the core operations of the central platform are reduced to Hamming distance and difference calculations. All modules reuse status word sequences and preset rule labels, avoiding the introduction of heavy algorithms such as neural networks. This architecture, which transforms nonlinear stability problems into lightweight time series analysis, enables large-scale energy networks to achieve real-time crash warnings on edge devices and servers. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0027] Figure 2 A comparison of the probability density distribution of consensus volatility before and after adaptive calibration, serving as the background of this invention;

[0028] Figure 3 This is a functional module and data flow diagram of the present invention.

[0029] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and should not be used to limit the scope of protection of the present invention.

[0031] This invention discloses a disaster prediction and assessment method for the energy industry based on multi-source data. The overall technical solution is deployed on a distributed edge computing and centralized analysis platform. The process sequentially includes four core stages: data localization specification, global state consensus monitoring, spatiotemporally coupled crash event verification, and diagnostic early warning and risk assessment. This method is applied to large-scale energy systems, such as wind farms and their transmission networks deployed in wide-area complex geographical environments. It aims to shift system safety monitoring from a passive response mode relying on probabilistic risk warnings to precise early warning and proactive monitoring based on deterministic system state evolution characteristics. Handling Mode; In a specific application, such as an energy system comprising hundreds of wind turbine generators and associated transmission and transformation facilities, this system faces the long-term risk of chain failures caused by the coupling of multiple factors such as icing, strong winds, and structural fatigue. The challenge in identifying this risk lies in the fact that local data anomalies at a single node, such as the instantaneous vibration of a single generator, are insufficient to predict system-wide instability. However, global environmental changes, such as the passage of a cold front, can easily trigger large-scale synchronous data fluctuations, creating false alarms that are difficult to effectively identify. To address this challenge, the key is to extract the true cause of system-wide collapse from massive distributed data sources. Qualitative precursors: The implementation of this invention begins with deploying a distributed edge computing node at key locations within the physical infrastructure of the energy system, such as the nacelle and blade roots of each wind turbine and critical structural points of each transmission tower. Each such node pre-stores its precise three-dimensional spatial location information and a set of logically fixed micro-heuristic rules. These rules are set based on the safe operating limits specified in the equipment's design specifications or empirical limits derived from statistical analysis of historical safe operating data, making fixed logical judgments. For example, a rule for blade icing risk could be set to determine when the blade surface temperature... When the temperature is below 2°C and the micro-climate humidity is above 90% for 5 consecutive minutes, state bit 1 outputs 1. Another rule can be set so that when a specific icing characteristic frequency appears in the cabin vibration spectrum, state bit 2 outputs 1. Within each predetermined time step, such as every 1 second, the node independently performs Boolean operations on this set of micro-heuristic rules based on its acquired local data source to generate a binary state word representing the local multi-dimensional state. In this way, each node transforms the high-dimensional heterogeneous raw sensor data stream into an extremely compressed and information-standardized binary format at the source, laying the data foundation for subsequent global analysis.

[0032] To discover the dynamic evolution trend of the system as a whole from isolated local states, the central platform located in the monitoring center aggregates binary status words from all distributed edge computing nodes at each time step and constructs them into a global state vector according to the fixed numbering order of the nodes. Over time, these vectors form a global state vector sequence. Therefore, the system needs an indicator to quantify the drastic changes in the global state. For this purpose, the central platform is configured to calculate the state at two adjacent time steps, for example, the first... Seconds and the The Hamming distance between global state vectors over a given second, which represents the total number of binary bits in the system that undergo state changes within that time step, is defined as the consensus volatility. A consistently stable consensus volatility sequence indicates stable system operation, while drastic changes indicate widespread state transitions within the system. To further capture the accelerating trend of state transitions, the central platform calculates the first-order difference in time based on the consensus volatility sequence to obtain the consensus collapse acceleration. This acceleration index directly reflects the rate of system instability, shifting the monitoring focus from the state itself to the dynamic trend of state changes. However, a sharp increase in consensus collapse acceleration alone cannot confirm an endogenous catastrophic collapse, as it may also be caused by external synchronous events affecting the entire domain. To filter out such false alarms, this invention designs a spatiotemporal coupling verification procedure. When the consensus collapse acceleration continuously exceeds a first threshold for a predetermined duration, the system does not immediately issue an alarm but concurrently executes two analysis tasks. First, the system acquires the global state vector sequence within a backtracking time window that caused the consensus collapse acceleration to exceed the threshold, and combines it with the edge... The system uses pre-stored spatial location information of nodes to perform topological path analysis on nodes whose states change within the window. Specifically, the algorithm identifies edge computing nodes whose states change within the backtracking time window and calculates the average proportion of these nodes whose direct neighbors in the preset physical space also change states in adjacent time steps. This average proportion serves as a topological propagation coherence index. A chain collapse caused by an internal fault will necessarily exhibit high propagation coherence in space, while an external synchronous disturbance will cause a large number of non-adjacent nodes to change simultaneously, resulting in a lower index. Secondly, the system pre-classifies rules based on the physical attributes of the local data source associated with the micro-heuristic logic rules. For example, rules related to wind speed and air pressure are classified as meteorological stress, rules related to equipment vibration and structural deformation as mechanical stress, and rules related to current, voltage, and insulation parameters as electrical anomalies. During periods when consensus collapse acceleration exceeds a threshold, the system statistically analyzes the domain classification of the binary bits that change in the consensus volatility to determine a dominant collapse mode.

[0033] The generation of decision instructions is based on the above verification results. Only when the topology propagation coherence index exceeds the second threshold does the system confirm the event as an endogenous system collapse and trigger a disaster warning containing the dominant collapse mode as diagnostic information. This allows emergency responders to know the fundamental nature of the disaster immediately. Conversely, if the consensus collapse acceleration exceeds the first threshold, but the topology propagation coherence index does not exceed the second threshold, the system judges the event as a synchronous external disturbance event, issues a non-disaster warning indicating an abnormal external environment, and suppresses any adaptive calibration actions on the micro-heuristic logic rules to prevent the model from being contaminated by abnormal environmental data. To achieve a forward-looking assessment of disaster intensity, this invention also includes a risk rating module based on historical knowledge. After confirming an endogenous system collapse, the system records the dominant collapse mode of the event and the peak value of the consensus volatility. The system internally maintains a historical event knowledge base that stores the mapping relationship between historical endogenous system collapse event parameters and the actual loss level determined after the fact. For the current event, the system determines the risk level based on its dominant collapse mode. Query and extract the corresponding risk weight coefficients from the knowledge base. This coefficient reflects the average destructive power of this type of crash mode historically, and then the system operates according to the calculation rules. The estimated risk level of the current event is calculated. The risk level, along with the dominant collapse mode and disaster warning, are combined to form a composite decision instruction. This instruction can then be used to match and output an emergency response plan from the emergency response plan library. Finally, to ensure the system can adapt to slow drift factors such as equipment aging and seasonal environmental changes during long-term operation, this invention also designs an adaptive calibration mechanism. On the central platform, the system statistically analyzes the consensus volatility sequence during normal operation periods without triggering disaster warnings within a preset statistical period to obtain the statistical distribution characteristics of the background consensus volatility. This statistical distribution characteristic is then compared with an ideal distribution range that represents the system's health status. When it exceeds the ideal distribution range, the system generates a global threshold adjustment factor and applies it to the micro-heuristic logic rules of all edge computing nodes to incrementally adjust their judgment thresholds. The effectiveness of this process is ensured by the successful filtering of external disturbance events by the aforementioned topology analysis module, improving the effectiveness of the calibration data and enabling the entire system to self-optimize during continuous operation.

[0034] Example 1: In the actual operation of a large-scale wind farm deployed in a mountainous area, the system faces a disaster scenario coupled with multiple factors in late winter and early spring. Initially, due to nighttime radiative cooling, thin, uneven ice buildup occurred on the blades of some turbines located in leeward mountain valleys. Simultaneously, a relatively weak frontal system was passing through the entire wind farm, causing continuous and irregular turbulent fluctuations in wind speed and direction across the entire area. Under these conditions, the micro-heuristic logic rules of a large number of the hundreds of distributed edge computing nodes across the farm were triggered in response to the wind field fluctuations. This was mainly manifested in the frequent flipping of binary state bits under the meteorological stress domain classification. This is reflected on the central platform as the consensus volatility oscillating at a relatively high baseline, but its first difference, i.e., the consensus collapse acceleration, fluctuates narrowly around zero and does not continuously exceed the first threshold. As a result, the system does not respond to this global but non-catastrophic condition. As the blade icing of local units gradually intensifies, the uneven distribution of its mass begins to trigger perceptible mechanical vibrations during unit operation. Correspondingly, the binary state bits of these specific unit nodes belonging to the mechanical stress domain begin to be triggered. In the initial stage, these local state bit changes are completely submerged in the background noise of consensus volatility caused by the wind disturbance across the entire field, making it impossible to effectively identify them by observing the absolute value of consensus volatility.

[0035] The consensus collapse acceleration calculation mechanism of this invention provides a way to discover the evolution of such hidden faults. Although the state position changes caused by icing account for a small proportion in the early stage, their evolution has the physical characteristic of continuous acceleration. That is, the heavier the icing, the faster the vibration intensifies and the faster the state position changes. This continuous acceleration trend causes the first difference of the consensus volatility, i.e., the consensus collapse acceleration, to show a continuous and unidirectional positive deviation. When this deviation accumulates within a predetermined duration and eventually exceeds the first threshold, the system triggers the analysis process of potential collapse events. At this time, the system's topology propagation coherence verification mechanism and the dominant collapse mode analysis mechanism are executed concurrently. The synergy of these two mechanisms constitutes an orthogonal verification system. The consensus collapse acceleration captures the destructive trend of system stability in the time dimension, while the calculation of the topology propagation coherence index verifies the source of this trend in the spatial dimension. Through backtracking analysis, it is found that the spatial source of the state position changes that cause the acceleration to exceed the threshold is highly concentrated in the aforementioned specific group of units and within this group. The state changes of nodes and their physical neighbors in adjacent time steps exhibited sequential propagation characteristics, causing the topology propagation coherence index to exceed the second threshold. Meanwhile, the analysis of the dominant collapse mode determined that the nature of this event was mechanical stress-driven. This series of analyses separated the event from the background noise of wind disturbances across the entire field, confirming it as an endogenous system collapse. Finally, the central platform generated and issued a composite decision instruction, which not only included a disaster warning signal but also indicated that the dominant collapse mode was mechanical stress. Based on the consensus volatility peak of this event and the historical event knowledge base, it provided an estimated risk level. In accordance with this instruction, the operation and maintenance team directly focused inspection and disposal resources on this specific unit group, performing targeted de-icing and vibration re-inspection operations. Thus, the risk was identified and eliminated before a potential major mechanical failure caused by blade imbalance occurred. Meanwhile, the rest of the entire energy system maintained normal operation during this period without being disturbed by unnecessary shutdowns for inspection.

[0036] Example 2: To objectively verify the effectiveness of the method of the present invention in distinguishing between endogenous system collapse and synchronous external disturbance events, an experimental platform consisting of 100 software-simulated distributed edge computing nodes was built. These 100 nodes were arranged in a 10x10 two-dimensional grid in virtual space. Each node was assigned its coordinates in the grid as spatial location information and pre-stored the identifiers of its directly adjacent nodes. A central server served as the central platform, responsible for aggregating and processing the binary status words generated by all nodes within the discrete time step. The setting of key experimental parameters followed the following procedure: the time step was set to 100 milliseconds. This value was the result of a trade-off between ensuring sufficient sampling resolution for the millisecond-level mechanical or electrical fault evolution process and the data processing load of the control center platform. The calibration of the first and second thresholds was generated by applying a background signal containing only stationary Gaussian white noise to the platform for 1 hour before the experiment. The system recorded the statistical distribution of consensus collapse acceleration and topology propagation coherence index during this period, and took the values ​​at the 99.9% quantile of their probability density function as the judgment benchmark.

[0037] The experiment included two control cases. Case A simulated the collapse of an endogenous system, which was set up as follows: At time 1, the node at the center of the grid experiences a state bit flip due to a simulated fault. In every subsequent time step, this fault propagates to all its neighboring nodes that have not yet failed, following a Manhattan distance of 1. Case B simulates a synchronicity-related external disturbance event, which is set to occur at... At a certain time, 50 nodes are randomly selected from 100 nodes, causing them to simultaneously flip their state bits. During the experiment, the two cases exhibited significantly different data characteristics. In case A, the consensus volatility started at 0 and gradually accelerated as the fault propagated layer by layer, with the corresponding consensus collapse acceleration consistently showing a positive value. In case B, the consensus volatility... The instantaneous jump from 0 to 50 generated a consensus collapse acceleration pulse with a large value but lasting only one time step. Subsequently, the acceleration immediately returned to zero. Table 1 shows excerpts of monitoring data at key time points in the two cases.

[0038] Table 1: Comparison of key data between Case A and Case B.

[0039]

[0040] The data in Table 1 reflects the inherent discrimination mechanism of the method of this invention. In Case A, when the consensus collapse acceleration continuously exceeds the first threshold, the topology propagation coherence index of the system's concurrent computation remains above 0.8 due to the high spatiotemporal correlation between the state-changing nodes and their neighboring nodes, exceeding the second threshold. Therefore, the system is judged to be an endogenous system collapse. In Case B, although the consensus collapse acceleration is... The pulse value at any given moment is enormous, but because the 50 nodes undergoing state changes are randomly distributed in space, the average proportion of their direct neighbors also experiencing state changes is extremely low. This results in a topology propagation coherence index of only 0.08, which does not reach the second threshold. Consequently, the system does not trigger a collapse warning but instead classifies it as a synchronous external disturbance event. The data from this experiment confirms that by concurrently verifying the temporal evolution rate of system state changes, i.e., consensus collapse acceleration, and the spatial topology structure, i.e., the topology propagation coherence index, the method disclosed in this invention can effectively distinguish between endogenous fault chains with physical propagation characteristics and global synchronous noise that does not possess such characteristics, thereby improving the accuracy of disaster early warning decision-making.

[0041] Example 3: This example combines Figures 1 to 3 This section explains the methods for disaster prediction and assessment in the energy industry based on multi-source data, such as... Figure 1 As shown, the complete logical chain from data processing to decision output begins with distributed edge nodes processing local data and fixed rules to generate binary status words. The central platform then aggregates the status words from all nodes to form a global state vector sequence, and performs core indicator calculations to obtain consensus volatility and consensus collapse acceleration. The core of the process lies in a two-stage judgment mechanism: first, it determines whether the consensus collapse acceleration continuously exceeds a preset threshold; otherwise, it returns to the monitoring loop; if so, it concurrently initiates topology propagation coherence verification (analyzing the spatial propagation coherence of state-changing nodes) and dominant collapse mode analysis (statistically counting the domains to which the changed rules belong). The system classifies domains and then makes a secondary judgment based on whether the topology propagation coherence index exceeds a threshold. If the index exceeds the threshold, it is confirmed as an endogenous system collapse, triggering a disaster warning containing diagnostic information. At the same time, it combines historical knowledge base to predict the risk level and finally outputs a composite decision instruction to match and output an emergency response plan. Conversely, if the coherence index does not exceed the threshold, it is determined as a synchronous external disturbance, issuing a non-disaster alarm and suppressing adaptive calibration to prevent the system from being contaminated by abnormal data. In addition, an adaptive calibration module continuously monitors the background consensus volatility. When it deviates from the ideal state, it generates a global threshold adjustment factor to dynamically optimize the rules of edge nodes, forming a feedback loop.

[0042] like Figure 2As shown, the horizontal axis represents the background consensus volatility, and the vertical axis represents the probability density. The ideal distribution is depicted by the bold solid line. The gray dashed line represents the statistical benchmark of the system's volatility under healthy conditions; the gray dashed line depicts the actual distribution. Initially, it shows that the volatility distribution of the system, before calibration, has deviated from the ideal state due to equipment aging or environmental drift; while the gray dotted line depicts the actual distribution. After calibration, it is clearly shown that after the adaptive calibration mechanism takes effect, by applying a global threshold adjustment factor, the volatility distribution of the system is corrected and significantly moves towards the ideal distribution. This figure visually demonstrates that the method of the present invention can quantify the deviation of the system state and perform effective incremental self-calibration accordingly to maintain the long-term accuracy of monitoring.

[0043] like Figure 3 As shown, the process begins with equipment within the energy system. Its local data source is processed by Module 1.0's localized state specification based on D3: edge node rules and micro-heuristic logic rules stored in the location library, generating standardized binary state words. Module 2.0's global stability monitoring aggregates these state words and calculates the consensus collapse acceleration of key indicators. This acceleration triggers Module 3.0's spatiotemporal coupling verification. This module calls the spatial location information in D3 to perform topology analysis and transmits the verification results to Module 4.0 for disaster early warning and assessment. Module 4.0 outputs a disaster early warning based on the verification results. The system can either trigger a crash mode or output a non-disaster alert. After a disaster is confirmed, Module 5.0, Risk Rating and Response, is initiated. It connects to the D1: Historical Event Knowledge Base to obtain risk weight coefficients and historical loss information, performs risk rating on the current event, generates composite decision instructions, and can match corresponding plans from the D2: Emergency Response Plan Library. Finally, the instructions are delivered to the operations and maintenance team. Module 6.0, Adaptive Calibration, runs through the entire process. It analyzes the background consensus volatility output by Module 2.0, generates a global threshold adjustment factor, and feeds it back to the system front end to achieve closed-loop adjustment of the rules.

[0044] Example 4: During the initial deployment and long-term operation of the method of the present invention, the calibration of its core parameters and the construction of the knowledge base follow a set of internal procedures. When this procedure is applied to a newly commissioned transmission network, the challenge is: how to effectively initialize the risk weight coefficients and establish a self-improving historical event knowledge base in the absence of long-term historical operating data. To address this challenge, the system first performs a structured definition and initialization of the historical event knowledge base. This knowledge base consists of a series of structured event record objects. Each event record object corresponds to a past endogenous system collapse event and includes a unique event identifier, an event occurrence timestamp, and the dominant collapse mode determined by the system. Peak consensus volatility during the event And an actual level of damage determined manually after an investigation. During the initial deployment phase of the system, the knowledge base is populated with seed data by importing publicly available typical accident case data in the field or by running high-fidelity physical simulation models to simulate specific fault scenarios.

[0045] Risk weighting coefficient The determination is a dynamic statistical process based on this knowledge base. During the daily background maintenance cycle, the system traverses all valid historical event knowledge bases. Value event logging, and according to the dominant crash mode. Grouping, for each defined crash mode The system calculates all event records belonging to this group. and The ratios, and the arithmetic mean of these ratios are updated to the crash mode. Risk weighting coefficient at the current moment When a new endogenous system crash event occurs, the system utilizes the current... With this incident Calculate the estimated risk level awaiting the new event Once manually tagged and added to the knowledge base, its data will be incorporated into calculations in the next maintenance cycle to achieve [the desired outcome]. The value is dynamically corrected; in parallel, the judgment threshold in the micro-heuristic logic rules of the edge nodes is adaptively calibrated. The process follows the calculation procedure as follows: During normal operation, the central platform fits the probability density distribution of the collected background consensus volatility sequence with a statistical period of 24 hours to obtain the actual volatility distribution of the current period. The system also pre-stores a baseline distribution representing an ideal state of health. The system periodically calculates the Gibbs relative entropy between these two probability distributions. The entropy value quantifies the deviation of the actual system state from the ideal state. A global threshold adjustment factor is defined as a function that is linearly proportional to the entropy value and is applied to the judgment threshold of all edge nodes for uniform incremental adjustment. This procedure ensures that the adjustment of the threshold is directly related to the quantification drift of the overall system state.

[0046] Example 5: When deploying the method of this invention for the first time in a geothermal power plant lacking prior fault data, the system needs to execute a standardized on-site calibration procedure. This procedure first requires the operations engineer to identify the key potential failure modes of the power plant in conjunction with the equipment design specifications. Subsequently, under controlled conditions, small perturbations simulating early failures are applied to the equipment, and high-frequency sensor data is collected simultaneously. Offline data analysis is used to establish a mapping relationship between specific sensor parameters and early failure symptoms, and these mapping relationships are solidified into micro-heuristic logic rules applicable to the geothermal power plant. After completing the configuration of the above rules and... After being deployed across all distributed edge computing nodes, the entire system will enter a 72-hour background data learning and operation phase. During this period, the geothermal power plant will maintain its most stable and typical normal operating conditions. The central platform will continuously record and statistically analyze the aggregated consensus volatility sequence and generate an initial ideal distribution range characterizing the health status of the specific plant based on this dataset. After the learning phase ends, the system will switch to formal operation mode and undergo final verification by executing a set of pre-set non-destructive disturbance tests. The initial deployment of the system will be complete when the system output matches the expected results of the disturbance tests.

[0047] Example 6: To determine the operating point of several core control parameters in the method of the present invention, the system executes an offline simulation optimization procedure before formal delivery. This procedure uses historical data or simulation models to generate a large number of endogenous system collapse event sequences and pure background noise sequences with different evolution rates. The purpose is to balance the immediacy of detection with immunity to instantaneous noise. For the two parameters of predetermined duration and backtracking time window, the procedure scans different parameter combinations within a preset range and evaluates the system performance under each combination. The evaluation indicators include the average detection delay time for the collapse event sequence and the false trigger rate for the background noise sequence. Finally, based on the preset requirements of detection immediacy and warning confidence for specific application scenarios, an operating point is selected from the Pareto optimal boundary of the response surface formed by these evaluation indicators, and the parameter combination corresponding to the operating point is used as the fixed configuration for the scenario.

[0048] After the system is put into online operation, in order to cope with the intermittent data interruption caused by communication failures or failures of distributed edge computing nodes, the system integrates an online fault tolerance mechanism. The central platform monitors the data upload status of each edge computing node. When one or more nodes fail to upload their binary status words within a preset timeout threshold, the central platform marks these missing binary bits as invalid in the global state vector of the current time step. When calculating the Hamming distance between the global state vectors of adjacent time steps, all bits marked as invalid and their corresponding bits in the previous time step are not included in the distance calculation. Furthermore, the final calculated Hamming distance value is normalized according to the total number of currently valid nodes. This mechanism can avoid abrupt artifacts in consensus volatility and consensus collapse acceleration caused by the missing data of some nodes, thereby ensuring the effectiveness of the overall system stability monitoring.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for disaster prediction and assessment in the energy industry based on multi-source data, characterized in that, Includes the following steps: Step a: Deploy multiple distributed edge computing nodes within the energy system and pre-store their respective spatial location information. Based on their respective local data sources and a set of logically fixed micro-heuristic logic rules stored within the edge computing nodes, generate binary state words representing local states independently within each predetermined time step. Step b: On the central platform, binary status words generated by multiple distributed edge computing nodes within a predetermined time step are aggregated to form a global status vector sequence. Step c: Based on the global state vector sequence, calculate the Hamming distance between global state vectors of adjacent time steps to obtain a consensus volatility sequence, and calculate its first-order difference in time based on the consensus volatility sequence to obtain the consensus collapse acceleration. Step d: When the consensus collapse acceleration exceeds a first threshold within a predetermined duration, the following steps are executed concurrently: Step d1: Obtain the global state vector sequence within a backtracking time window that caused the consensus collapse acceleration to exceed the first threshold, and combine it with the spatial location information of the edge computing nodes to calculate the propagation coherence of the edge computing nodes whose states changed within the backtracking time window in the spatial topology, and obtain the topology propagation coherence index; Step d2: Based on the physical attributes of the local data source associated with the micro-heuristic logic rules, classify the micro-heuristic logic rules into domains in advance, and count the domain classifications of the binary bits that changed during the consensus collapse acceleration exceeding the first threshold, to determine a dominant collapse mode; Step e: When the topology propagation coherence index exceeds the second threshold, the event is confirmed as an endogenous system collapse, and a disaster warning containing the dominant collapse mode as diagnostic information is triggered. Among them, the micro-heuristic logic rule in step a is a fixed logic judgment set based on the safety operation limit of the design specifications of the equipment in the energy system, or based on the empirical limit derived from the statistical analysis of historical safety operation data. In addition, step d1 calculates the topology propagation coherence index, which specifically includes: identifying edge computing nodes that undergo state changes within the backtracking time window; calculating the average proportion of the nodes that undergo state changes within the backtracking time window whose direct neighbor nodes in the preset physical space also undergo state changes in adjacent time steps, and using this average proportion as the topology propagation coherence index.

2. The energy industry disaster prediction and assessment method based on multi-source data according to claim 1, characterized in that, The method also includes an adaptive calibration step for the judgment threshold of the micro-heuristic logic rules. The steps include: on the central platform, statistical analysis of the consensus volatility sequence within a preset statistical period during the operation period when no disaster warning is triggered, to obtain the statistical distribution characteristics of the background consensus volatility; comparing the statistical distribution characteristics of the background consensus volatility with a stored ideal distribution interval that characterizes the health status of the system; when the statistical distribution characteristics of the background consensus volatility exceed the ideal distribution interval, generating a global threshold adjustment factor and applying it to the micro-heuristic logic rules of all edge computing nodes to incrementally adjust their judgment thresholds.

3. The energy industry disaster prediction and assessment method based on multi-source data according to claim 1, characterized in that, The domain classification of the micro-heuristic logic rules in step d2 includes classifying rules related to wind speed and air pressure into the meteorological stress domain, rules related to equipment vibration and structural deformation into the mechanical stress domain, and rules related to current, voltage, and insulation parameters into the electrical anomaly domain.

4. The energy industry disaster prediction and assessment method based on multi-source data according to claim 1, characterized in that, The method also includes: after confirming that the event is an endogenous system crash, recording the dominant crash mode of the event and the peak value of the consensus volatility during the event; associating and matching the dominant crash mode and the peak value of the consensus volatility with a historical event knowledge base that stores the mapping relationship between parameters of historical endogenous system crash events and the actual loss level determined after the fact; and determining and outputting the estimated risk level of the current endogenous system crash event based on the association matching results.

5. The energy industry disaster prediction and assessment method based on multi-source data according to claim 4, characterized in that, Estimated risk level The determination of is based on the following calculation rules: ,in, This represents the peak value of consensus volatility in the current endogenous system crash event. The dominant crash pattern for the current event. To extract data from a historical event knowledge base based on the dominant crash pattern Query and extract the risk weight coefficients.

6. The energy industry disaster prediction and assessment method based on multi-source data according to claim 1, characterized in that, The method also includes: when the consensus collapse acceleration exceeds a first threshold within a predetermined duration, but the topology propagation coherence index does not exceed a second threshold, the event is judged as a synchronous external disturbance event; when judged as a synchronous external disturbance event, a non-disaster alert indicating an abnormal external environment is issued, and any adaptive calibration action on the micro-heuristic logic rules is suppressed.

7. The method for disaster prediction and assessment in the energy industry based on multi-source data according to claim 1, characterized in that, The central platform in step b is a central server or cloud computing platform that has the computational capability to process global state vector sequences and stores the first threshold and the second threshold.

8. The method for disaster prediction and assessment in the energy industry based on multi-source data according to claim 1, characterized in that, The method also includes combining disaster warning, dominant collapse mode and estimated risk level into a composite decision instruction, and matching and outputting an emergency response plan from the emergency response plan library based on the composite decision instruction.

Images