Dynamic planning and abnormity identification method for intelligent inspection path of thermal power plant

Abstract

The invention discloses a thermal power plant intelligent inspection path dynamic planning and anomaly recognition method, and particularly relates to the technical field of thermal power plants, and the method comprises the following steps: S1, constructing a dynamic three-dimensional risk thermodynamic diagram and an equipment health gene code; s2, multi-target dynamic path planning; s3, multi-modal data collaborative acquisition and anomaly detection are carried out; s4, performing abnormal root cause association and risk level judgment; s5, re-planning a closed-loop path of abnormal driving; and S6, executing and dynamically updating. According to the invention, through a core technology architecture of dynamic risk thermodynamic diagram-equipment health gene code-multi-modal anomaly detection-closed-loop path planning, full-process optimization of intelligent inspection of the thermal power plant from risk perception, path decision-making to anomaly handling is realized, and the inspection efficiency, the equipment reliability and the emergency response capability are significantly improved; meanwhile, risks of man-made misjudgment and system failure are reduced, and core technical support is provided for intelligent transformation of the thermal power plant.

Description

Technical Field

[0001] This invention relates to the field of thermal power plant technology, and more specifically, to a method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants. Background Technology

[0002] As the core infrastructure for energy supply, the stability of thermal power plant equipment operation is directly related to the security of power supply and production efficiency. With the expansion of equipment capacity (such as the popularization of megawatt-class units) and the increase in process complexity (such as supercritical / ultra-supercritical boilers and steam turbine systems), the cascading risks of equipment failures have increased significantly.

[0003] Traditional inspection methods rely on regular manual checks, which have the following prominent problems:

[0004] 1. Existing intelligent inspection systems mostly adopt fixed routes or path planning based on a single rule (such as equipment importance ranking), which cannot dynamically respond to changes in equipment status. For example, when the water-cooled wall tubes of an old boiler are at increased risk of corrosion, the traditional system still inspects them according to a fixed cycle, which may lead to missed inspection of high-risk areas, while the inspection frequency of key equipment (such as turbine bearings) may be excessive, wasting resources.

[0005] 2. Traditional risk assessment relies on current equipment operating parameters (such as temperature and vibration), lacking comprehensive quantification of environmental risks (such as gas leakage and diffusion, and heat radiation) and historical deterioration trends. For example, if a local high temperature is caused by valve leakage in a certain area, but the traditional system only monitors it through point temperature sensors, it cannot provide early warning of secondary risks caused by gas diffusion. At the same time, historical equipment failure data is not linked to the real-time risk model, leading to the contradiction of "over-inspection of new equipment and insufficient inspection of old equipment".

[0006] 3. Existing anomaly detection methods mostly rely on data from a single sensor (such as infrared thermal imaging or vibration spectrum), which are easily affected by environmental interference (such as dust obscuring the infrared lens or background noise masking acoustic anomalies), leading to false alarms. For example, if a bearing experiences excessive vibration due to poor lubrication, the traditional system may misjudge it as "mechanical loosening" because it does not correlate with temperature data (lubrication failure is usually accompanied by temperature rise). At the same time, root cause analysis relies on human experience and lacks structured tools such as knowledge graphs, resulting in a long time consumption for troubleshooting. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art, this invention provides a method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants, comprising the following steps:

[0009] S1. Construct a dynamic 3D risk heat map and equipment health genotype: Based on real-time data from an IoT sensor array and a digital twin model of the equipment, generate a spatial-temporal-risk 3D heat map, integrating safety risks, equipment failure risks, and environmental risks into the risk dimensions; simultaneously construct an equipment health genotype, with encoded fields including installation process deviation values, cumulative equivalent operating hours, historical similar failure counts, and local microenvironment corrosion coefficients; establish a two-way feedback mechanism.

[0010] a. Real-time risk values ​​from a three-dimensional space-time-risk heatmap drive the weighting of device health gene codes;

[0011] b. Equipment failure risk model based on dynamic correction of space-time-risk three-dimensional heat map using equipment health genomics;

[0012] S2. Multi-objective dynamic path planning: Using the heat map and gene code as input, an improved multi-objective optimization algorithm is used to generate inspection paths. The optimization objectives include minimizing risk exposure, maximizing the coverage priority of key equipment, and optimizing inspection efficiency.

[0013] S3. Multimodal data collaborative acquisition and anomaly detection: Control the inspection equipment to move along the path, synchronously acquire multimodal data and input it into the fusion analysis model, and output the abnormal status and confidence level of the equipment;

[0014] S4. Anomaly Root Cause Association and Risk Level Determination: Based on the knowledge graph, the anomaly state is associated with the potential root cause of the failure, and the risk level is calculated in combination with the real-time parameters of the equipment.

[0015] S5. Anomaly-driven closed-loop path replanning: If the risk level exceeds the threshold, path replanning is triggered to generate a new path with the core objectives of anomaly point confirmation, impact range assessment, and high-risk area avoidance.

[0016] S6, Execution and Dynamic Update: Controls the inspection equipment to execute new paths and feeds back the collected data to S1 to update the risk weight value of the heat map.

[0017] As a further improvement to the technical solution of the present invention, the execution of the bidirectional feedback mechanism in S1 includes:

[0018] When the equipment risk value remains abnormal, the weight of the historical failure count in the genetic code is automatically adjusted.

[0019] When the corrosion coefficient in the genetic code exceeds the limit, the area of ​​environmental risk diffusion is marked on the heat map;

[0020] After the genetic code is updated, the equipment failure risk items in the heat map are recalculated.

[0021] As a further improvement to the technical solution of the present invention, the improvements to the improved multi-objective optimization algorithm in S2 include:

[0022] The device health genomic code is introduced as a dynamic weighting factor, and the genomic code field is quantified in the following way:

[0023] a. Installation process deviation: calculated by comparing 3D scanning with the digital model;

[0024] b. Cumulative equivalent operating hours: calculated based on load spectrum integration;

[0025] The optimization objective meets the preset conditions:

[0026] a. The path risk exposure value does not exceed the preset proportion of equipment downtime losses;

[0027] b. The coverage density of critical equipment is not lower than the relevant threshold for the failure rate level;

[0028] c. The ratio of the path length to the ideal shortest path does not exceed the preset upper limit.

[0029] As a further improvement to the technical solution of the present invention, the multimodal data collaborative acquisition and anomaly detection in S3 includes:

[0030] The inspection equipment collects the following data:

[0031] a. Visible light / infrared images;

[0032] b. Acoustic fingerprint data;

[0033] c. Vibration spectrum;

[0034] d. Gas concentration data;

[0035] A fusion analysis model is used to process multimodal data. The model architecture includes a spatial feature extraction layer, a temporal modeling layer, and a cross-modal fusion layer.

[0036] As a further improvement to the technical solution of the present invention, the abnormal root cause association and risk level determination in S4 includes:

[0037] Construct a knowledge graph that includes equipment types, failure modes, and process nodes;

[0038] Candidate root causes are matched by probability calculation, and the probability values ​​are adjusted by combining historical device data.

[0039] The risk level is divided into three levels, and different response mechanisms are triggered based on the anomaly score.

[0040] As a further improvement to the technical solution of the present invention, the dynamic path replanning decision in S5 includes:

[0041] Mark the risk spread area according to the risk level;

[0042] Generate a new path that includes elements such as anomaly detection, associated device scanning, and emergency evacuation.

[0043] Task instructions are sent to the inspection equipment via a dedicated network, and multimodal positioning and navigation are enabled.

[0044] As a further improvement to the technical solution of the present invention, the risk weight update rule in S6 is as follows:

[0045] If the anomaly confirmation result is a false alarm, the risk weight coefficient for that equipment type will be lowered.

[0046] If the equipment that was missed malfunctions, the risk value of the associated area will be amplified;

[0047] The updated risk weight values ​​are synchronized to the inspection equipment in real time, triggering local path adjustments.

[0048] As a further improvement to the technical solution of the present invention, if a communication interruption occurs during the execution of S6, a resilience inspection strategy is initiated. The resilience inspection strategy includes:

[0049] A safe corridor (width ≥ 2m) is generated based on the last received valid heat map;

[0050] Switch to acoustic-vibration coupled obstacle sensing mode;

[0051] Perform inspections according to the list of life-saving devices in the device's health genomic code.

[0052] The beneficial effects of this invention are:

[0053] This invention, through its core technical architecture of dynamic risk heatmap, equipment health genomics, multimodal anomaly detection, and closed-loop path planning, achieves end-to-end optimization of intelligent inspection in thermal power plants, from risk perception and path decision-making to anomaly handling. Through the synergy of five key technologies—dynamic risk quantification, intelligent path planning, precise anomaly detection, rapid root cause localization, and system robustness enhancement—it achieves a leap in thermal power plant inspection from "passive and periodic" to "proactive and dynamic," from "single detection" to "root cause closed-loop," and from "fixed strategies" to "personalized adaptation." This significantly improves inspection efficiency, equipment reliability, and emergency response capabilities, while reducing the risk of human error and system failure, providing core technical support for the intelligent transformation of thermal power plants. Attached Figure Description

[0054] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] As attached Figure 1 The method for dynamic planning and anomaly identification of intelligent inspection paths in thermal power plants, as shown, includes the following steps:

[0057] S1. Construct a dynamic 3D risk heat map and equipment health genotype: Based on real-time data from an IoT sensor array and a digital twin model of the equipment, generate a spatial-temporal-risk 3D heat map, integrating safety risks, equipment failure risks, and environmental risks into the risk dimensions; simultaneously construct an equipment health genotype, with encoded fields including installation process deviation values, cumulative equivalent operating hours, historical similar failure counts, and local microenvironment corrosion coefficients; establish a two-way feedback mechanism.

[0058] a. Real-time risk values ​​from a three-dimensional space-time-risk heatmap drive the weighting of device health gene codes;

[0059] b. Equipment failure risk model based on dynamic correction of space-time-risk three-dimensional heat map using equipment health genomics;

[0060] S2. Multi-objective dynamic path planning: Using the heat map and gene code as input, an improved multi-objective optimization algorithm is used to generate inspection paths. The optimization objectives include minimizing risk exposure, maximizing the coverage priority of key equipment, and optimizing inspection efficiency.

[0061] S3. Multimodal data collaborative acquisition and anomaly detection: Control the inspection equipment to move along the path, synchronously acquire multimodal data and input it into the fusion analysis model, and output the abnormal status and confidence level of the equipment;

[0062] S4. Anomaly Root Cause Association and Risk Level Determination: Based on the knowledge graph, the anomaly state is associated with the potential root cause of the failure, and the risk level is calculated in combination with the real-time parameters of the equipment.

[0063] S5. Anomaly-driven closed-loop path replanning: If the risk level exceeds the threshold, path replanning is triggered to generate a new path with the core objectives of anomaly point confirmation, impact range assessment, and high-risk area avoidance.

[0064] S6, Execution and Dynamic Update: Controls the inspection equipment to execute new paths and feeds back the collected data to S1 to update the risk weight value of the heat map.

[0065] The execution of the two-way feedback mechanism in S1 includes:

[0066] When the equipment risk value remains abnormal, the weight of the historical failure count in the genetic code is automatically adjusted.

[0067] When the corrosion coefficient in the genetic code exceeds the limit, the area of ​​environmental risk diffusion is marked on the heat map;

[0068] After the genetic code is updated, the equipment failure risk items in the heat map are recalculated.

[0069] The traditional proportional risk model takes the following form:

[0070]

[0071] Where h(t) is the risk function, h0(t) is the baseline risk (group level), and x i For covariates (such as equipment age, operating time), β i It is a fixed coefficient.

[0072] The revised model uses the genetic code field as a dynamic covariate and introduces time dependence and individual specificity, in the form of:

[0073]

[0074] in:

[0075] D install The installation process deviation value (standardized to 0-1, where 0 represents no deviation and 1 represents the maximum permissible deviation) is used as an "inherent risk factor". Its coefficient β1(t) changes with time (e.g., the deviation has a large impact in the early stage of equipment operation, and the impact weakens in the later stage due to wear covering the initial deviation).

[0076] H eq (t) / H design The ratio of cumulative equivalent operating hours to design life (reflecting the proportion of operating intensity; a value exceeding 1 indicates "overload"), the coefficient β2(t) increases significantly as the equipment approaches its design life (the risk of failure increases exponentially);

[0077] N fault (t) represents the number of similar failures in history (dynamically updated, such as incrementing by 1 if a new failure is added), and the coefficient β3(t) increases with the number of failures (the more failures there are, the stronger the "failure inheritance tendency").

[0078] C env (t) is the local microenvironment corrosion coefficient (updated in real time by IoT sensors, such as humidity sensors and corrosion rate sensors). The coefficient β4(t) is related to environmental parameters (e.g., β4 increases under high humidity, accelerating material degradation).

[0079] The improvements to the improved multi-objective optimization algorithm in S2 include:

[0080] (1) Introduce the device health gene code as a dynamic weighting factor. The gene code field is quantified in the following way:

[0081] a. Installation process deviation: calculated by comparing 3D scanning with the digital model;

[0082] b. Cumulative equivalent operating hours: calculated based on load spectrum integration;

[0083] The optimization objective meets the preset conditions:

[0084] a. The path risk exposure value does not exceed the preset proportion of equipment downtime losses;

[0085] b. The coverage density of critical equipment is not lower than the relevant threshold for the failure rate level;

[0086] c. The ratio of the path length to the ideal shortest path does not exceed the preset upper limit.

[0087] Among them, the multimodal data collaborative acquisition and anomaly detection in S3 include:

[0088] The inspection equipment collects the following data:

[0089] a. Visible light / infrared images;

[0090] b. Acoustic fingerprint data;

[0091] c. Vibration spectrum;

[0092] d. Gas concentration data;

[0093] A fusion analysis model is used to process multimodal data. The model architecture includes a spatial feature extraction layer, a temporal modeling layer, and a cross-modal fusion layer.

[0094] Among them, the abnormal root cause association and risk level determination in S4 include:

[0095] Construct a knowledge graph that includes equipment types, failure modes, and process nodes;

[0096] Candidate root causes are matched by probability calculation, and the probability values ​​are adjusted by combining historical device data.

[0097] The risk level is divided into three levels, and different response mechanisms are triggered based on the anomaly score.

[0098] The dynamic path replanning decision in S5 includes:

[0099] Mark the risk spread area according to the risk level;

[0100] Generate a new path that includes elements such as anomaly detection, associated device scanning, and emergency evacuation.

[0101] Task instructions are sent to the inspection equipment via a dedicated network, and multimodal positioning and navigation are enabled.

[0102] The risk weight update rule in S6 is as follows:

[0103] If the anomaly confirmation result is a false alarm, the risk weight coefficient for that equipment type will be lowered.

[0104] If the equipment that was missed malfunctions, the risk value of the associated area will be amplified;

[0105] The updated risk weight values ​​are synchronized to the inspection equipment in real time, triggering local path adjustments.

[0106] Furthermore, if a communication interruption occurs during S6 execution, a resilience inspection strategy will be initiated. The resilience inspection strategy includes:

[0107] A safe corridor (width ≥ 2m) is generated based on the last received valid heat map;

[0108] Switch to acoustic-vibration coupled obstacle sensing mode;

[0109] Perform inspections according to the list of life-saving devices in the device's health genomic code.

[0110] In summary, this invention, through its core technical architecture of dynamic risk heatmap, equipment health genomics, multimodal anomaly detection, and closed-loop path planning, achieves end-to-end optimization of intelligent inspection processes in thermal power plants, from risk perception and path decision-making to anomaly handling. The specific technical effects can be summarized as follows:

[0111] 1. The precision and dynamism of risk assessment

[0112] Dynamic 3D Risk Heat Map: By integrating real-time sensor data (temperature, vibration, gas concentration), environmental models (CFD diffusion simulation), and historical fault databases, a spatial-temporal-risk 3D distribution model is constructed. This solves the problem of blindness in traditional periodic inspections with "fixed cycles and fixed routes," quantifies risk distribution in real time, automatically marks high-risk areas, and prioritizes inspection resources. Environmental risks (such as gas diffusion and thermal radiation) are dynamically visualized, avoiding errors in human experience-based judgment.

[0113] Equipment health genomics: By using fields such as installation process deviations, cumulative operating hours, historical failure counts, and micro-environment corrosion coefficients, a "digital health record" is built for each piece of equipment, enabling personalized risk assessment at the equipment level (such as automatically increasing the inspection priority for old equipment); and dynamically correcting the risk model (such as automatically amplifying the environmental risk weight when the corrosion coefficient exceeds the limit), avoiding "one-size-fits-all" management.

[0114] 2. The intelligence and adaptability of path planning

[0115] Multi-objective optimization algorithm: Taking heat map and genetic code as input, it balances three conflicting objectives: "minimizing risk exposure", "maximizing coverage of key equipment" and "optimizing inspection efficiency", avoiding the defects of single-objective optimization (such as pursuing efficiency alone may result in missing high-risk equipment, and focusing only on risk may result in path redundancy); dynamic weight adjustment (such as automatically increasing the path priority of equipment with a high number of historical failures) to achieve precise resource allocation.

[0116] Closed-loop path replanning: When the abnormal risk level exceeds the threshold, a new path is automatically generated with "abnormal point confirmation - impact range assessment - high-risk area avoidance" as the core, responding to abnormalities in real time (such as automatically bypassing the diffusion area in the event of a gas leak), shortening the emergency response time; avoiding the "lag" of fixed paths, ensuring that high-risk areas are inspected first, and reducing ineffective inspections in low-risk areas.

[0117] 3. Reliability of anomaly detection and efficiency of root cause localization

[0118] Multimodal data fusion analysis: Simultaneously collect multimodal data such as visual, infrared, acoustic, vibration, and gas concentration data, extract related features through spatiotemporal residual networks and cross-modal attention mechanisms; cross-validate anomalies (such as when infrared hot spots and acoustic anomaly frequency bands coincide in spatiotemporal, to improve anomaly confidence) and reduce the false alarm rate of single sensors; dominate the output of anomaly modal labels (such as "vibration + infrared" combined anomalies) to provide direction for root cause analysis.

[0119] Knowledge graph root cause association: Construct a knowledge graph of equipment type, failure mode, and process chain relationship, and calculate the probability of candidate root causes by combining real-time parameters to quickly locate the root cause of failure (e.g., "abnormal bearing vibration + temperature rise" is associated with "lubrication failure"), reducing troubleshooting time; combine the maintenance history coefficient in the equipment gene code to correct the probability (e.g., automatically increase the weight of "aging" root cause for equipment with multiple failures) to improve diagnostic accuracy.

[0120] 4. System robustness and industrial adaptability

[0121] Communication interruption resilience strategy: When communication is interrupted, a safety corridor is generated based on the last valid heat map, the system switches to acoustic-vibration coupled sensing mode, and performs limited target inspections according to the list of life-saving equipment (such as main steam pipelines and emergency cooling valves) to avoid complete system failure and ensure basic safety functions (such as critical equipment status confirmation and emergency passage inspection); acoustic-vibration mode matching (pre-stores 200 typical obstacle features) ensures that the environment can still be sensed under low bandwidth.

[0122] Adaptation to special environments in thermal power plants: Deploy intrinsically safe sensors (such as explosion-proof gas sensors) and explosion-proof inspection robots (six-degree-of-freedom robotic arms) to support high dust and strong electromagnetic interference environments, meeting the reliability requirements of thermal power plants in "high temperature, high pressure, flammable and explosive" scenarios; the robotic arms are adapted to complex equipment detection (such as valve status confirmation in narrow spaces), expanding the inspection range.

[0123] 5. The overall system's closed-loop iteration and self-optimization capabilities

[0124] The "heatmap-gene code-path" closed-loop mechanism updates the risk weights of the heatmap after the path is executed, and the gene code is dynamically adjusted according to abnormal results, forming a continuous iteration of "perception-decision-execution-optimization". The system is gradually optimized as the operating data accumulates (such as automatically lowering the risk weight of false alarm equipment and amplifying the risk of related areas for missed detection equipment), achieving the adaptive capability of "becoming more accurate with use"; avoiding the "model rigidity" problem of traditional static systems, and adapting to dynamic scenarios such as equipment aging and process changes.

[0125] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants, characterized in that, Includes the following steps: S1. Construct a dynamic three-dimensional risk heat map and equipment health genomics: Based on real-time data from IoT sensor arrays and digital twin models of equipment, generate a spatial-temporal-risk three-dimensional heat map, integrating safety risks, equipment failure risks, and environmental risks in the risk dimensions; Simultaneously construct the equipment health gene code, with coded fields including installation process deviation value, cumulative equivalent operating hours, number of similar historical failures, and local micro-environment corrosion coefficient; Establish a two-way feedback mechanism: a. Real-time risk values ​​from a three-dimensional space-time-risk heatmap drive the weighting of device health gene codes; b. Equipment failure risk model based on dynamic correction of space-time-risk three-dimensional heat map using equipment health genomics; S2. Multi-objective dynamic path planning: Using the heat map and gene code as input, an improved multi-objective optimization algorithm is used to generate inspection paths. The optimization objectives include minimizing risk exposure, maximizing the coverage priority of key equipment, and optimizing inspection efficiency. S3. Multimodal data collaborative acquisition and anomaly detection: Control the inspection equipment to move along the path, synchronously acquire multimodal data and input it into the fusion analysis model, and output the abnormal status and confidence level of the equipment; S4. Anomaly Root Cause Association and Risk Level Determination: Based on the knowledge graph, the anomaly state is associated with the potential root cause of the failure, and the risk level is calculated in combination with the real-time parameters of the equipment. S5. Anomaly-driven closed-loop path replanning: If the risk level exceeds the threshold, path replanning is triggered to generate a new path with the core objectives of anomaly point confirmation, impact range assessment, and high-risk area avoidance. S6, Execution and Dynamic Update: Controls the inspection equipment to execute new paths and feeds back the collected data to S1 to update the risk weight value of the heat map.

2. The method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants according to claim 1, characterized in that: The execution of the bidirectional feedback mechanism in S1 includes: (1) When the equipment risk value is continuously abnormal, the weight of the number of historical failures in the gene code is automatically adjusted; (2) When the corrosion coefficient in the gene code exceeds the limit, mark the area of ​​environmental risk diffusion in the heat map; (3) After the gene code is updated, the equipment failure risk item of the heat map is recalculated.

3. The method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants according to claim 1, characterized in that: The improvements to the improved multi-objective optimization algorithm in S2 include: (1) Introduce the device health gene code as a dynamic weighting factor. The gene code field is quantified in the following way: a. Installation process deviation: calculated by comparing 3D scanning with the digital model; b. Cumulative equivalent operating hours: calculated based on load spectrum integration; (2) The optimization objective meets the preset conditions: a. The path risk exposure value does not exceed the preset proportion of equipment downtime losses; b. The coverage density of critical equipment is not lower than the relevant threshold for the failure rate level; c. The ratio of the path length to the ideal shortest path does not exceed the preset upper limit.

4. The method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants according to claim 1, characterized in that: The multimodal data collaborative acquisition and anomaly detection in S3 include: (1) Control the inspection equipment to collect the following data: a. Visible light / infrared images; b. Acoustic fingerprint data; c. Vibration spectrum; d. Gas concentration data; (2) A fusion analysis model is used to process multimodal data. The model architecture includes a spatial feature extraction layer, a temporal modeling layer, and a cross-modal fusion layer.

5. The method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants according to claim 1, characterized in that: The abnormal root cause association and risk level determination in S4 include: (1) Construct a knowledge graph that includes equipment types, failure modes, and process nodes; (2) Match candidate root causes by probability calculation and adjust the probability value by combining historical data of the equipment; (3) The risk level is divided into three levels, and different response mechanisms are triggered according to the anomaly score.

6. The method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants according to claim 1, characterized in that: The dynamic path replanning decision in S5 includes: (1) Mark the risk diffusion area according to the risk level; (2) Generate a new path that includes elements such as anomaly detection, associated device scanning, and emergency evacuation; (3) Send task instructions to the inspection equipment through a dedicated network and enable multimodal positioning and navigation.

7. The method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants according to claim 1, characterized in that: The risk weight update rule in S6 is as follows: (1) If the abnormality confirmation result is a false alarm, the risk weight coefficient of the equipment type shall be reduced; (2) If the equipment that was missed malfunctions, the risk value of the associated area will be increased; (3) The updated risk weight value is synchronized to the inspection equipment in real time, triggering local path adjustment.

8. The method for dynamic planning of intelligent inspection paths and anomaly identification in thermal power plants according to claim 1, characterized in that: If a communication interruption occurs during the execution of S6, a resilience inspection strategy will be initiated. The resilience inspection strategy includes: (1) Generate a safe corridor based on the last received valid heat map; (2) Switch to the acoustic-vibration coupled obstacle sensing mode; (3) Perform inspections according to the list of life-saving equipment in the equipment health gene code.

Images