Method and system for robot monitoring data interaction based on eDNA and artificial intelligence

Abstract

The present application belongs to the field of robot monitoring interaction, and particularly relates to a robot monitoring data interaction method and system based on eDNA and artificial intelligence, which comprises the following steps: constructing an equipment data graph, a global task pool and a robot task execution state graph in an eDNA real-time database; dynamically generating an inspection instruction with a priority score by monitoring equipment abnormal state index values and combining a time sequence prediction model and an event-task rule library; allocating tasks for robots and updating the task execution state graph by using a multi-objective optimization algorithm; then performing data consistency evaluation on the perception result data packets collected by the robots, triggering a review inspection instruction for the data that does not meet the evaluation results; and finally writing the data packets that meet the evaluation results back to the eDNA real-time database; the present application realizes full-process automation from abnormal perception, task generation, optimization scheduling to data checking and feedback, and significantly improves the monitoring efficiency and data reliability.

Description

Technical Field

[0001] This invention belongs to the field of robot monitoring and interaction, and particularly relates to a robot monitoring data interaction method and system based on eDNA and artificial intelligence. Background Technology

[0002] As a core infrastructure for energy supply, the operating status of power plants directly determines the stability of power supply and the safety of production. With the expansion of modern power plant unit capacity and the improvement of equipment intelligence, the equipment to be monitored includes core equipment such as boilers, steam turbines, generators, and transformers, as well as auxiliary systems such as circulating water and condensate. Traditional monitoring methods are no longer sufficient to meet the needs of real-time processing of massive amounts of data. The need to use digital and intelligent systems for real-time monitoring, fault analysis, and rapid response of equipment operating status is becoming increasingly urgent.

[0003] Currently, some smart power plants use IoT technology and robots to assist in equipment monitoring. Existing technologies mainly rely on distributed control systems to collect and store equipment parameters, and then send alarms to operators' terminals based on thresholds. Operators then make judgments before performing maintenance. In existing technologies, robot inspections are set to fixed times and routes, or are remotely operated by humans to complete daily passive inspection processes.

[0004] The existing technologies mentioned above have the following problems: the monitoring data results rely on manual judgment and processing, which has high latency, is not only time-consuming and labor-intensive, but also easily leads to task backlog or omissions; the robot inspection tasks are assigned manually, resulting in low scheduling efficiency; the accuracy of equipment sensors is low, leading to false alarms or missed information. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a robot monitoring data interaction method and system based on eDNA and artificial intelligence. The method includes: constructing a device data graph, a global task pool, and a robot task execution status graph in the eDNA real-time database; dynamically generating inspection instructions with priority scores by monitoring device anomaly status indicators and combining a time-series prediction model with an event-task rule base; allocating tasks to the robot and updating the task execution status graph using a multi-objective optimization algorithm; then performing data consistency evaluation on the sensory result data packets collected by the robot, triggering review inspection instructions for data with substandard evaluation results; and finally writing back the qualified data packets to the eDNA real-time database. This invention achieves full-process automation from anomaly detection, task generation, optimized scheduling to data verification and feedback, significantly improving monitoring efficiency and data reliability.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The robot monitoring data interaction method based on eDNA and artificial intelligence includes: S1. Based on all the equipment and robots to be monitored in the power plant, create equipment data diagrams, global task pools and robot task execution status diagrams in the eDNA real-time database. S2. Monitor the abnormal status index values ​​representing the operating status of all devices to be monitored in the device data graph, combine the time series prediction model with the preset event-task rule library, generate inspection instructions with priority scores, and write the inspection instructions into the global task pool. S3. Track and record the task execution status diagram, and update the task execution status diagram by combining the inspection instructions in the global task pool and through a multi-objective optimization algorithm; the task execution status diagram includes the identifier, real-time position, inspection instructions to be executed, task execution status and start timestamp of each robot; S4. Each idle robot in the updated task execution state diagram collects real-time data of the target device to be monitored and obtains the perception result data packet of the target device to be monitored; the idle robot and the target device to be monitored correspond one-to-one under each inspection instruction to be executed. S5. Perform a data consistency assessment based on the perception result data packet. If the data consistency assessment fails to meet the standard, issue a data review and inspection instruction and update the perception result data packet. S6. If the data consistency assessment meets the standard, the qualified perception result data packet is written back to the eDNA real-time database through the eDNA interface, and the device data map is updated.

[0007] Specifically, the steps for generating inspection instructions with priority scores include: S201. Based on the real-time collected abnormal state index values, the index sequence of N consecutive sampling periods is recorded through a sliding time window, and the abnormal state index values ​​of adjacent sampling periods are differentially calculated to obtain the abnormal state index change rate. S202. Based on the abnormal state index value, the abnormal state index change rate and the equipment historical abnormal data sequence, combined with the time series prediction model, output a state index prediction sequence that represents the future state evolution trend of the equipment. S203. If the abnormal state index value exceeds the first preset safety threshold within N consecutive sampling periods, or the state index prediction sequence exceeds the second preset safety threshold, or the abnormal state index change rate exceeds the preset change rate threshold, then an event trigger command is output.

[0008] Specifically, the steps for generating inspection instructions with priority scores also include: S204. Based on the target device type and abnormal status indicator type corresponding to the event triggering instruction, and combined with the mapping relationship in the event-task rule base, the inspection instruction of the target device to be monitored is obtained; S205. Based on the criticality of the target monitored device corresponding to the event triggering instruction and the deviation ratio of the abnormal state index value, an impact factor is calculated through a weighted evaluation algorithm; the impact factor characterizes the degree of influence of the abnormal state index value on the stability and safety of the target monitored device; the criticality of the device characterizes the importance of the target monitored device in the production chain; S206. Based on the impact factor, a normalization operation is performed to obtain the priority score of the inspection instruction; based on the inspection instruction and its priority score, a global task pool is constructed; the task status in the global task pool includes waiting to be assigned and assigned.

[0009] Specifically, the task execution state diagram is used to assign inspection instructions to be executed to each robot, and the specific steps include: S301. Monitor the task execution status of all robots and obtain the task type, target equipment type to be monitored, and duration of operation of the currently executing inspection instructions; S302. Input the task type and the target equipment type to be monitored into a preset prediction model, and output the estimated total operation time of the inspection instruction; the prediction model is obtained by training based on the historical execution records of the same type of inspection instruction; S303. Calculate the difference between the estimated total operation time and the operation time to obtain the remaining operation time; when the remaining operation time is less than a preset time threshold, mark the corresponding robot as an idle robot.

[0010] Specifically, the steps for assigning inspection instructions to each robot also include: S304. Based on the inspection instructions waiting to be assigned in the global task pool, extract the three-dimensional spatial coordinates of the target monitoring device corresponding to the inspection instruction and the real-time position of the idle robot, and calculate the path distance from the idle robot to each target monitoring device. S305. Based on the path distance, the priority score, and the historical average confidence score, the adaptation score of each inspection instruction is calculated using a multi-objective optimization algorithm; the historical average confidence score is used to characterize the execution reliability of the idle robot for the data types to be collected in the inspection instructions corresponding to each target device to be monitored; S306. Select the inspection instruction with the highest adaptation score as the inspection instruction to be executed by the idle robot.

[0011] Specifically, each robot is assigned an inspection instruction to be executed. The specific steps also include: continuously monitoring the abnormal status index value associated with the inspection instruction and synchronously updating the priority score of the inspection instruction; if the waiting time of any inspection instruction waiting to be assigned exceeds the preset maximum tolerance time threshold, or the deviation rate of the abnormal status index value exceeds the preset risk threshold, then the inspection instruction is forcibly selected as the inspection instruction to be executed for the robot with the shortest remaining operation time, and the task status of the inspection instructions already assigned to the robot is marked as waiting to be assigned.

[0012] Specifically, the data consistency assessment based on the perceived result data packet includes the following steps: The perception result data package includes the collected data and confidence level of the target monitored device. If the confidence level is lower than the corresponding preset confidence threshold, the data consistency assessment is deemed unsatisfactory. Alternatively, based on the collected data and the abnormal state index value, the device parameter deviation ratio is calculated using a proportional difference function. If the device parameter deviation ratio exceeds a preset error ratio, the data consistency assessment is deemed unsatisfactory. The confidence level is the probability score output by the inference model after semantic parsing the collected data, representing the degree of certainty of the inference model regarding the parsing results.

[0013] Specifically, the robot monitoring data interaction method based on eDNA and artificial intelligence also includes: S801. Based on historical fault data of all equipment in the power plant, the sliding time window method is used to obtain the number of faults of each equipment in each time window, and the Jaccard similarity coefficient algorithm is combined to calculate the temporal correlation strength of each equipment; the number of faults includes the total number of faults of each equipment in the collection period and the number of common faults of any two equipment with correlation. S802. Based on the equipment connection relationship and process flow diagram in the equipment data diagram, analyze the propagation path of medium flow and energy flow in the production link through a directed graph model, and calculate the process dependency between equipment. S803. Obtain the shortest directed path between each device based on the device connection relationship, and calculate the propagation coefficient between devices by combining the exponential decay function; S804. Perform normalized multiplication on the temporal correlation strength, the process dependency, and the propagation coefficient to obtain the fault correlation degree between each device, and embed it into the device data graph.

[0014] Specifically, generating inspection instructions with priority scores also includes: S901. Based on the sensing result data packet that meets the data consistency assessment standard, determine whether the target monitored device is faulty; S902. If the target device to be monitored is determined to be in a fault state, obtain all devices whose fault correlation with the target device to be monitored is greater than a preset correlation threshold, and form a set of associated devices. S903. Based on the abnormal operating parameter types associated with the associated devices in the associated device set and the target device to be monitored, obtain the inspection instructions and corresponding priority scores of the associated devices, and write them into the global task pool.

[0015] A robot-based data interaction system for monitoring inventory based on eDNA and artificial intelligence includes: eDNA real-time database server is used to store equipment data diagrams, global task pools, and robot task execution status diagrams within the power plant. The instruction generation module is used to monitor the abnormal status index values ​​representing the operating status of all devices to be monitored in the device data graph, combine the time series prediction model with the preset event-task rule library, generate inspection instructions with priority scores, and write the inspection instructions into the global task pool. The task scheduling module is used to track and record the task execution status graph of all robots, and update the task execution status graph by combining the inspection instructions in the global task pool and through a multi-objective optimization algorithm. The task execution status graph includes the identifier, real-time position, inspection instructions to be executed, task execution status and start timestamp of each robot. The data acquisition module is used to collect real-time data of the target device to be monitored and obtain the sensing result data packet of the target device to be monitored. The data consistency assessment module is used to assess the data consistency of the sensing result data packet. If the data consistency assessment fails to meet the standard, a data review and inspection instruction is issued, and the sensing result data packet is updated. The data write-back module is used to write the data packet of the perception result that has passed the data consistency assessment back to the eDNA real-time database through the eDNA interface, and update the device data map.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention automatically analyzes and generates inspection instructions through a real-time eDNA database and a built-in event-task rule base, communicating directly with the robot to replace the manual decision-making process and improve the response speed and accuracy of task processing. It employs a multi-objective optimization algorithm to assign tasks to each robot, considering not only path distance but also the urgency of the inspection instructions and the execution capabilities of different robots, maximizing resource utilization and processing efficiency. Furthermore, this invention introduces a data consistency assessment mechanism, which can effectively identify and correct abnormal sensing data through write-back, ensuring the accuracy and reliability of monitoring results and reducing false alarms or missed alarms caused by inaccurate sensors. Attached Figure Description

[0017] Figure 1 This is a flowchart of a robot monitoring data interaction method based on eDNA and artificial intelligence, according to an embodiment of the present invention. Figure 2 A flowchart for generating event triggering instructions in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating an example of constructing a directed graph of power plant system equipment according to an embodiment of the present invention; Figure 4 This is a block diagram of a robot monitoring data interaction system based on eDNA and artificial intelligence, according to an embodiment of the present invention. Detailed Implementation

[0018] Example 1: Please see Figure 1 One embodiment of the present invention provides a method for robot monitoring data interaction based on eDNA and artificial intelligence, comprising the following steps: S1. Based on all the equipment and robots to be monitored in the power plant, create equipment data diagrams, global task pools and robot task execution status diagrams in the eDNA real-time database. S2. Monitor the abnormal status index values ​​representing the operating status of all devices to be monitored in the device data graph, combine the time series prediction model with the preset event-task rule library, generate inspection instructions with priority scores, and write the inspection instructions into the global task pool. S3. Track and record the task execution status graph, and update the task execution status graph by combining the inspection instructions in the global task pool and using a multi-objective optimization algorithm. S4. Each idle robot in the updated task execution state diagram collects real-time data of the target device to be monitored and obtains the perception result data packet of the target device to be monitored; the idle robot and the target device to be monitored correspond one-to-one under each inspection instruction to be executed. S5. Perform a data consistency assessment based on the perception result data packet. If the data consistency assessment fails to meet the standard, issue a data review and inspection instruction and update the perception result data packet. S6. If the data consistency assessment meets the standard, the qualified perception result data packet is written back to the eDNA real-time database through the eDNA interface, and the device data map is updated.

[0019] Furthermore, in this embodiment, the eDNA real-time database serves as a central data interface platform, communicating in real-time with the devices and robots via the eDNA interface. The device data graph includes graph nodes and graph edges, where graph edges represent device connections and graph nodes represent device entities. Each graph node includes a device identifier, an abnormal status index value, and three-dimensional spatial coordinates. The task execution status graph includes the identifier of each robot, its real-time location, pending inspection instructions, task execution status, and start timestamp. The global task pool is used to centrally manage inspection instructions or data verification inspection instructions triggered by abnormal events, including all instructions and their generation timestamps, task status, and scheduling decisions. Instructions include the identifier of the target monitored device, task type, and priority score. A visual monitoring interface is built based on the eDNA real-time database to support human-computer interaction functions such as information query, instruction issuance, and maintenance operations.

[0020] In this embodiment, abnormal status index values ​​of key components of the target monitored equipment are collected in real time through the power plant's DCS or edge sensor network. The collected data is transmitted to the eDNA real-time database via a communication interface, and an encryption protocol is used during the transmission process to ensure data security. The abnormal status index values ​​include one or more of the following: temperature, pressure, flow rate, vibration amplitude, current, voltage, and speed, which characterize the operating status of the boiler, turbine, generator, and auxiliary equipment systems.

[0021] Furthermore, in this embodiment, based on real-time and historical data of equipment operating status, combined with a time-series prediction model, the system proactively identifies potential anomalies and triggers events, thereby generating inspection instructions with priority scores. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 The flowchart for generating event triggering instructions in this embodiment of the invention includes the following steps: S201. Based on the real-time collected abnormal state index values, the index sequence of N consecutive sampling periods is recorded through a sliding time window, and the abnormal state index values ​​of adjacent sampling periods are differentially calculated to obtain the abnormal state index change rate; wherein the sampling period N is dynamically adjusted according to the equipment type and sampling frequency. For fast-response equipment, such as frequency converters and regulating valves, N takes a smaller value; for equipment with slow-changing processes, such as boilers and heat exchangers, N takes a larger value.

[0022] In this embodiment, at the arrival of each new sampling period, the window slides forward one position, removing the oldest data and adding the latest abnormal state indicator value, forming a continuous indicator sequence. Specifically, the indicator sequence is as follows: ,in Let N be the abnormal state index value at the p-th sampling time; there are a total of N abnormal state index values; the rate of change sequence is... Among them, the rate of change of abnormal state indicators There are a total of N-1 abnormal state indicator change rates, among which, and These are the abnormal state index values ​​at the q-th and q+1-th sampling times, respectively; the average value of the rate of change sequence is taken as the rate of change of the abnormal state index in the current period, which is used to reflect the average severity of the change in the abnormal state index value.

[0023] S202. Based on the abnormal state index values, the abnormal state index change rate, and the equipment historical abnormal data sequence, and combined with the time series prediction model, output a state index prediction sequence representing the future state evolution trend of the equipment. In this embodiment, the equipment historical abnormal data sequence is composed of abnormal data extracted from the eDNA database over the past three years, including the historical abnormal state index value sequence, historical change rate sequence, and abnormal event markers for each device to be monitored, wherein the abnormal event markers are the confirmed times of occurrence of abnormalities or faults. The equipment historical abnormal data sequences are grouped and stored in groups according to the same equipment type and the same abnormal state index type, aligned with the same timestamp, and the sampling period is the same as the real-time data of similar devices to be monitored.

[0024] Specifically, in this embodiment, historical anomaly data is first standardized to eliminate dimensional differences. A Long Short-Term Memory (LSTM) network is selected as the time-series prediction model. The sequence of anomaly state index values ​​and the rate of change of the anomaly state index collected in the current window are used as inputs to the model, and the output is a predicted sequence of state indices for the next M time periods. In this embodiment, the LSTM network structure includes an input layer, two LSTM layers, and a fully connected output layer. The input layer has the same dimension as the input features. The number of hidden units in the first LSTM layer is set to 128, and the number of hidden units in the second LSTM layer is set to 64. The number of neurons in the fully connected output layer corresponds to the prediction step size for the next M time periods. The network is trained using historical anomaly data and a mean squared error loss function, employing the Adam optimizer with an initial learning rate of 0.001 and a batch size of 64. An early stopping mechanism is introduced: training stops when the validation set loss value does not decrease for 10 consecutive training epochs to prevent overfitting.

[0025] S203. If the abnormal state indicator value exceeds the first preset safety threshold within N consecutive sampling periods, or the predicted sequence of the state indicator exceeds the second preset safety threshold, or the rate of change of the abnormal state indicator exceeds the preset rate of change threshold, then an event trigger command is output. The first preset safety threshold, the second preset safety threshold, and the preset rate of change threshold are set according to the equipment type and operating specifications. The second preset safety threshold is not less than the first preset safety threshold. The event trigger command also includes an equipment identifier and an abnormal state indicator type, used to analyze and obtain the corresponding inspection command, and synchronize it to the operation and maintenance monitoring platform for early warning reminders. The abnormal state indicator types include, but are not limited to, temperature, pressure, liquid level, and flow rate.

[0026] Furthermore, the specific steps for generating inspection instructions with priority scores also include: S204. Based on the target device type and abnormal status indicator type corresponding to the event triggering instruction, and combined with the mapping relationship in the event-task rule base, the inspection instruction of the target device to be monitored is obtained; the event-task rule base is a mapping relationship library between event types and inspection instructions generated and dynamically updated by clustering algorithm based on historical abnormal status indicator values ​​and corresponding task logs. Specifically, in this embodiment, when constructing the event-task rule base, abnormal state index values, trigger event types, executed inspection instructions, and judgment results after instruction execution are obtained from historical records. An unsupervised clustering algorithm is used to group abnormal events, identifying several event clusters with similar response patterns. The number of clusters is set according to the actual event handling types of the power plant. Then, for each event cluster, the inspection instruction logs corresponding to all abnormal events within the cluster are statistically analyzed. One or more tasks with the highest execution frequency and optimal task completion effect are selected as the standard tasks for that cluster, establishing an event-task mapping relationship. The task completion effect reflects the value of the data collected after executing the corresponding inspection instruction. In this embodiment, the ratio of the number of times the collected data can directly confirm equipment abnormalities or false alarms to the total number of times the task is executed is used as the representation. The selection of one or several tasks must consider both the task collaborative judgment capability and the reduction of energy waste. For example, if the event type is temperature exceeding limits and the equipment type is a boiler, and the data collection and fault analysis of boiler infrared temperature measurement and close-range visual inspection yields the best results, then it is set as the standard task for this abnormal event.

[0027] S205. Based on the criticality of the target monitored device corresponding to the event triggering instruction and the deviation ratio of the abnormal state index value, an impact factor is calculated through a weighted evaluation algorithm; the impact factor characterizes the degree of influence of the abnormal state index value on the stability and safety of the target monitored device; the criticality of the device characterizes the importance of the target monitored device in the production chain; Specifically, in this embodiment, the production chain refers to the complete production process chain of a power plant from energy input to power output. It is a closed-loop system formed by connecting core equipment, auxiliary equipment, and process links, and is used to intuitively represent the functional positioning of equipment and its upstream and downstream dependencies in the power generation process. The criticality of the aforementioned equipment is set by those skilled in the art based on the core position of the equipment in the power plant's production chain, the degree of impact caused by shutdown, and the proportion of operation and maintenance costs. The deviation ratio is the ratio of the difference between the average value of the abnormal state indicator and the first preset safety threshold to the range of the safety threshold, where the safety threshold range is the interval formed by the first preset safety threshold and zero. Furthermore, when using a weighted evaluation algorithm to calculate the impact factor, the weighting coefficient is dynamically set according to the power plant's operation strategy, and the weighting coefficient of the equipment criticality is larger during high load periods; during maintenance and repair periods, the weighting coefficient of the deviation ratio of the abnormal state indicator is larger.

[0028] S206. Based on the impact factor, a normalization operation is performed to obtain the priority score of the inspection instruction; based on the inspection instruction and its priority score, a global task pool is constructed; the task status in the global task pool includes waiting to be assigned and assigned.

[0029] Specifically, in this embodiment, the steps for calculating the priority score include: extracting the maximum and minimum values ​​of the impact factors in all inspection instructions, and then using the maximum-minimum normalization algorithm to restrict all impact factors to the range of [0,1] to obtain a priority score that reflects the urgency of each inspection instruction.

[0030] Each inspection instruction is immediately stored in the global task pool upon generation. The initial task status of the inspection instruction is "waiting for allocation". After being allocated to a robot by the scheduling decision, its task status is changed to "allocated", and the robot identifier and estimated completion time are recorded. During the waiting for allocation process, the priority score is updated in real time according to the abnormal status index value monitored in the eDNA real-time database, so that allocation can be carried out in advance when the abnormal situation reaches a certain level of deterioration.

[0031] Furthermore, the task execution state diagram is used to assign inspection instructions to be executed to each robot. The specific steps include: S301. Monitor the task execution status of all robots and obtain the task type, target equipment type to be monitored, and working time of the inspection command currently being executed; wherein, the task execution status includes the task currently being executed by the robot and the assigned task; the working time is calculated based on the start execution timestamp and the collection timestamp in the task execution status diagram; S302. Input the task type and the target equipment type to be monitored into a preset prediction model, and output the estimated total operation time of the inspection instruction; the prediction model is obtained by training based on the historical execution records of the same type of inspection instruction; Specifically, in this embodiment, the pre-set prediction model is constructed using a gradient boosting decision tree algorithm. It takes a two-dimensional feature vector composed of task type encoding and target monitored device type encoding as input, and outputs the estimated total operation time as a continuous value. The model is trained using historical execution records stored in the eDNA real-time database. Each execution record corresponds to a successfully executed inspection instruction by the robot, including the task type, target monitored device type, robot identifier, and actual task execution time. The prediction model has 200 trees, searched within the range of 100-500 to ensure a gradual decrease in the root mean square error of the validation set. The maximum tree depth is set to 6 and searched within the range of 3-10. The learning rate is set to 0.05, and the subsampling rate is set to 0.8. During model training, the task execution records completed in the past two years in the eDNA real-time database are divided into a training set and a validation set in an 8:2 ratio, with the root mean square error used as the loss function.

[0032] S303. Calculate the difference between the estimated total task duration and the completed task duration to obtain the remaining task duration. When the remaining task duration is less than a preset time threshold, mark the corresponding robot as an idle robot. Specifically, in this embodiment, an idle robot refers to a robot that is about to complete its current task and can be assigned to the next task, rather than a robot that is strictly in an idle state. The preset time threshold is determined based on the average task switching time cost of the robot and the path planning response delay, and is usually set to a short period of a few seconds so that the pre-assignment of the next task can be started before the current inspection instruction is completed, reducing the scheduling window period.

[0033] Furthermore, the specific steps for assigning inspection instructions to each robot also include: S304. Based on the inspection instructions waiting to be assigned in the global task pool, extract the three-dimensional spatial coordinates of the target monitoring equipment corresponding to the inspection instructions and the real-time position of the idle robot, and calculate the path distance from the idle robot to each target monitoring equipment; specifically, the path distance is the shortest distance from the idle robot to each target monitoring equipment calculated based on the robot movement routes pre-deployed in the power plant and the Dijkstra algorithm. S305. Based on the path distance, the priority score, and the historical average confidence score, a multi-objective optimization algorithm is used to calculate the fit score of each inspection instruction. The historical average confidence score is used to characterize the execution reliability of the idle robot for the data types to be collected in the inspection instructions corresponding to each target monitored device. Specifically, in this embodiment, the confidence scores of the same robot performing the same type of task multiple times in the historical record are averaged to obtain the historical average confidence score. The higher the score, the more proficient and reliable the corresponding robot is in performing this type of task. The lower the score, the more interference there is when the corresponding robot performs this type of task, or the robot's data acquisition module has poor performance, and the data acquisition reliability is not high. When a robot has a high historical average confidence score for performing a certain type of inspection instruction, it is given priority in assigning the robot to perform the same type of inspection instruction when allocating tasks, thereby increasing the possibility of efficient completion. Specifically, in this embodiment, the multi-objective optimization algorithm stipulates that the shorter the path distance, the higher the fit score. Therefore, after the path distance is processed by maximum-minimum normalization, the path score is obtained by combining it with the inverse proportional function. Then, the path score, priority score and historical average confidence score are weighted and summed to obtain the fit score between each inspection instruction to be assigned and the current idle robot. The weighting coefficient of the weighted summation is determined based on the experiment with the cumulative delay time of task assignment and the number of review triggers as optimization objectives. The weighting coefficient of the priority score is the largest, the weighting coefficient of the path score is the smallest, and the sum of the weighting coefficients of the three is 1.

[0034] S306. Select the inspection instruction with the highest adaptation score as the inspection instruction to be executed by the idle robot.

[0035] Specifically, in this embodiment, each idle robot iterates through all waiting inspection instructions and selects the inspection instruction with the highest fit score as the corresponding inspection instruction to be executed. If multiple inspection instructions have the same score, the inspection instruction generated earlier is selected based on its timestamp. If the number of idle robots exceeds the number of waiting tasks, inspection instructions with higher priority scores are assigned to idle robots with higher historical average confidence scores. Furthermore, after a waiting inspection instruction is successfully assigned, its task status changes from waiting to be assigned to assigned, and the eDNA real-time database is updated synchronously.

[0036] Furthermore, the urgency of inspection instructions evolves in real time with the equipment's operating status to avoid response delays due to static priorities. Therefore, each robot is assigned an inspection instruction to be executed. Specific steps include: continuously monitoring the abnormal status indicator values ​​associated with the inspection instruction and synchronously updating the priority score of the inspection instruction; if the waiting time of any waiting inspection instruction exceeds a preset maximum tolerance time threshold, or the deviation rate of its abnormal status indicator value exceeds a preset risk threshold, then the inspection instruction is forcibly selected as the waiting inspection instruction for the robot with the shortest remaining operation time, and the task status of the inspection instructions already assigned to the robot is marked as awaiting assignment. Specifically, in this embodiment, the waiting time of the inspection instruction is calculated from the monitoring time and the timestamp of the inspection instruction generation. The maximum tolerance time threshold is set according to different task types and equipment criticality. For example, the maximum tolerance time threshold for critical equipment such as boilers and steam turbines is 30s, and the maximum tolerance time threshold for auxiliary equipment is 200s. The risk threshold is typically set to 50%~80%.

[0037] Furthermore, a data consistency assessment is performed based on the perceived data packet, specifically including the following steps: The perception result data package includes the collected data and confidence level of the target monitored device. If the confidence level is lower than the corresponding preset confidence threshold, the data consistency assessment is deemed unsatisfactory. Alternatively, based on the collected data and the abnormal state index value, the device parameter deviation ratio is calculated using a proportional difference function. If the device parameter deviation ratio exceeds a preset error ratio, the data consistency assessment is deemed unsatisfactory. The confidence level is the probability score output by the inference model after semantic parsing the collected data, representing the degree of certainty of the inference model regarding the parsing results.

[0038] Specifically, in this embodiment, the perception result data packet consists of image or audio data of the corresponding target monitoring device collected by the idle robot according to different inspection instructions, as well as the confidence level of the corresponding acquisition module output by the built-in inference model; the preset confidence threshold is the minimum confidence requirement set according to different task types and data types, which is determined by the performance statistics in historical similar tasks, and is usually taken as the confidence score corresponding to the 5th percentile in the historical records; the preset error ratio is determined according to the normal deviation distribution between the eDNA real-time database monitoring value and the robot's on-site acquisition value under steady-state working conditions, and is usually taken as the upper limit of the 99% confidence interval of the normal deviation ratio. The specific structure of the inference model needs to be dynamically set according to the different types of data collected. For example, when the collected data is an image, the inference model adopts an image classification or object detection model built with a convolutional neural network, such as a model based on the MobileNet-v3 architecture. The input is the on-site image of the device to be monitored collected by the robot, and the output is the category label of the device status represented by the image and the corresponding confidence score. When the collected data is audio, the inference model adopts a voiceprint recognition model based on Mel-frequency cepstral coefficient features and a one-dimensional convolutional neural network. The input is the audio clip of the device to be monitored collected by the robot, and the output is the classification label of the device's operating status (such as normal, abnormal noise, friction, impact, etc.) and its confidence score.

[0039] In this embodiment, if the data consistency assessment meets the standard, meaning the data collected by the robot during inspection matches the abnormal state indicator value monitored by the eDNA real-time database, the target device to be monitored is determined to be abnormal. The sensing result data packet is written back to the eDNA real-time database, and a device abnormality alarm message is sent to the monitoring platform. If the data consistency assessment fails to meet the standard, a data review inspection command is triggered. The current inspection command and real-time priority score are rewritten into the global task pool, waiting for a robot to be assigned for review. The review sensing result data packet is then subjected to a data consistency assessment. If the review result matches the data in the eDNA real-time database, the review sensing result data packet is written back to the eDNA real-time database, and a device abnormality alarm message and a robot acquisition module fault reminder are sent to the monitoring platform. If the review result consistency assessment still fails to meet the standard, a device abnormal state indicator value monitoring sensor fault or environmental interference reminder is sent to the monitoring platform, and the data packet is not written back to the eDNA real-time database for the time being.

[0040] Furthermore, considering the potential chain reactions caused by equipment failures within a power plant, a correlation degree for equipment failures is constructed. When a failure is detected in the target monitored equipment, other potentially failing equipment is predicted. Therefore, the data interaction method for robot monitoring based on eDNA and artificial intelligence also includes: S801. Based on historical fault data of all equipment in the power plant, the sliding time window method is used to obtain the number of faults of each equipment within each time window, and the temporal correlation strength of each equipment is calculated by combining the Jaccard similarity coefficient algorithm. The number of faults includes the total number of faults of each equipment within the collection period and the number of common faults of any two related equipment. Specifically, in this embodiment, historical fault data of all equipment in the past three years is extracted from the eDNA database, including fault equipment identifier, fault start time stamp, fault end time stamp, and fault type. The window length can be set to 60 days, and the sliding step can be set to 2 days, dividing the entire historical period into continuous and partially overlapping window sequences. All window sequences are traversed, and the total number of faults of equipment i and equipment j within each time window is counted. and and the number of common failures Then, the temporal association strength between device i and device j can be calculated using the Jaccard similarity coefficient algorithm. ; S802. Based on the equipment connection relationships and process flow diagram in the equipment data diagram, the propagation paths of medium flow and energy flow in the production link are analyzed using a directed graph model, and the process dependency between equipment is calculated. Specifically, in this embodiment, the propagation path of medium flow includes the propagation direction and path of water, fuel, steam, etc., and the propagation path of energy flow includes the propagation direction and path of electrical energy, thermal energy, and mechanical energy. The directed graph model is used to describe the topological structure of the propagation paths of medium flow and energy flow in the power plant production link. Equipment is used as graph nodes, and the propagation direction of medium flow / energy flow is used as directed edges. The power plant system is modeled as a directed graph, where the weight values ​​of the directed edges are set according to the directness and influence of the medium / energy transfer between equipment, which is the process dependency between upstream and downstream equipment. For example, consider a simplified power plant system consisting of a feedwater pump (P), a heater (H), a boiler (B), a steam turbine (T), and a generator (G). The process flow is as follows: the feedwater pump P pressurizes water and sends it to the heater H. The heated water then enters the boiler B and is heated into steam. The steam drives the steam turbine T to rotate, and the steam turbine T drives the generator G to generate electricity. The steam turbine T also extracts steam to the heater H. Therefore, see [reference needed]. Figure 3 , Figure 3 This is a schematic diagram illustrating an example of constructing a directed graph for a power plant system according to an embodiment of the present invention. Based on the propagation directions of the medium flow and energy flow, a corresponding directed graph can be drawn as follows: Figure 3 As shown, a, b, c, d, and f represent the process dependency between any two devices, with values ​​ranging from [0,1]. For example, if the feedwater pump P is the only water source for the heater H, then a is set to 0.95; the steam extraction from the turbine T to the heater H is part of the regenerative cycle and belongs to the auxiliary loop, so f is set to 0.4.

[0041] S803. Obtain the shortest directed path between each device based on the device connection relationship. The propagation coefficient between devices is calculated using the exponential decay function. ;in Here is the attenuation rate parameter, and The shortest directed path is determined by fitting the historical fault propagation distance distribution. If there is no reachable path between two devices, the shortest directed path is infinite and the propagation coefficient is 0.

[0042] S804. Perform normalized multiplication on the temporal correlation strength, the process dependency, and the propagation coefficient to obtain the fault correlation degree between each device, and embed it into the device data graph.

[0043] Specifically, in this embodiment, Dijkstra's algorithm is used based on the device connection relationship to calculate the path hop count between each pair of devices as the shortest directed path. For example, based on... Figure 3 The directed graph shown has a shortest directed path of 4 hops from pump P to generator G. The propagation coefficient is 4. Assuming the feedwater pump P fails, the process dependence of heater H is a, and the propagation coefficient is... Then its fault correlation degree is , The time-series correlation strength between pump P and heater H; the process dependence of generator G is... The propagation coefficient is Its fault correlation degree is , Let P be the temporal correlation strength between the water pump P and the generator G.

[0044] Furthermore, generating inspection instructions with priority scores also includes: S901. Based on the perception result data packet that meets the data consistency assessment standard, determine whether the target device to be monitored is faulty; specifically, when the perception result data packet collected by the robot is consistent with the abnormal state index value in the eDNA real-time database, the target device to be monitored is determined to be faulty. S902. If the target device to be monitored is determined to be in a fault state, acquire all devices whose fault correlation with the target device to be monitored is greater than a preset correlation threshold, and form a set of associated devices; wherein the preset correlation threshold is determined according to the statistical distribution of historical fault records, and the fault correlation covering 95% of fault events can be taken as the correlation threshold, which is used to screen strongly associated devices that may cause a chain reaction of faults. S903. Based on the abnormal operating parameter types associated with the associated devices in the associated device set and the target device to be monitored, obtain the inspection instructions and corresponding priority scores of the associated devices, and write them into the global task pool. Specifically, in this embodiment, the associated abnormal operating parameters are the parameters of the associated devices affected by the fault of the target device to be monitored. For example, assuming the target device to be monitored is a boiler, the fault parameter is steam pressure, the associated devices include a steam turbine, and the parameters of the steam turbine affected by abnormal steam pressure are inlet air pressure and vibration acceleration. Specifically, this embodiment obtains the inspection instructions for the associated devices based on the types of parameters affected by the associated devices and in conjunction with a preset event-task rule base. Then, according to steps S205-S206, the priority score of the associated device inspection instructions is calculated. It should be noted that the operating parameters of the associated devices have not been abnormal at the current moment, and it is a preventive task, so the initial value of the deviation ratio is 0. Further, in this embodiment, when calculating the impact factor of the associated devices, the fault correlation degree is introduced, and the impact factor is obtained by weighted summation with the device criticality. At this time, the weighting coefficient of the device criticality is relatively large. Finally, the priority score of the inspection instructions is obtained by normalization algorithm, and the inspection instructions and priority scores of the associated devices are written into the global task pool for allocation.

[0045] Example 2: Please see Figure 4 Another embodiment of the present invention provides: a robot monitoring data interaction system based on eDNA and artificial intelligence, comprising: eDNA real-time database server is used to store equipment data diagrams, global task pools, and robot task execution status diagrams within the power plant. The instruction generation module is used to monitor the abnormal status index values ​​representing the operating status of all devices to be monitored in the device data graph, combine the time series prediction model with the preset event-task rule library, generate inspection instructions with priority scores, and write the inspection instructions into the global task pool. The task scheduling module is used to track and record the task execution status graph of all robots, and update the task execution status graph by combining the inspection instructions in the global task pool and through a multi-objective optimization algorithm. The task execution status graph includes the identifier, real-time position, inspection instructions to be executed, task execution status and start timestamp of each robot. The data acquisition module is used to collect real-time data of the target device to be monitored and obtain the sensing result data packet of the target device to be monitored. The data consistency assessment module is used to assess the data consistency of the sensing result data packet. If the data consistency assessment fails to meet the standard, a data review and inspection instruction is issued, and the sensing result data packet is updated. The data write-back module is used to write the data packet of the perception result that has passed the data consistency assessment back to the eDNA real-time database through the eDNA interface, and update the device data map.

[0046] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

Claims

1. A robot monitoring data interaction method based on eDNA and artificial intelligence, characterized in that, include: S1. Based on all the equipment and robots to be monitored in the power plant, create equipment data diagrams, global task pools and robot task execution status diagrams in the eDNA real-time database. S2. Monitor the abnormal status index values ​​representing the operating status of all devices to be monitored in the device data graph, combine the time series prediction model with the preset event-task rule library, generate inspection instructions with priority scores, and write the inspection instructions into the global task pool. S3. Track and record the task execution status diagram, and update the task execution status diagram by combining the inspection instructions in the global task pool and through a multi-objective optimization algorithm; the task execution status diagram includes the identifier, real-time position, inspection instructions to be executed, task execution status and start timestamp of each robot; S4. Each idle robot in the updated task execution state diagram collects real-time data of the target device to be monitored and obtains the perception result data packet of the target device to be monitored; the idle robot and the target device to be monitored correspond one-to-one under each inspection instruction to be executed. S5. Perform a data consistency assessment based on the perception result data packet. If the data consistency assessment fails to meet the standard, issue a data review and inspection instruction and update the perception result data packet. S6. If the data consistency assessment meets the standard, the qualified perception result data packet is written back to the eDNA real-time database through the eDNA interface, and the device data map is updated.

2. The robot monitoring data interaction method based on eDNA and artificial intelligence as described in claim 1, characterized in that, The specific steps for generating inspection instructions with priority scores include: S201. Based on the real-time collected abnormal state index values, the index sequence of N consecutive sampling periods is recorded through a sliding time window, and the abnormal state index values ​​of adjacent sampling periods are differentially calculated to obtain the abnormal state index change rate. S202. Based on the abnormal state index value, the abnormal state index change rate and the equipment historical abnormal data sequence, combined with the time series prediction model, output a state index prediction sequence that represents the future state evolution trend of the equipment. S203. If the abnormal state index value exceeds the first preset safety threshold within N consecutive sampling periods, or the state index prediction sequence exceeds the second preset safety threshold, or the abnormal state index change rate exceeds the preset change rate threshold, then an event trigger command is output.

3. The robot monitoring data interaction method based on eDNA and artificial intelligence as described in claim 2, characterized in that, The specific steps for generating inspection instructions with priority scores also include: S204. Based on the target device type and abnormal status indicator type corresponding to the event triggering instruction, and combined with the mapping relationship in the event-task rule base, the inspection instruction of the target device to be monitored is obtained; S205. Based on the criticality of the target monitored device corresponding to the event triggering instruction and the deviation ratio of the abnormal state index value, an impact factor is calculated through a weighted evaluation algorithm; the impact factor characterizes the degree of influence of the abnormal state index value on the stability and safety of the target monitored device; the criticality of the device characterizes the importance of the target monitored device in the production chain; S206. Based on the impact factor, a normalization operation is performed to obtain the priority score of the inspection instruction; based on the inspection instruction and its priority score, a global task pool is constructed; the task status in the global task pool includes waiting to be assigned and assigned.

4. The robot monitoring data interaction method based on eDNA and artificial intelligence as described in claim 3, characterized in that, The task execution state diagram is used to assign inspection instructions to be executed to each robot. The specific steps include: S301. Monitor the task execution status of all robots and obtain the task type, target equipment type to be monitored, and duration of operation of the currently executing inspection instructions; S302. Input the task type and the target equipment type to be monitored into a preset prediction model, and output the estimated total operation time of the inspection instruction; the prediction model is obtained by training based on the historical execution records of the same type of inspection instruction; S303. Calculate the difference between the estimated total operation time and the operation time to obtain the remaining operation time; when the remaining operation time is less than a preset time threshold, mark the corresponding robot as an idle robot.

5. The robot monitoring data interaction method based on eDNA and artificial intelligence as described in claim 4, characterized in that, The specific steps for assigning inspection instructions to each robot also include: S304. Based on the inspection instructions waiting to be assigned in the global task pool, extract the three-dimensional spatial coordinates of the target monitoring device corresponding to the inspection instruction and the real-time position of the idle robot, and calculate the path distance from the idle robot to each target monitoring device. S305. Based on the path distance, the priority score, and the historical average confidence score, the adaptation score of each inspection instruction is calculated using a multi-objective optimization algorithm; the historical average confidence score is used to characterize the execution reliability of the idle robot for the data types to be collected in the inspection instructions corresponding to each target monitored device; S306. Select the inspection instruction with the highest adaptation score as the inspection instruction to be executed by the idle robot.

6. The robot monitoring data interaction method based on eDNA and artificial intelligence as described in claim 5, characterized in that, The specific steps of assigning inspection instructions to each robot include: continuously monitoring the abnormal status index values ​​associated with the inspection instructions and synchronously updating the priority score of the inspection instructions; if the waiting time of any inspection instruction waiting to be assigned exceeds a preset maximum tolerance time threshold, or the deviation rate of its abnormal status index value exceeds a preset risk threshold, then the inspection instruction is forcibly selected as the inspection instruction to be executed for the robot with the shortest remaining operation time, and the task status of the inspection instructions already assigned to the robot is marked as waiting to be assigned.

7. The robot monitoring data interaction method based on eDNA and artificial intelligence as described in claim 6, characterized in that, The specific steps for performing data consistency assessment based on the perceived data packet include: The perception result data package includes the collected data and confidence level of the target monitored device. If the confidence level is lower than the corresponding preset confidence threshold, the data consistency assessment is deemed unsatisfactory. Alternatively, based on the collected data and the abnormal state index value, the device parameter deviation ratio is calculated using a proportional difference function. If the device parameter deviation ratio exceeds a preset error ratio, the data consistency assessment is deemed unsatisfactory. The confidence level is the probability score output by the inference model after semantic parsing the collected data, representing the degree of certainty of the inference model regarding the parsing results.

8. The robot monitoring data interaction method based on eDNA and artificial intelligence as described in claim 7, characterized in that, Also includes: S801. Based on historical fault data of all equipment in the power plant, the sliding time window method is used to obtain the number of faults of each equipment in each time window, and the Jaccard similarity coefficient algorithm is combined to calculate the temporal correlation strength of each equipment; the number of faults includes the total number of faults of each equipment in the collection period and the number of common faults of any two equipment with correlation. S802. Based on the equipment connection relationship and process flow diagram in the equipment data diagram, analyze the propagation path of medium flow and energy flow in the production link through a directed graph model, and calculate the process dependency between equipment. S803. Obtain the shortest directed path between each device based on the device connection relationship, and calculate the propagation coefficient between devices by combining the exponential decay function; S804. Perform normalized multiplication on the temporal correlation strength, the process dependency, and the propagation coefficient to obtain the fault correlation degree between each device, and embed it into the device data graph.

9. The robot monitoring data interaction method based on eDNA and artificial intelligence as described in claim 8, characterized in that, The generation of inspection instructions with priority scores also includes: S901. Based on the sensing result data packet that meets the data consistency assessment standard, determine whether the target monitored device is faulty; S902. If the target device to be monitored is determined to be in a fault state, obtain all devices whose fault correlation with the target device to be monitored is greater than a preset correlation threshold, and form a set of associated devices. S903. Based on the abnormal operating parameter types associated with the associated devices in the associated device set and the target device to be monitored, obtain the inspection instructions and corresponding priority scores of the associated devices, and write them into the global task pool.

10. A robot monitoring data interaction system based on eDNA and artificial intelligence, used to implement the robot monitoring data interaction method based on eDNA and artificial intelligence as described in any one of claims 1-9, characterized in that, include: eDNA real-time database server is used to store equipment data diagrams, global task pools, and robot task execution status diagrams within the power plant. The instruction generation module is used to monitor the abnormal status index values ​​representing the operating status of all devices to be monitored in the device data graph, combine the time series prediction model with the preset event-task rule library, generate inspection instructions with priority scores, and write the inspection instructions into the global task pool. The task scheduling module is used to track and record the task execution status graph of all robots, and update the task execution status graph by combining the inspection instructions in the global task pool and through a multi-objective optimization algorithm. The task execution status graph includes the identifier, real-time position, inspection instructions to be executed, task execution status and start timestamp of each robot. The data acquisition module is used to collect real-time data of the target device to be monitored and obtain the sensing result data packet of the target device to be monitored. The data consistency assessment module is used to assess the data consistency of the sensing result data packet. If the data consistency assessment fails to meet the standard, a data review and inspection instruction is issued, and the sensing result data packet is updated. The data write-back module is used to write the data packet of the perception result that has passed the data consistency assessment back to the eDNA real-time database through the eDNA interface, and update the device data map.

Images