An intelligent inspection system for power plant air cooling island

Abstract

The application discloses an intelligent inspection system for an air cooling island of a power plant, and belongs to the technical field of monitoring of air cooling islands of power plants, and comprises the following steps: acquiring abnormal inspection variables in real time, and performing classification processing to divide bottom-line variables and elastic variables, so that decision-making errors caused by fuzzy variable boundaries are avoided; a scheduling task set is constructed, a priority sorting method is set, the real-time urgency of the scheduling task is calculated, and priority sorting and conflict resolution are performed; based on a path anchor point topology graph, a candidate path set is generated, a candidate prediction method is set, the result of each candidate path is predicted, an inspection instruction is generated based on comprehensive scoring, a responsibility transfer mechanism or a reporting mechanism is triggered, and efficient execution of the scheduling task and timely response to risks are realized.

Description

TECHNICAL FIELD

[0001] The application relates to an intelligent inspection system for an air cooling island of a power plant, and belongs to the technical field of air cooling island monitoring of power plants. BACKGROUND

[0002] In a thermal power plant, the air cooling island, as a core heat dissipation device of the air cooling system, undertakes the important function of condensing turbine exhaust steam into water, and is usually composed of large-scale arrayed finned tube bundles, fan groups and supporting frames. Because the air cooling island occupies a wide area and is exposed to complex outdoor environments for a long time, the equipment state of the air cooling island affects the safe and efficient operation of the power plant. At present, the industry generally relies on a combination of regular manual climbing inspection and fixed-point video monitoring to implement inspection. However, this method has limited coverage and cannot achieve high-frequency, dead-angle-free fine observation.

[0003] At present, through mobile monitoring technology such as cable traction type mobile monitoring platform and track type inspection robot, temperature field mapping and pollution analysis are realized by carrying infrared thermal imagers and visual sensors.

[0004] However, the prior art does not consider path reconstruction in extreme emergency scenarios, especially in the case of sudden multivariate coupling scenarios during inspection, such as real-time detection of a sudden temperature rise in the target area during inspection, local damage to the track, and insufficient remaining power of the equipment. Specifically, in extreme emergency scenarios, variables are not isolated, the temperature increase rate of the target area directly affects the time window of the emergency response, such as a sudden temperature rise that shortens the optimal intervention time, the degree of track damage determines the safety boundary of the path, such as a severely deformed section that requires detouring to increase the path length, and the remaining power of the equipment limits the range of executable paths, such as low power that cannot support long-distance detouring. If the decision logic only focuses on a single dimension, it cannot dynamically adjust the priorities of each target based on real-time data, nor can it find a balance critical point when multiple targets conflict, such as a shortest path that needs to pass through a high-risk section or a safe path that exceeds the power support range of the equipment, resulting in a missed opportunity to handle the target optimally due to the mechanical adherence to a single rule or a task interruption due to the lack of consideration of equipment endurance, ultimately making it difficult to achieve efficiency in emergency response while ensuring system safety. SUMMARY

[0005] In view of the deficiencies of the prior art, the purpose of the present application is to provide an intelligent inspection system for an air cooling island of a power plant, which clearly defines the boundaries of classified baseline variables and flexible variables, quantifies the urgency and resource adaptation degree, schedules to prevent delays, generates an adaptive path and dynamically corrects it, and solves the problems of variable ambiguity, task delay, path inadaptation and information lag by combining a responsibility transfer mechanism and a reporting mechanism.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] An intelligent inspection system for air-cooled islands in power plants includes:

[0008] Real-time acquisition of abnormal inspection variables, classification and processing, and division into bottom-line variables and elastic variables;

[0009] Construct a set of scheduling tasks, set a priority ranking method, calculate the real-time urgency of the scheduling tasks, and perform priority ranking and conflict resolution. When scheduling tasks have priority conflicts, introduce resource adaptability to dynamically adjust the priority of the scheduling tasks.

[0010] Based on the path anchor point topology map, a candidate path set is generated, a candidate prediction method is set, the result prediction is performed for each candidate path, an inspection instruction is generated based on the comprehensive score, and a responsibility transfer mechanism or a reporting mechanism is triggered. When responsibility is transferred, the collaborative equipment pool is retrieved, the task is split and the connection path is planned. The reporting mechanism integrates information according to the real-time urgency level and pushes it to the responsible entity through multiple channels.

[0011] Specifically, the classification process includes:

[0012] Acquire real-time inspection data during the air-cooled island inspection process, and preprocess the data to generate an inspection dataset.

[0013] Anomaly identification is performed on the actual values ​​in the inspection dataset to generate a list of abnormal inspection variables. ;in, The number of abnormal inspection variables, For the first One abnormal inspection variable, The corresponding outlier is ;

[0014] For abnormal inspection variables Calculate abnormal change trends Configure the corresponding security parameters and adjustable coefficients, and set the first-level anomaly threshold. and Level 2 anomaly threshold At the same time, set the trend threshold. To determine the category of abnormal inspection variables;

[0015] like and If so, then the abnormal inspection variable is determined to be an elastic variable; if or If so, the abnormal inspection variable is determined to be a bottom-line variable;

[0016] Otherwise, continue collecting. The actual values ​​of the abnormal inspection variables mentioned above are then used to determine the category again. If they are continuous... The average does not meet the requirements. and determining that the variable is a bottom-line variable, otherwise, determining that the variable is a flexible variable.

[0017] Specifically, the prioritization method comprises:

[0018] Based on the mapping rule library, the abnormal inspection variable is converted into a scheduling task;

[0019] Based on the results of the classification processing, the scheduling task is layered, if there is a bottom-line variable in the scheduling task, it is classified as a bottom-line task, otherwise, it is a flexible task, and a layered task list is generated;

[0020] For the bottom-line task, the number of associated variables is obtained, the task coverage coefficient is set, the trend deterioration coefficient is calculated based on the ratio of the abnormal change trend to the trend threshold value, the highest basic urgency is combined for product operation, and the real-time urgency of the bottom-line task is generated;

[0021] For the flexible task, the risk reduction amount is calculated based on the difference between the safety parameter and the abnormal value, the resource consumption amount is obtained by calculating the weighted sum of the predicted processing time and the predicted processing energy consumption of the flexible variable, and the real-time urgency of the flexible task is generated by calculating the ratio of the risk reduction amount to the resource consumption amount and introducing the flexible change coefficient;

[0022] The scheduling tasks are sorted, the priority of the bottom-line task is higher than that of the flexible task, and the same type of scheduling task is arranged in descending order of real-time urgency, thereby generating a preliminary scheduling queue.

[0023] Specifically, the prioritization method further comprises:

[0024] The adaptation degree of the scheduling task is calculated by the difference between the current resource of the device and the resource demand of the task, and an adaptation threshold is set to determine whether to adjust the queue order of the current scheduling task;

[0025] If the adaptation degree is less than the adaptation threshold, it is marked as a resource-limited task, and the queue order of the corresponding task is lowered to the end of the same type of variable, and for multiple resource-limited tasks, they are sorted in descending order based on the adaptation degree; otherwise, the original order is retained;

[0026] Set the queue update interval, update the priority queue based on the real-time collected data, and generate a real-time scheduling queue.

[0027] Specifically, the step of generating a candidate path set comprises:

[0028] Obtain the real-time state of the device, construct an association matrix, and establish a path database in combination with the real-time scheduling queue and the results of the classification processing;

[0029] Set the bottom line variable as a mandatory anchor point, set the elastic variable as a mobile anchor point, call a task time threshold mapping table based on real-time urgency, match the time threshold for each anchor point, and generate an anchor point type division table;

[0030] Based on the track coordinate system, a path anchor point topology graph is constructed.

[0031] The safety weight of the track is calculated, and a track safety threshold is set to determine whether the current track is safe.

[0032] If the safety weight is less than the track safety threshold, the current track segment is removed, otherwise, the current track segment is retained to update the path anchor point topology graph.

[0033] Specifically, the step of generating a candidate path set further comprises:

[0034] For each scheduling task, the mandatory anchor point is the primary coverage anchor point, and the mobile anchor point is the secondary coverage anchor point. For multiple anchor points of the same type, they are arranged in descending order according to real-time urgency to generate an anchor point queue.

[0035] The location of the inspection device executing the scheduling task is obtained, which is defined as a starting node, and the anchor point ranked first in the anchor point queue is defined as the target anchor point of the scheduling task.

[0036] The heuristic safety coefficient and the heuristic time coefficient are calculated, and a heuristic function is constructed by weighted calculation combined with the historical failure frequency.

[0037] Using A algorithm, the target anchor point is used as a guide, and other coverage anchor points of the scheduling task are sequentially embedded based on the order of the anchor point queue to generate a candidate path.

[0038] A correction interval is set to locally adjust the candidate path to generate a candidate path set.

[0039] Specifically, the candidate prediction method comprises:

[0040] Extract the path parameters from the candidate path set, associate the abnormal inspection variables and the scheduling task information, and generate a path variable association table.

[0041] For a candidate path, the path from the starting node to any anchor point is split by track segment, the physical parameters of each track segment are extracted from the path anchor point topology graph, and the estimated time to reach each anchor point is calculated.

[0042] Based on the anchor point type, the anchor point weight is given, and the proportion of the estimated time and the time threshold and the trend correction coefficient are combined to calculate the completion degree of a single anchor point, and the task estimated completion degree is accumulated.

[0043] The real-time remaining power of the inspection device is obtained, a redundancy coefficient is set, and the redundant power is calculated.

[0044] Obtaining the predicted energy consumption, combining the redundant power, calculating the energy consumption deviation rate;

[0045] Synchronously obtaining the predicted total time consumption, combining the time threshold of the candidate path, calculating the time consumption deviation rate.

[0046] Specifically, the candidate prediction method further comprises:

[0047] Based on the energy consumption deviation rate, the time consumption deviation rate, dividing the resource adaptation level, including complete adaptation, basic adaptation and serious shortage;

[0048] For each anchor point in the candidate path, based on the anchor point abnormal value and abnormal change trend, predicting the predicted state of the anchor point after the path execution;

[0049] Comparing the predicted state with the safety parameter of the anchor point, calculating the risk exceeding probability, integrating the risk exceeding probability of all anchor points, determining the overall risk level of the path;

[0050] Integrating the task predicted completion degree, the resource adaptation level and the overall risk level of the path, calculating the comprehensive score, forming the comprehensive prediction result, setting the threshold trigger condition, triggering the responsibility transfer mechanism, the reporting mechanism, otherwise, setting the prediction update interval, updating the prediction result, generating the trigger signal.

[0051] Specifically, the responsibility transfer mechanism comprises:

[0052] Querying the inspection equipment cooperation pool, extracting the real-time state of the equipment, including the basic state and the historical cooperation record;

[0053] According to the anchor point position, the resource demand and the risk level of the task to be transferred, calculating the equipment adaptation degree, screening the equipment whose equipment adaptation degree exceeds the adaptation threshold, generating the candidate list of cooperative equipment;

[0054] According to the type and the ability of the candidate equipment, splitting the original task into subtasks, resetting the priority to form the allocation table;

[0055] Planning the path for the cooperative equipment, setting the task handover point;

[0056] Calculating the time difference of the equipment reaching the task handover point, adjusting the speed when the error threshold is exceeded;

[0057] After the equipment reaches the task handover point, completing the identity verification and data synchronization through the double-machine interaction protocol, monitoring the task progress;

[0058] After the task is completed, evaluating the transfer effect, optimizing the adaptation degree model.

[0059] Specifically, the reporting mechanism comprises:

[0060] Real-time monitoring report trigger signal, divided into one-level report, two-level report, three-level report according to the emergency degree, for different levels of report signal, automatically integrated into a standardized report information package;

[0061] According to the reporting level, automatically select the reporting channel and follow the flow rule;

[0062] After the responsibility subject receives the report information, the emergency command end is issued, the instruction content is recorded and forwarded to the execution module, and the instruction execution progress is tracked in real time;

[0063] If the instruction execution of the first-level report is overdue, the upgrade report is automatically triggered, and the overdue reminder is displayed on the large screen, and the processing result is recorded after the execution is completed;

[0064] After the processing is completed, the report closed-loop report is automatically generated, and the report mechanism is continuously optimized.

[0065] The beneficial effects of the present application are:

[0066] By real-time acquisition of abnormal inspection variables and classified processing, the bottom line variable and the elastic variable are divided, the safety bottom line is clear and cannot be broken, and the space for elastic adjustment is reserved, the decision-making error caused by the fuzzy variable boundary is avoided from the source, the dynamic baseline and the static threshold are combined, the periodic and non-periodic variables are judged respectively, the missed detection and false alarm are reduced, and the accuracy of the abnormal inspection variable screening is ensured; through the combination of multiple variables and the avoidance of repeated inspection, the priority is quantified based on the different emergency degree models of the bottom line and the elastic task, the conflict is eliminated combined with the resource adaptation degree, the high emergency degree task is processed preferentially, the delay problem caused by the mechanical processing in the traditional sorting is solved; the candidate prediction assesses the task completion degree, the resource adaptation and the risk level, identifies the execution hidden danger in advance, triggers the responsibility transfer or the report mechanism, the responsibility transfer is optimized by means of equipment cooperation and path connection, the task is seamlessly handed over, the report mechanism transmits information according to the emergency degree grading, ensures that the risk is responded in time, realizes the safety, efficiency and reliability of the air cooling island inspection, and comprehensively improves the emergency response capability. BRIEF DESCRIPTION OF DRAWINGS

[0067] Fig. 1 It is a flow chart of an intelligent inspection system for a power plant air cooling island;

[0068] Fig. 2 It is a flow chart of the priority sorting method in the present application;

[0069] Fig. 3 It is a flow chart of generating a candidate path set in the present application;

[0070] Fig. 4 It is a flow chart of the candidate prediction method in the present application. DETAILED DESCRIPTION

[0071] The technical scheme of the present application will be described in detail below with reference to the drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical scheme of the present application, but not limitations of the technical scheme of the present application. In the case of no conflict, the technical features in the embodiments of the present application and the embodiments can be combined with each other.

[0072] Reference Figs. 1 to 4 As shown in the drawings, the embodiment introduces an intelligent inspection system for an air cooling island of a power plant, which comprises a sensing module, a dynamic evaluation module and a decision module.

[0073] The sensing module is used to acquire abnormal inspection variables in the inspection process in real time, including task variables, environmental variables and device state variables. The abnormal inspection variables are classified and processed by edge computing to divide bottom line variables and elastic variables, so as to ensure that the safety bottom line cannot be broken, and at the same time, adjustment space is reserved for elastic variables, so as to avoid decision-making errors caused by variable boundary ambiguity from the source.

[0074] Among them, the task variable is an abnormal task target of the air cooling island equipment identified by the inspection device during the inspection, including emergency point position, fin temperature value and speed increase, track damage area and depth, fin surface dust coverage rate and pollutant accumulation thickness. For the task variable, the inspection device only performs processing within its own capability range, such as starting the self-provided cleaning device to process pollutant accumulation. For the abnormality beyond the capability range, such as fin temperature abnormality requiring cooling and track deformation requiring maintenance, abnormal information is generated and synchronized to the central control end, which is responded by a special device. The environmental variable is the external environmental interference encountered by the inspection device during the inspection process, including real-time wind pressure, track deformation amount and environmental temperature. The device state variable is the state parameter of the inspection device itself, including remaining power, cleaning device working condition, current position, driving motor load and historical fault record.

[0075] The dynamic evaluation module is used to extract key tasks requiring emergency scheduling from the collected abnormal inspection variables, generate a scheduling task set, and set a priority sorting method. By dynamic marginal benefit, the scheduling task is quantified as an emergency degree index to calculate the emergency degree of the scheduling task in real time, break through the fixed weight limit, and at the same time, the priority of the scheduling task is sorted and conflict is resolved in combination with the device resource state. When the priority of the scheduling task conflicts, resource adaptation degree is introduced to dynamically negotiate the priority of the scheduling task. For the problem of delay of high emergency degree task caused by multi-task mechanical sorting, the inspection resource is inclined to the most critical task.

[0076] The decision module is used for dividing the mandatory anchor point and the maneuvering anchor point, matching the time threshold and converting into three-dimensional coordinates, generating the path anchor topological graph of the air cooling island, adopting the hybrid algorithm of rules and heuristics to generate the candidate path set, setting the candidate pre-judgment method, pre-judging the task completion degree, resource consumption and conflict risk of each candidate path, calculating the emergency degree reduction ratio, the predicted remaining power, the time consumption and the weighted sum of the safety hazards after the execution of the path through the lightweight simulation engine, quantifying the abstract path into the result index to avoid the task execution deviation problem caused by the lack of result pre-judgment, and generating the inspection instruction based on the comprehensive score, triggering the responsibility transfer mechanism or the reporting mechanism through the multi-path comparison and quantitative evaluation, improving the reliability of the decision, searching the collaborative equipment pool when the responsibility is transferred, splitting the task and planning the connection path to complete the task handover and tracking optimization; the reporting mechanism integrates the information according to the emergency level, pushes it to the responsible subject through multiple channels, tracks the instruction execution and forms the closed-loop report optimization mechanism, and finally realizes the efficient execution of the scheduling task and the timely response to the risk.

[0077] In the embodiment, after the inspection equipment is started, initial abnormalities are identified. Since these initial abnormalities are dynamically changing, if only relying on the initial detection data for decision-making, improper handling will be caused due to information lag. For example, if the temperature of the fin reaches the critical value in the initial detection, and the routine inspection is still performed, the best cooling opportunity will be missed. In order to avoid information lag caused by static data, a scheduling task is generated based on the initial abnormalities to retest the abnormalities immediately, track the abnormal change trend and processing effect in real time, such as whether the dust coverage rate rebounds after cleaning, to ensure the timeliness of the basis for decision-making and avoid the blind area of not confirming the effect after abnormal processing. At the same time, for the immediate retest, if it is realized by single-device residence tracking, the original inspection plan will be disrupted, and the information of the new abnormal inspection variable will also be missed due to the lack of idle equipment. For example, the inspection equipment responsible for the B area detects an abnormality when passing through the A area, stops in the A area for tracking, and the new abnormality in the B area will be missed. Through dynamic replacement of multi-device collaborative scheduling, it is ensured that the abnormal tracking and the global inspection do not conflict, and the lag of retesting after the original plan is completed is avoided. For example, the track deformation in the A area expands to an unrecoverable state before retesting, realizing the comprehensive coverage of the initial abnormalities and the new abnormalities, and solving the problem of incomplete information collection.

[0078] Specifically, the specific steps of the classification processing include:

[0079] Real-time inspection data in the air cooling island inspection process is obtained by the multiple types of sensors carried by the inspection equipment, and preprocessed, including removing noise data by using 3σ rule, time alignment and unit standardization, so as to generate a standardized inspection data set containing inspection variable name, acquisition timestamp and value;

[0080] The actual values in the inspection data set are subjected to anomaly identification, and the inspection variables include periodic and non-periodic changes. For the periodic change of the inspection variable, such as the environmental temperature and the regular energy consumption of the equipment, a dynamic baseline is used for comparison. The historical inspection data without anomaly is selected as the baseline sample. A sliding window is set, including the window size and the sliding interval. The mean value and the standard deviation in the latest window are calculated. The baseline range of the inspection variable is generated by combining the standard coefficient. The inspection variable with the real-time value exceeding the baseline range is determined as the abnormal inspection variable. For the non-periodic change of the inspection variable, such as the track deformation and the equipment current, an initial abnormal threshold is set based on the equipment manual and the air cooling island safety regulations. The inspection variable with the actual value greater than the initial abnormal threshold is determined as the abnormal inspection variable, so as to generate the list of abnormal inspection variables , including the inspection variable name, the abnormal value, the collection position, the collection timestamp and the trigger reason; is the number of abnormal inspection variables, is the th abnormal inspection variable, , the corresponding abnormal value is ;

[0081] For the abnormal inspection variable , three equally spaced sampling points in the window are taken. The linear fitting is performed by the least square method to obtain the slope, which is defined as the abnormal change trend , and the corresponding safety parameter and the adjustable coefficient are configured. The product of the safety parameter and the adjustable coefficient is set as the first-level abnormal threshold , the safety parameter is the second-level abnormal threshold , and . At the same time, the trend threshold is set to determine the category of the abnormal inspection variable; wherein the safety parameter is an unbreakable threshold determined based on the air cooling island safety regulations and the equipment hardware limit, which is pre-recorded and locked by the engineer. The adjustable coefficient is a flexible coefficient dynamically adjusted according to the real-time running load of the air cooling island, which is issued by the central control center. The trend threshold is set based on the historical fault data.

[0082] If and , the abnormal inspection variable is determined as the flexible variable; if or , the abnormal inspection variable is determined as the bottom line variable; otherwise, the actual values of the abnormal inspection variable are continuously collected for times, and the category determination is performed again. If the continuous times do not satisfy and , directly determine as a bottom line variable, otherwise, determine as a flexible variable; wherein .

[0083] Specifically, the specific steps of the prioritization method include:

[0084] Based on the mapping rule library of the abnormal inspection variable and the task, the abnormal inspection variable is converted into a structured scheduling task to avoid task repetition or omission, including a single variable independent task and a multi-variable associated task. For the single variable independent task, one abnormal inspection variable corresponds to one scheduling task, such as a temperature exceeding the standard corresponding to a temperature detection task, and the variable attribute is directly extracted as the task basic information. For the multi-variable associated task, when the collection positions of multiple abnormal inspection variables are less than a preset range, they are combined into one scheduling task, such as the temperature exceeding the standard at point A and the slight deformation of the track at point A being combined into an A area comprehensive inspection task to avoid repeated inspection of the same area, and each scheduling task is labeled with attributes including a task ID, an associated variable name, and a required resource, and the scheduling task set is generated by sorting the abnormal inspection variable detection time;

[0085] Based on the classification processing result of the abnormal inspection variable, the scheduling tasks in the scheduling task set are divided into bottom line tasks and flexible tasks to realize accurate stratification of the risk level. If there is a bottom line variable in the scheduling task, it is classified as a bottom line task regardless of whether it contains a flexible variable, and the safety bottom line is prioritized, otherwise it is a flexible task, thereby generating a stratified task list that clearly labels the task type and the classification attributes of the associated variable;

[0086] Different emergency quantification models are designed for the two types of scheduling tasks to break through the traditional single-dimensional evaluation logic. For the bottom line task, the emergency is quantified in two dimensions of associated coverage and trend deterioration to highlight the safety priority, the number of associated variables is obtained, the task coverage coefficient is set, the task coverage coefficient is obtained by multiplying the number of associated variables and the correlation coefficient, and adding the basic coefficient, if it is a single variable independent task, the number of associated variables is zero, at this time the task coverage coefficient is the basic coefficient, based on the ratio of the abnormal change trend to the trend threshold, the trend deterioration coefficient is calculated by adding the basic coefficient, the real-time emergency of the bottom line task is obtained by multiplying the highest basic emergency, the task coverage coefficient and the trend deterioration coefficient, and the bottom line task emergency table is generated; wherein, in this embodiment, the basic coefficient is taken as 1 and the highest basic emergency is a fixed value to ensure that the priority is higher than that of the flexible task;

[0087] For the elastic task, the risk reduction amount is used to quantify the urgency based on the unit resource input, the abnormal value of the elastic variable in the task is obtained, and the difference between the safety parameter and the abnormal value is calculated, which is defined as the risk reduction amount. The weighted sum of the predicted processing time and the predicted processing energy consumption of the elastic variable is calculated to obtain the resource consumption amount, and the ratio of the risk reduction amount to the resource consumption amount is calculated. At the same time, the elastic change coefficient is introduced, and the product is obtained. The real-time urgency of the elastic task is obtained. For the multi-variable related task, the maximum value of the quantitative result is taken as the real-time urgency, and the most urgent sub-variable is highlighted, so as to generate the elastic task urgency table;

[0088] Based on the real-time urgency, the scheduling tasks are sorted, the bottom line task queue is arranged before the elastic task queue, the priority of the bottom line task is higher than that of the elastic task, and the same type of scheduling task is arranged in descending order according to the real-time urgency, so as to generate a preliminary scheduling queue including task ID, real-time urgency and queue position.

[0089] The adaptability is checked in combination with the equipment resource state to ensure that the priority has both urgency and feasibility. The adaptability of the scheduling task is calculated through the difference between the current resource of the equipment and the resource demand of the task, and the adaptability threshold is set to determine whether the queue order of the current scheduling task needs to be adjusted. The available resources of the equipment are obtained in real time by the power sensor in the equipment, and the allocation of the equipment is random allocation.

[0090] If the adaptability is less than the adaptability threshold, it indicates that the queue order of the current scheduling task needs to be adjusted, and is marked as a resource-limited task, and the queue order of the current scheduling task is lowered to the end of the same variable. For multiple resource-limited tasks, the descending order is sorted based on the adaptability. Otherwise, the original order is retained.

[0091] The queue update interval is set, the priority queue is updated based on the real-time collected data including the updated abnormal patrol variable, equipment resource consumption and task execution state, and the real-time scheduling queue is finally generated.

[0092] Specifically, the specific steps of generating the candidate path set include:

[0093] The path planning basic data integration is performed, the real-time scheduling queue issued is received through the industrial Ethernet, the attributes of each scheduling task are parsed one by one, including task ID, task priority (bottom line task or flexible task), position description and resource requirement of the associated abnormal inspection variable, meanwhile, the category of each abnormal inspection variable is extracted, the air cooling island track topology map is loaded, including the slope, friction coefficient and historical fault frequency of each track, meanwhile, the real-time state of the equipment is acquired, including the current position, remaining power and maximum endurance time, the associated matrix including the task coordinates, track parameters and equipment capacity is constructed, the data is fused, the path database is established, the task ID is taken as the index, the abnormal inspection variable attributes, track parameters and equipment state are associated, and the structured data table is formed;

[0094] For each scheduling task, according to the classification attribute of the associated abnormal inspection variable, the anchor point type is divided, including a forced anchor point and a mobile anchor point, the forced anchor point is a core node that must be covered in path planning, and if it is not covered, a safety risk will be directly caused, the position of the bottom line variable associated in the bottom line task is set as the forced anchor point and marked as a red solid point, the attribute is that the core node must be covered and if it is not covered, a safety alarm will be triggered, the mobile anchor point is a node that can be flexibly covered or bypassed in path planning, and if it is not covered, only the inspection efficiency is affected, and no safety risk is directly caused, the position of the elastic variable associated in the bottom line task and the position of all variables in the elastic task are defined as the mobile anchor point and marked as a yellow hollow point, the attribute is that the core node must be covered and if it is not covered, only the inspection integrity is affected, meanwhile, based on the real-time urgency of the task, a task time threshold mapping table is called, a corresponding interval is found, a time threshold value is matched for each anchor point, an anchor point type division table is generated, including a task ID, an anchor point type and a time threshold value, in the case that there are a plurality of anchor points under a task ID, all the anchor points need to be listed; wherein, the task time threshold mapping table is a corresponding relationship table of the real-time urgency of the task and the maximum allowed arrival time, which is generated by statistical history emergency data;

[0095] A special track coordinate system of the air cooling island is adopted, the center of the cooling tower is taken as the origin, the X axis is along the track extension direction, and the Y axis is perpendicular to the track, all the track codes and equipment numbers are bound with the track coordinate system, and the position description extracted in the task node is converted into specific three-dimensional coordinate values, the current track position of the equipment is calibrated in combination with the GPS and the RFID tag on the track, and finally a path anchor point topology map including anchor point coordinates, color markers and time constraints is generated, which is used as a spatial coordinate reference of path planning;

[0096] The track deformation amount and the distance from the current position to the high-temperature area are obtained, the ratio of the track deformation amount to the safety parameter and the ratio of the distance from the high-temperature area to the high-temperature safety distance are calculated respectively, a weighted calculation is performed to obtain an anomaly coefficient, a difference between a basic coefficient and the anomaly coefficient is calculated to obtain a safety weight of the track, and a track safety threshold is set to screen a safe track section; wherein, by obtaining the average temperature of each region in real time, the region whose average temperature exceeds a preset temperature threshold is defined as a high-temperature area;

[0097] If the safety weight is less than the track safety threshold, it is determined that the current track section is not passable, and the current track section is excluded, otherwise, the current track section is retained to update the path anchor point topology graph and label the energy consumption coefficient of each track section and the time limit threshold of each anchor point, providing a safety boundary and energy consumption reference for path generation;

[0098] For each scheduling task, an anchor point queue is generated according to the anchor point type and the real-time urgency, so that the primary coverage anchor point is forced to be a first-level coverage anchor point and the mobile anchor point is a second-level coverage anchor point. For multiple anchor points of the same type, they are arranged in descending order according to the real-time urgency. Since the anchor point is determined by the abnormal inspection variable, the real-time urgency of the anchor point is the real-time urgency of the corresponding abnormal inspection variable. Finally, the anchor point queue of the scheduling task is generated, and the anchor point ranked first in the anchor point queue is defined as the target anchor point of the scheduling task as the primary arrival node of path planning. For flexible tasks, only mobile anchor points are included, so there is no first-level coverage anchor point, and the second-level coverage anchor point is sorted, and the first-level coverage anchor point is not considered in subsequent calculations;

[0099] For a single scheduling task, A algorithm is used to generate a candidate path in the path anchor point topology graph, the position of the inspection device executing the scheduling task is obtained and defined as a starting node, the straight-line distance from the starting node to the target anchor point is obtained, and the ratio of the straight-line distance to the safety weight of the starting node is calculated to obtain a heuristic safety coefficient. Meanwhile, based on the product of the straight-line distance and the energy consumption coefficient of the starting node, the ratio of the product to the maximum moving speed of the device is calculated to obtain a heuristic time efficiency coefficient. The heuristic safety coefficient, the heuristic time efficiency coefficient and the historical failure frequency of the starting node are integrated, and a heuristic function is constructed by weighted calculation to evaluate the comprehensive cost from the starting node to the target anchor point;

[0100] With the target anchor point as the guide, the adjacent track nodes are explored according to the heuristic function value from low to high, an initial path from the starting node to the target anchor point is generated, other anchor points of the scheduling task are embedded along the initial path in turn, the shortest distance of each subsequent anchor point from the initial path is calculated according to the anchor point queue order, once the detour distance does not exceed the search proportion of the length of the initial path, the anchor point is embedded into the path, the path node sequence is updated, and the constraint check is performed synchronously to ensure that the total energy consumption of the path does not exceed the remaining power of the device, and the time to reach the target anchor point does not exceed the time threshold, finally three differentiated candidate paths are generated, which are a safety priority path, a comprehensive optimization path and a resource adaptation path, covering different scene requirements, and the generated candidate paths are stored according to task ID, and each path contains: track node sequence, all covered anchor points, total energy consumption, total time consumption, and safety weight average;

[0101] A correction interval is set to locally adjust the candidate path to generate a set of corrected candidate paths , ensuring that the path always matches the current scene; if a new bottom line task is added, the anchor point corresponding to the new bottom line task is directly inserted into the current path and the detour route is re-planned; if the remaining power of the device is less than 1.2 times the energy consumption of the path, replace the high energy consumption track segment in the path to reduce energy consumption; if the real-time urgency increase of the forced anchor point exceeds the warning proportion, increase the weight of the time efficiency score in the comprehensive score, recalculate the score and adjust the path priority; wherein, for the first candidate path of the scheduling task, in this embodiment, = 3.

[0102] Specifically, the specific steps of constructing the association matrix include:

[0103] Decompose the core data items of the three dimensions to clarify the basic composition of the matrix. The data of the task coordinate dimension comes from the real-time scheduling queue and the abnormal inspection variable classification result, including the unique identifier, priority, three-dimensional coordinates of the associated abnormal inspection variable, and the time constraint of the task. The data of the track parameter dimension comes from the air cooling island track topology map and the safety weight calculation result, including the unique identifier, physical parameters, safety attributes and energy consumption coefficient of the track. The data of the device capability dimension comes from the device real-time state and historical performance database, including the current three-dimensional position, resource constraint, motion performance and safety threshold of the device.

[0104] Establish the spatial association between the task coordinates and the track parameters to determine the path carrier from the task to the track. For the three-dimensional coordinates of each scheduling task, locate the specific track segment it is in through the air cooling island track topology map, and then trace back from the current position of the device to the scheduling task coordinates to form a mapping relationship containing the task and the necessary track segments.

[0105] Establish a matching association between track parameters and equipment capabilities, and determine the feasibility of the equipment passing through the track. For each task and the combination of the necessary track segments, check the matching from the equipment capability dimension. On the safe matching, check whether the track slope is less than the maximum allowable slope of the equipment and whether the track safety weight is not less than 0.5 to ensure safety and feasibility, and mark the results. On the energy consumption matching, calculate the energy consumption of passing through the track segment according to the track length, energy consumption coefficient, and energy consumption rate of the equipment on the slope. On the time matching, calculate the passing time according to the track length and the maximum speed of the equipment on the track segment.

[0106] Integrate the data of the three dimensions to form an association matrix. The matrix takes the task coordinates as the row index and the track parameters as the column index. The cell records the association information of the task, track, and equipment, including whether the track segment is a necessary segment for the task, the safety state of the equipment passing through the track segment, the specific energy consumption and time, and whether it is associated with the bottom line variable.

[0107] Specifically, the specific steps of the candidate pre-judgment method include:

[0108] Extract the path parameters of each path from the candidate path set, including the track node sequence, the list of covered anchor points, the total energy consumption, the total time consumption, and the average safety weight. Associate the abnormal inspection variables through the list of anchor points to determine the specific properties of the bottom line variables and the elastic variables covered by each path, such as safety parameters, abnormal values, and change trend rates. At the same time, associate the scheduling task information to obtain the task priority, time threshold, and resource demand standard corresponding to the path, thereby generating a path variable association table containing the anchor point type, associated variable properties, and task constraints of each path.

[0109] For each candidate path Split the path from the starting node to any anchor point by track segment. Extract the physical parameters of each track segment from the path anchor point topology graph, including length, slope, and friction coefficient. The moving speed of the inspection equipment on each track segment is dynamically adjusted based on the track parameters. Different slopes correspond to different moving speeds. Calculate the single-segment time of each track segment. Add the moving time of all track segments from the starting node to the anchor point and the operation time of the equipment at the anchor point to obtain the estimated time of the path to reach the anchor point.

[0110] For each anchor point covered by the path, assign different anchor point weights according to the anchor point type. Calculate the proportion of the estimated time of each anchor point to the time threshold at the anchor point. At the same time, based on the abnormal change trend in the classification processing Determine the trend correction coefficient. In this embodiment, the anchor point weights of the forced anchor point and the mobile anchor point are set to 1 and 0.7, respectively. The trend correction coefficient is determined based on the abnormal change trend in the classification processing When , set the trend correction coefficient to 1, otherwise, set the trend correction coefficient to 0.9;

[0111] The anchor weight, time ratio, and trend correction coefficient of each anchor point are multiplied to obtain the completion degree of a single anchor point, and the completion degrees of all anchor points are added to obtain the task expected completion degree of the candidate path , to comprehensively evaluate the coverage quality of the candidate path to all anchor points; the expression is as follows:

[0112]

[0113] In the formula, is the task expected completion degree of the th candidate path, is the number of anchor points covered in the candidate path, is the weight of the th anchor point in the candidate path, is the expected time to reach the anchor point , is the time limit of the anchor point , is the trend correction coefficient of the anchor point ;

[0114] Based on the candidate path , the expected energy consumption and the real-time remaining power of the inspection device are obtained, the redundancy coefficient is set to reserve power redundancy and avoid sudden consumption, the redundant power is calculated, based on the expected energy consumption and the redundant power, the energy consumption deviation rate is calculated, the expected total consumption time is obtained synchronously, combined with the time limit of the candidate path , the consumption time deviation rate is calculated, and the expression is as follows:

[0115]

[0116]

[0117] Based on the energy consumption deviation rate and the consumption time deviation rate, the first deviation threshold and the second deviation threshold are set to divide the resource adaptation level, including complete adaptation, basic adaptation, and serious shortage; if and , it means complete adaptation, indicating that the resource consumption is within the device capability range and has sufficient redundancy; if or , it means basic adaptation, indicating that the resource consumption is close to the upper limit of the device capability, which needs to be vigilant; if and At this time, the serious shortage indicates that the resources are obviously insufficient, resulting in the task being unable to be completed;

[0118] For each anchor point in the candidate path, based on the abnormal value at the anchor point and the abnormal change trend , the predicted state of the anchor point after the path is executed is predicted The predicted state is obtained by multiplying the abnormal change trend and the predicted total time consumption, and then adding the abnormal value;

[0119] The predicted state is compared with the safety parameter of the anchor point, and the risk exceeding probability is calculated, the risk exceeding probabilities of all anchor points are integrated, and the overall risk level of the path is determined, which is divided into three levels of low, medium and high;

[0120] The task predicted completion degree, resource adaptation level and overall risk level of the path are integrated, the comprehensive score of the candidate path is calculated, the comprehensive predicted result of each path is formed, and the conditions for triggering the responsibility transfer and reporting mechanism are set; when the task predicted completion degree is lower than the set score or the resource adaptation level is serious shortage or the overall risk level is high, the responsibility transfer mechanism is triggered; when the task predicted completion degree is lower than the set score or there is an anchor point with a risk exceeding probability of 1 or the comprehensive score is lower than the score threshold, the reporting mechanism is triggered; otherwise, set the predicted update interval, update the predicted result based on the latest variable data, at this time, generate a trigger signal, ensure that the predicted result can reflect the current actual situation in real time, so that the triggering of the responsibility transfer mechanism and the reporting mechanism is more accurate and always matches the latest scene.

[0121] Specifically, the specific steps of the responsibility transfer mechanism include:

[0122] The industrial Internet of Things is used to query the inspection equipment coordination pool in the air cooling island area, including all devices with inspection capabilities, including track robots, unmanned aerial vehicles, and ground inspection vehicles, and the real-time state of the devices is extracted, including the basic state, the historical cooperation record, the basic state including whether it is idle, the remaining power, the current position, and the device type, and the historical cooperation record including the cooperation success rate with local devices in the past 3 times and the processing efficiency for specific tasks;

[0123] According to the anchor point position, resource requirement, risk level of the task to be transferred, the adaptability of each device is calculated, including: based on the ratio of the remaining power of the device to the energy consumption required by the task, the resource matching degree is calculated, and the distance proportion of the straight line distance from the current position of the device to the anchor point of the task to the maximum radius of the air cooling island is calculated, the distance adaptability is obtained by subtracting the distance proportion from the basic coefficient, the historical cooperation success times proportion is calculated to obtain the historical cooperation score, the resource matching degree, the distance adaptability and the historical cooperation score are weighted and calculated to obtain the device adaptability, the devices with adaptability exceeding the adaptability threshold are screened out, and the collaborative device candidate list is generated in descending order of device adaptability score, containing device ID, device adaptability score and basic state;

[0124] According to the type and ability of the candidate device, the original task is split into multiple sub-tasks, the anchor point is preferentially assigned to the adapted device, the flexible anchor point is assigned to the flexible device, the anchor points within the preset distance range are merged into one sub-task to reduce the device cross-regional round trip, and the priority of the split sub-tasks is reset, wherein the anchor point corresponding to the sub-task of the original task retains the original task urgency, the anchor point corresponding to the sub-task of the flexible anchor point is 70% of the original task, the new path connection sub-task urgency is 50% of the original task, and finally a sub-task assignment table is formed to clearly indicate the device, anchor point list, priority and resource requirement of the sub-task;

[0125] An independent path is planned for each collaborative device, taking the current position of the device as the starting point and the anchor point of the sub-task as the target, a safe path is generated, and the energy consumption coefficient and the expected time consumption are marked, a task handover point is set at the intersection area of the paths of the original device and the collaborative device, which requires that the distance from the nearest anchor point of both parties does not exceed the preset distance, the safety weight exceeds the preset weight, and is equipped with an RFID tag for identity verification;

[0126] Based on the estimated time of the original device to reach the task handover point and the estimated time of the collaborative device to reach the task handover point, the time difference of both parties to reach the task handover point is calculated, once the time difference exceeds the preset error threshold, the driving speed of the collaborative device is adjusted to ensure time synchronization and avoid handover delay, and a collaborative device path planning diagram is generated, including task handover point coordinates, time synchronization error and energy consumption;

[0127] After the device reaches the task handover point, a double-machine interaction protocol is started, both parties verify their identities through RFID tags and device ID encryption, confirm that there is no error, the original device synchronizes the collected abnormal patrol variable data to the collaborative device through 5G, the collaborative device returns a task takeover confirmation signal after receiving the data, the original device marks the sub-task as transferred and removes it from the local task list, after the handover, the position and task progress of the collaborative device are synchronized, if the device appears abnormal, a standby device is called from the collaborative device pool for secondary transfer; after all sub-tasks are completed, the collaborative device summarizes the results to the central database, and the original device generates a task transfer summary report;

[0128] After the task is completed, the completion rate of the subtask after the transfer is calculated, the resource consumption is calculated through the ratio of the total energy consumption of multiple devices to the expected energy consumption of a single device, the time delay is calculated based on the difference between the actual total time consumption and the originally planned time consumption, the effect of responsibility transfer is evaluated, the collaborative device adaptation model is optimized according to the evaluation result, the historical cooperation points of the device with good performance are increased, the resource matching degree weight of the device causing delay is reduced, and the efficiency of future responsibility transfer is improved.

[0129] Specifically, the specific steps of the reporting mechanism include:

[0130] Real-time monitoring of reporting trigger signals is divided into three levels according to the emergency level, including first-level reporting, second-level reporting and third-level reporting, and at the same time, different levels of reporting signals are automatically integrated into standardized reporting information packages to ensure data integrity and format uniformity;

[0131] According to the reporting level, the reporting channel is automatically selected and the flow rule is followed to ensure that the information reaches the responsible subject directly;

[0132] After the responsible subject receives the reported information, the processing instruction is issued through the emergency command terminal, the instruction content is immediately recorded and forwarded to the execution module, and the instruction execution progress is tracked in real time; if the instruction involves device scheduling, the position and state of the device are updated in real time, and if it involves shutdown maintenance, the node state is recorded, such as shutdown and maintenance;

[0133] If the execution of the first-level reported instruction is overdue, automatic escalation reporting is triggered, and an overdue reminder is displayed on the large screen; after execution is completed, the processing result is recorded;

[0134] After processing is completed, a reporting closed-loop report is automatically generated, including the comparison of original reporting information and processing instructions, the difference between actual processing time and standard response time, risk mitigation effect, responsibility trace record, and continuous optimization of the reporting mechanism; if the same type of abnormality triggers first-level reporting multiple times, the safety parameters of the abnormality are downgraded; if the response of a terminal in a certain area is overdue multiple times, the reporting channel of the area is adjusted; the false positive rate of each level of reporting is calculated, and if the false positive rate is too high, the identification logic of the abnormality inspection variable is optimized.

[0135] Working principle and effect:

[0136] By collecting abnormal inspection variables in air cooling island inspection in real time, the bottom line and the boundary of elastic variables are determined through classification processing, the bottom line variable locks the safety red line, the elastic variable reserves adjustment space, the scheduling task is generated based on the abnormal inspection variable, the related tasks are combined to avoid repetition, the emergency degree is quantified through the dynamic model and combined with the resource adaptation degree to sort, the high priority task is processed in priority, the path is planned based on the task anchor point, the safe and adaptive candidate path is generated by integrating the track parameters and the equipment state, and the path is dynamically corrected to adapt to changes. The execution effect of the path is predicted, the responsibility transfer is triggered to realize the equipment cooperation or hierarchical reporting to ensure the risk response, the information lag is avoided through retesting and replacement, finally, the safety bottom line is clear, the task is efficiently executed, the path is dynamically adapted, the problems of variable ambiguity, task delay and information lag are solved, and the inspection safety and efficiency are improved.

[0137] The above only describes the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall also be considered as the protection scope of the present application.

Claims

1. An intelligent inspection system for air-cooled islands in power plants, characterized in that, include: Real-time acquisition of abnormal inspection variables, classification and processing, and division into bottom-line variables and elastic variables; Construct a set of scheduling tasks, set a priority ranking method, calculate the real-time urgency of the scheduling tasks, and perform priority ranking and conflict resolution. When scheduling tasks have priority conflicts, introduce resource adaptability to dynamically adjust the priority of the scheduling tasks. Based on the path anchor point topology map, a candidate path set is generated, a candidate prediction method is set, the result prediction is performed for each candidate path, an inspection instruction is generated based on the comprehensive score, and a responsibility transfer mechanism or a reporting mechanism is triggered. When responsibility is transferred, the collaborative equipment pool is retrieved, the task is split and the connection path is planned. The reporting mechanism integrates information according to the real-time urgency level and pushes it to the responsible entity through multiple channels. The classification process includes the following steps: Acquire real-time inspection data during the air-cooled island inspection process, and preprocess the data to generate an inspection dataset. Anomaly identification is performed on the actual values ​​in the inspection dataset to generate a list of abnormal inspection variables. ;in, The number of abnormal inspection variables. For the first One abnormal inspection variable, The corresponding outlier is ; For abnormal inspection variables Calculate abnormal change trends Configure the corresponding security parameters and adjustable coefficients, and set the first-level anomaly threshold. and Level 2 anomaly threshold At the same time, set the trend threshold. To determine the category of abnormal inspection variables; like and If so, then the abnormal inspection variable is determined to be an elastic variable; if or If so, the abnormal inspection variable is determined to be a bottom-line variable; Otherwise, continue collecting. The actual values ​​of the abnormal inspection variables mentioned above are then used to determine the category again. If they are continuous... The average does not meet the requirements. and If it is determined to be a baseline variable, it is determined to be an elastic variable; otherwise, it is determined to be an elastic variable.

2. The intelligent inspection system for air-cooled islands in power plants according to claim 1, characterized in that, The priority sorting method includes: Based on the mapping rule base, the abnormal inspection variables are converted into scheduling tasks; Based on the results of the classification process, the scheduling tasks are stratified. If there are bottom-line variables in the scheduling tasks, they are classified as bottom-line tasks; otherwise, they are elastic tasks. A stratified task list is generated. For bottom-line tasks, obtain the number of related variables, set the task coverage coefficient, and calculate the trend deterioration coefficient by superimposing the base coefficient based on the ratio of abnormal change trend to trend threshold, and then multiply it with the highest base urgency to generate the real-time urgency of the bottom-line tasks. For elastic tasks, the risk reduction is calculated based on the difference between safety parameters and outliers. The resource consumption is obtained by calculating the weighted sum of the expected processing time and expected processing energy consumption of elastic variables. The ratio of risk reduction to resource consumption is calculated. At the same time, the elasticity change coefficient is introduced to generate the real-time urgency of elastic tasks. The scheduling tasks are sorted, with bottom-line tasks having higher priority than elastic tasks. Tasks of the same type are sorted in descending order of real-time urgency to generate a preliminary scheduling queue.

3. The intelligent inspection system for air-cooled islands in power plants according to claim 2, characterized in that, The priority ranking method also includes: The suitability of the scheduled task is calculated by the difference between the current resource requirements of the device and the resource requirements of the task, and a suitability threshold is set to determine whether the queue order of the current scheduled task should be adjusted. If the fit is less than the fit threshold, it is marked as a resource-constrained task, and the queue order of the corresponding task is reduced to the end of the same type of variable. For multiple resource-constrained tasks, they are sorted in descending order based on the fit; otherwise, the original order is retained. Set the queue update interval, update the priority queue based on real-time collected data, and generate a real-time scheduling queue.

4. The intelligent inspection system for air-cooled islands in power plants according to claim 3, characterized in that, The steps for generating a candidate path set include: Obtain the real-time status of the device, construct an association matrix, and establish a path database by combining the real-time scheduling queue and the results of classification processing; Set the bottom-line variable as the mandatory anchor point and the elastic variable as the mobile anchor point. Based on the real-time urgency, call the task timeliness threshold mapping table to match the timeliness threshold for each anchor point and generate an anchor point type classification table. Construct a path anchor point topology map based on the orbital coordinate system; Calculate the safety weight of the track and set the track safety threshold to determine whether the current track is safe; If the safety weight is less than the track safety threshold, the current track segment is removed; otherwise, the current track segment is retained to update the path anchor point topology.

5. The intelligent inspection system for air-cooled islands in power plants according to claim 4, characterized in that, The steps for generating a candidate path set also include: For each scheduling task, mandatory anchor points are used as primary coverage anchor points and mobile anchor points as secondary coverage anchor points. For multiple anchor points of the same type, they are arranged in descending order according to real-time urgency to generate an anchor point queue. The location of the inspection equipment executing the scheduling task is obtained and defined as the starting node. The anchor point at the top of the anchor point queue is defined as the target anchor point of the scheduling task. The heuristic safety factor and heuristic timeliness factor are calculated, and a heuristic function is constructed by weighted calculation in combination with historical failure frequency. Guided by the target anchor point, explore adjacent orbital nodes from low to high according to the heuristic function value to generate an initial path; Other anchor points of the scheduled task are embedded sequentially along the initial path. Based on the order of the anchor point queue, other covering anchor points of the scheduled task are embedded sequentially, and constraint checks are performed simultaneously to generate candidate paths. Set a correction interval to locally adjust the candidate paths and generate a candidate path set.

6. The intelligent inspection system for air-cooled islands in power plants according to claim 5, characterized in that, The candidate prediction method includes: Extract path parameters from the candidate path set, associate them with anomaly inspection variables and scheduling task information, and generate a path variable association table; For candidate paths, the path from the starting node to any anchor point is divided into track segments. The physical parameters of each track segment are extracted from the path anchor point topology map, and the estimated time to reach each anchor point is calculated. Anchor points are assigned weights based on their type, and the completion rate of a single anchor point is calculated by combining the ratio of the estimated time to the timeliness threshold and the trend correction coefficient. The results are then accumulated to obtain the estimated completion rate of the task. Obtain the real-time remaining power of the inspection equipment, set the redundancy coefficient, and calculate the redundant power. Obtain the estimated energy consumption and, in conjunction with the redundant power, calculate the energy consumption deviation rate; The estimated total time is obtained synchronously, and the time deviation rate is calculated by combining the timeliness threshold of the candidate path.

7. The intelligent inspection system for air-cooled islands in power plants according to claim 6, characterized in that, The candidate prediction method also includes: Based on the energy consumption deviation rate and the time consumption deviation rate, resource adaptation levels are divided into fully adapted, basically adapted, and severely insufficient. For each anchor point in the candidate path, based on the outliers and abnormal change trends at the anchor point, predict the predicted state of the anchor point after the path is executed. The predicted state is compared with the safety parameters of the anchor points to calculate the probability of risk exceeding the standard. The overall risk level of the path is determined by combining the probability of risk exceeding the standard of all anchor points. The system integrates the expected completion rate of the task, the resource suitability level, and the overall risk level of the path, calculates a comprehensive score, forms a comprehensive expected result, sets threshold trigger conditions to trigger the responsibility transfer mechanism and the reporting mechanism; otherwise, it sets the expected update interval to update the expected result and generate a trigger signal.

8. The intelligent inspection system for air-cooled islands in power plants according to claim 7, characterized in that, The liability transfer mechanism includes: Query the inspection equipment collaboration pool to extract the real-time status of the equipment, including basic status and historical collaboration records; Based on the anchor point location, resource requirements, and risk level of the task to be transferred, calculate the equipment compatibility, filter out equipment whose compatibility exceeds the compatibility threshold, and generate a candidate list of collaborative equipment. The original task is split into sub-tasks based on the candidate device type and capabilities, and the priorities are reset to form an allocation table; Plan paths for collaborative devices and set task handover points; Calculate the time difference between the arrival of the calculation device at the task handover point, and adjust the speed when the error threshold is exceeded; After the equipment arrives at the task handover point, it completes identity verification and data synchronization through a dual-machine interaction protocol to monitor task progress. After the task is completed, evaluate the transfer effect and optimize the adaptation model.

9. The intelligent inspection system for air-cooled islands in power plants according to claim 8, characterized in that, The reporting mechanism includes: Real-time monitoring and reporting of trigger signals are categorized into Level 1, Level 2, and Level 3 reporting based on urgency. Reporting signals of different levels are automatically integrated into standardized reporting information packages. The system automatically selects the reporting channel based on the reporting level and follows the circulation rules. After receiving the reported information, the responsible party issues processing instructions through the emergency command terminal, records the content of the instructions and forwards them to the execution module, and tracks the progress of instruction execution in real time. If the execution of a first-level reported instruction times out, an escalation report will be automatically triggered, and a timeout reminder will be displayed on the large screen. After the execution is completed, the processing result will be recorded. Once processing is complete, a closed-loop report is automatically generated, and the reporting mechanism is continuously optimized.

Images