Intelligent alarm system and method for monitoring power plant equipment

Abstract

The invention provides an intelligent alarm system and method for monitoring power plant equipment, and the method comprises the steps: collecting the temperature, pressure and vibration data of the equipment through a sensor, and generating a holographic image through the scanning of a laser radar, so as to complete the data collection and environment modeling. A space-time atlas is constructed and analyzed by means of a space-time network, faults are analyzed and positioned by means of topological data, and data fusion and deep analysis are achieved. And constructing a Bayesian network by adopting intelligent learning and adaptive optimization of evolutionary strategy optimization and memory enhanced neural network prediction, and carrying out fault diagnosis and causal inference in combination with anti-factual reasoning. And the group intelligent collaborative diagnosis and emotion calculation aided decision is used for intelligent decision and collaborative operation, and finally, an operator is assisted to work based on a man-machine interaction function. The system and the method can comprehensively and accurately monitor equipment, efficiently diagnose faults and assist in decision making, and improve operation safety and reliability of a power plant.

Description

Technical Field

[0001] The present invention relates to the field of intelligent alarm for power plant equipment monitoring, and in particular to an intelligent alarm system and method for power plant equipment monitoring. Background Art

[0002] In the field of power plant equipment monitoring, traditional alarm systems have many limitations. On the one hand, the data collection method is relatively simple, usually relying on conventional sensors to collect limited physical quantity data, and the sampling frequency is low, making it difficult to capture subtle changes in the equipment's operating status, and it is easy to miss early fault hazards. On the other hand, traditional analysis methods are mostly based on simple threshold judgments or empirical rules. For complex equipment operating conditions and failure modes, they lack effective in-depth analysis capabilities, cannot accurately predict the propagation and evolution of faults between equipment, and are difficult to locate the root cause of the fault. In addition, when facing new equipment operating conditions or operating condition changes, the traditional system has poor model adaptability and requires frequent manual parameter adjustments, resulting in low efficiency and difficulty in ensuring accuracy. At the same time, in the decision-making process, the impact of the operator's status on the decision is often ignored, which may lead to misjudgment or operational errors due to factors such as fatigue and tension of the operator, and the human-computer interaction is not friendly enough, and it is impossible to provide efficient assistance and support for the operator. Summary of the invention

[0003] The main purpose of the present invention is to provide an intelligent alarm system and method for monitoring power plant equipment. To solve the above technical problems, the technical solution adopted by the present invention is: an intelligent alarm system for power plant equipment monitoring, including a data acquisition and environment modeling module, a data fusion and deep analysis module, an intelligent learning and adaptive optimization component and a fault diagnosis and causal inference subsystem; The data acquisition and environment modeling module collects the operating parameters of the equipment, scans and fuses the data through laser radar and high-definition cameras, and generates a holographic image containing static and dynamic information of the equipment; The data fusion and deep analysis module combines time series data and spatial information to construct a multi-dimensional spatiotemporal map, analyze the spatiotemporal evolution of the equipment operation status, and use topological data analysis methods to explore the data topology structure and identify hidden patterns and anomalies. The intelligent learning and adaptive optimization components include an evolutionary strategy optimization module and a memory-enhanced neural network prediction module; The fault diagnosis and causal inference subsystem includes a Bayesian network fault assessment module and a counterfactual reasoning root cause analysis module.

[0004] In the preferred solution, in the intelligent learning and adaptive optimization components: the evolution strategy optimization module introduces the evolution strategy ES algorithm to build an intelligent learning framework, and compares the model parameters to the genes of individual organisms; By simulating biological evolution mechanisms such as natural selection and crossover mutation, the model can automatically select and optimize parameter combinations based on actual operating data feedback, so that the model can adapt to changes in power plant operating conditions and improve adaptability to changes in equipment status and prediction accuracy; Adopt MANN and equip it with an external memory module to store a large amount of historical operation data and fault cases. MANN is a memory-enhanced neural network. When predicting the future operating status of the equipment, the current input data is combined with similar historical patterns in the memory module to make judgments, enhance the ability to capture potential failure trends, and improve the reliability and foresight of predictions.

[0005] In the preferred scheme, in the fault diagnosis and causal inference subsystem: a Bayesian network fault diagnosis model is constructed based on historical data and expert experience to determine the probabilistic dependency relationship between equipment operating parameters, environmental factors and various types of faults, and to assign conditional probabilities to the connections between network node variables; After accessing the real-time monitoring data, the Bayesian inference engine is used to calculate the posterior probability of different faults based on the probability relationship; Among them, the possibility of various faults when the equipment temperature rises is evaluated by combining the changes of multiple parameters, providing a quantitative decision-making basis for fault diagnosis; Use counterfactual reasoning techniques to simulate hypothetical scenarios in equipment operation, including simulation of hypothetical scenarios in standby operation in the event of abnormalities in key equipment or sensor measurement deviations; By comparing the differences in system states between actual and hypothetical scenarios, we can gain a deep understanding of the fault causal mechanism and accurately determine the root cause of the fault. This can be used to diagnose faults caused by the interaction of multiple factors in complex systems and avoid misdiagnosis.

[0006] In the preferred solution, the data acquisition of the data acquisition and environmental modeling module uses sensors to collect temperature, pressure, and vibration physical quantity data, with a sampling frequency of 10-60ms once, to obtain high-precision raw data and monitor subtle changes in the operating status of the equipment in real time; Use laser radar, high-definition cameras or 3D scanning equipment to conduct a comprehensive scan of the power plant's internal structure layout, equipment appearance and operating status from multiple angles; By fusing the point cloud data and image data obtained through scanning through complex algorithms, a holographic image containing the equipment's geometric shape, positional relationship and real-time operation dynamic information is generated in real time, providing an intuitive visual basis for subsequent analysis.

[0007] In the preferred solution, in the data fusion and in-depth analysis module: With the help of space-time network technology, the time series data of equipment operating parameters are combined with information such as the spatial location and connection relationship of the equipment in the power plant to construct a multi-dimensional space-time map; Analyze the connection relationship and data propagation path between nodes in the space-time graph, observe the evolution of the equipment operation status in the space-time dimension, predict the impact of the fault on adjacent equipment in space, its propagation trend over time, and the potential impact range; Use topological data analysis methods to calculate the topological invariants of data points in high-dimensional space to explore the topological structure of data; By identifying hidden patterns in the topological structure, anomalies and abnormal patterns can be effectively discovered in high-dimensional nonlinear data sets, and faults can be located in a timely manner based on changes in the topological distribution of data points when equipment fails.

[0008] The preferred solution also includes an intelligent decision-making and collaborative operation module and a human-computer interaction and user experience improvement module; The intelligent decision-making and collaborative operation module includes a group intelligent collaborative diagnosis module and an emotional computing auxiliary decision-making module; The AI ​​computing predictive maintenance module provides natural language dialogue services to assist operators and improve user experience.

[0009] In the preferred solution, in the intelligent decision-making and collaborative operation module: Swarm intelligence collaborative diagnosis module: Multiple intelligent agents with autonomous decision-making and local information processing capabilities are deployed in the system to monitor and preliminarily diagnose the equipment or areas they are responsible for. Global collaborative diagnosis is achieved through collaborative mechanisms such as pheromone transmission and sharing of local optimal solutions. When equipment is abnormal, intelligent agents quickly gather together for joint analysis and make comprehensive judgments to draw more accurate and comprehensive fault diagnosis conclusions, thereby improving diagnostic efficiency and accuracy. Emotional computing decision-making support module: uses high-definition cameras, microphones and physiological signal monitoring equipment to collect multimodal data on operators’ facial expressions, voice intonation, and heart rate, and analyzes their emotional state and workload; The operator status is taken into account during fault diagnosis and decision-making to assist in determining the impact of human factors on operational errors or misjudgments, such as adjusting alarm displays or providing prompt information based on stressful conditions to help operators make accurate decisions.

[0010] In the preferred embodiment, S1. Data collection and environmental modeling: Use sensors to collect temperature, pressure, and vibration data of power plant equipment at a sampling frequency of 1060ms to obtain high-precision raw data and monitor subtle changes in equipment operating status in real time; Use laser radar, high-definition cameras or 3D scanning equipment to conduct a comprehensive scan of the power plant's internal structure layout, equipment appearance and operating status from multiple angles; The point cloud data and image data obtained by scanning are fused through algorithms to generate holographic images containing the geometric shape, position relationship and real-time operation dynamic information of the equipment, providing intuitive visual basis for subsequent analysis; S2, Data Fusion and In-depth Analysis Module Steps: With the help of spatiotemporal network technology, the time series data of equipment operation parameters are combined with the spatial location and connection relationship information of the equipment in the power plant to construct a multi-dimensional spatiotemporal map, where the construction of the spatiotemporal map should accurately reflect the spatiotemporal characteristics of the equipment operation data; Analyze the constructed spatiotemporal graph, including studying the connection relationship between nodes and the data propagation path, to observe the evolution of the equipment operation status in the spatiotemporal dimension, and then accurately predict the impact of the fault on adjacent equipment in space, the propagation trend over time, and the potential impact range; Use topological data analysis methods to calculate the topological invariants of data points in high-dimensional space to explore the topological structure of data. By identifying hidden patterns in the topological structure, abnormal points and abnormal patterns can be effectively found in high-dimensional nonlinear data sets. The fault can be located in time according to the changes in the topological distribution of data points when the equipment fails. The application of topological data analysis methods should have high accuracy and reliability. S3, Intelligent learning and adaptive optimization component steps: Evolution strategy optimization step, that is, introducing the evolution strategy ES algorithm to build an intelligent learning framework, analogizing the model parameters to the genes of individual organisms, simulating biological evolution mechanisms such as natural selection and crossover mutation, and automatically screening and optimizing parameter combinations based on actual operating data feedback, so that the model can adapt to changes in power plant operating conditions and improve the adaptability and prediction accuracy of equipment status changes. The implementation process of the evolution strategy should ensure the effectiveness and stability of parameter optimization; The memory-enhanced neural network prediction step uses MANN and is equipped with an external memory module to store a large amount of historical operation data and fault cases. When predicting the future operation status of the equipment, it combines the current input data with similar historical patterns in the memory module to make judgments, enhance the ability to capture potential fault trends, and improve the reliability and foresight of predictions. The prediction process of the memory-enhanced neural network should make full use of historical data and current information to ensure the reliability of the prediction results; S4, Fault diagnosis and causal inference module steps: Construct a Bayesian network fault diagnosis model based on historical data and expert experience, determine the probability dependency between equipment operating parameters, environmental factors and various types of faults, and assign conditional probabilities to the connections between network node variables. The construction of the Bayesian network model should be based on sufficient historical data and accurate expert experience; After accessing the real-time monitoring data, the Bayesian reasoning engine is used to calculate the posterior probability of different faults based on the probability relationship, and the possibility of various faults when the equipment temperature rises is evaluated by combining the changes in multiple parameters, providing a quantitative decision-making basis for fault diagnosis. The calculation process of the Bayesian reasoning engine should be accurate and efficient; Use counterfactual reasoning technology to simulate hypothetical scenarios in equipment operation, key equipment anomalies or sensor measurement deviations, etc. By comparing the differences in system states between actual and hypothetical scenarios, we can deeply understand the fault causal mechanism and accurately determine the root cause of the fault. This can be used to diagnose faults caused by the interaction of multiple factors in complex systems to avoid misdiagnosis. The application of counterfactual reasoning technology should be able to effectively reveal the causal relationship of faults; S5. Intelligent decision-making and collaborative operation steps: Group intelligent collaborative diagnosis steps, deploy multiple intelligent agents with autonomous decision-making and local information processing capabilities in the system, responsible for monitoring and preliminary diagnosis of the equipment or areas they are responsible for, and use collaborative mechanisms such as pheromone transmission and sharing of local optimal solutions to achieve global collaborative diagnosis. When the equipment is abnormal, the intelligent agents quickly gather together for joint analysis, and make comprehensive judgments to draw more accurate and comprehensive fault diagnosis conclusions, thereby improving diagnostic efficiency and accuracy. The process of group intelligent collaborative diagnosis should ensure smooth collaboration between intelligent agents and reliable diagnostic results; The emotional computing-assisted decision-making step uses high-definition cameras, microphones, and physiological signal monitoring equipment to collect multimodal data on operators’ facial expressions, voice intonation, and heart rate, analyze their emotional state and workload, consider the operator’s state during fault diagnosis and decision-making, and assist in determining the impact of human factors on operational errors or misjudgments; Adjust the alarm display or provide prompt information according to the stress state to help operators make accurate decisions. The process of emotional computing to assist decision-making should be able to accurately analyze the operator's state and provide effective auxiliary information; S6. Based on the human-computer interaction function of the system, efficient interaction between the operator and the system is achieved to assist the operator in his work.

[0011] In the preferred embodiment, steps S2-S4 are specifically as follows: S2.1. Data fusion and deep analysis module steps: With the help of spatiotemporal network technology, the time series data of equipment operating parameters are combined with the spatial location and connection relationship information of the equipment in the power plant to construct a multi-dimensional spatiotemporal map. ,in is a node set, is the edge set, is the time dimension information; define the node connection strength function for: ; in, and Node and In time The key operating parameters of is the time window size; this function is used to construct a spatiotemporal graph to accurately reflect the spatiotemporal characteristics of the equipment operation data; Analyze the constructed spatiotemporal graph and define the factors affecting fault propagation for: ; in, For Node In time The parameter change, To analyze the time window; through research To observe the evolution of the equipment's operating status in the time and space dimensions, and then accurately predict the impact of the fault on adjacent equipment in space, as well as its propagation trend over time and potential impact range; The topological data analysis method is used to calculate the topological invariant TI of data points in high-dimensional space. The data point set is , define the topological similarity metric function for: ; in, For data points In the The value under the feature map, is the adjustment coefficient, is the number of feature maps; by calculating TI, the topological structure of the data is explored, and by identifying the hidden patterns in the topological structure, abnormal points and abnormal patterns are effectively found in high-dimensional nonlinear data sets, and the fault is located in time according to the change in the topological distribution of data points when the equipment fails. The application of topological data analysis methods should have high accuracy and reliability; S3.1, Intelligent learning and adaptive optimization component steps: Evolution strategy optimization step, introduce the evolution strategy ES algorithm to build an intelligent learning framework, set the model parameter vector , define the fitness function for:

[0012] in, is the actual device operation status value, is the model prediction value, is the sample size, is the regularization coefficient; by simulating biological evolution mechanisms such as natural selection and crossover mutation, the optimization parameter combination is automatically screened based on actual operation data feedback, so that the model can adapt to the changes in power plant operating conditions and improve the adaptability and prediction accuracy of equipment status changes. The implementation process of the evolutionary strategy should ensure the effectiveness and stability of parameter optimization; The memory-enhanced neural network prediction step uses MANN and is equipped with an external memory module to store a large amount of historical operation data and fault cases; the current input data is , the historical pattern stored in the memory module is ( ), define the pattern matching function for: ; When predicting the future operating status of the device, the current input data is combined with similar historical patterns in the memory module to make a judgment; let the prediction value be , the calculation formula is: ; in, , Based on historical model The prediction value of the memory-enhanced neural network should be fully utilized to capture the potential fault trend and improve the prediction reliability and foresight. S4.1, Fault diagnosis and causal inference module steps: Based on historical data and expert experience, a Bayesian network fault diagnosis model is constructed to determine the probabilistic dependency between equipment operating parameters, environmental factors and various types of faults, and to assign conditional probabilities to the connections between network node variables; the equipment operating parameter vector is , the fault variable is , the environmental factor vector is , determine the conditional probability distribution based on a large amount of historical data and expert knowledge ; After accessing the real-time monitoring data, the Bayesian inference engine is used to calculate the posterior probability of different faults according to the probability relationship; assuming that the real-time monitoring data is and , according to Bayes' theorem, the fault The posterior probability The calculation formula is: ; in, The number of fault types; Combined with the changes in multiple parameters, the possibility of various faults when the equipment temperature rises is evaluated to provide a quantitative decision-making basis for fault diagnosis. The calculation process of the Bayesian reasoning engine should be accurate and efficient; The counterfactual reasoning technique is used to simulate the hypothetical scenario of equipment operation. The change of key equipment parameters is assumed to be , the sensor measurement deviation is , define the system state difference measurement function for: ; in, , and is the weight coefficient, It is the fault prediction value under normal circumstances; by comparing the differences in system states under actual and hypothetical scenarios, we can deeply understand the fault causal mechanism and accurately determine the root cause of the fault. It is used for fault diagnosis caused by the interaction of multiple factors in complex systems to avoid misdiagnosis. The application of counterfactual reasoning technology should be able to effectively reveal the fault causal relationship.

[0013] In the preferred solution, step S5 is specifically as follows: Group intelligence collaborative diagnosis step: the system is deployed with Agents, each agent The data obtained from monitoring the equipment or area for which it is responsible is , the fault probability vector obtained after local preliminary diagnosis is ,in is the number of fault types; Defining Agents and Agent The pheromone concentration between , and its update formula is: ; in, is the pheromone volatility coefficient, ranging from arrive between, For the agent in this iteration and Agent The increment of pheromone between Based on the local diagnosis results, it is considered that When the diagnostic results are highly correlated, ,in Increase the strength factor for pheromones. and is the adjustment parameter; this formula is used to measure the degree of information correlation between agents and guide the collaboration between agents by updating the pheromone concentration; When the intelligent Based on the local diagnosis results, it is considered that The correlation of the diagnosis results is mainly calculated by calculating the local fault probability vector and the agent The similarity of the fault possibility vector is assumed to be ,when , is the set similarity threshold; Agent Update your own diagnosis results according to the pheromone concentration and the fault probability vector of other agents. The update formula is: ; in, , for the intelligent agent For Agents After multiple iterations, until the fault probability vectors of all agents converge, the results of all agents are finally combined to obtain an accurate and comprehensive fault diagnosis conclusion. During the whole process, the communication between agents is stable and the data sharing is timely, which improves the efficiency and accuracy of diagnosis. Emotional computing aided decision-making steps: Assume that the facial expression feature vector of the operator collected by high-definition camera, microphone and physiological signal monitoring equipment is , the speech intonation feature vector is , the heart rate characteristic value is HR; Define the emotional state evaluation function as , the calculation formula is: ; in, , and is the weight coefficient of different modal data, is the average heart rate of the operator during normal work. is the standard deviation of heart rate; this function is used to comprehensively evaluate the emotional state of the operator, and its value range is arrive The closer It means the mood is more tense or unstable; Assume that the operator's workload evaluation function is ,in is the amount of equipment data and task complexity index that the current operator needs to process, and the calculation formula is: ; in, For the Quantitative indicators of data or tasks, and is the adjustment index, is the total number of data or tasks; this function is used to evaluate the workload of operators, and its value range is arrive The closer It means the workload is greater; The decision support coefficient is defined as ,when When the system believes that the operator may be affected by emotions or workload, the decision risk increases. is the set threshold; at this time, the system adjusts the alarm display intensity according to the following formula : ; in, Displays the intensity of the original alarm. is the intensity adjustment coefficient; at the same time, the system provides prompt information according to the predefined rule base; When DAC is in a certain range, it prompts the operator to check specific equipment or review the operation steps to help the operator make accurate decisions. The process of emotional computing-assisted decision-making should be able to accurately analyze the operator's status and provide effective auxiliary information.

[0014] The present invention provides an intelligent alarm system and method for power plant equipment monitoring. First, by using high-frequency sensors to collect physical quantity data such as temperature, pressure, vibration, etc., and combining laser radar, high-definition camera and other equipment for comprehensive scanning and data fusion to generate holographic images, more comprehensive and high-precision equipment operation information can be obtained, which greatly improves the accuracy and sensitivity of equipment status monitoring and helps to timely discover potential faults. Secondly, using space-time network technology to construct multi-dimensional space-time maps and topological data analysis methods can deeply explore the space-time characteristics and hidden patterns of equipment operation data, accurately predict the trend and scope of fault propagation, and effectively locate the fault point, providing a strong basis for fault diagnosis and improving the efficiency and accuracy of fault diagnosis. Furthermore, the evolutionary strategy optimization and memory-enhanced neural network prediction functions in the intelligent learning and adaptive optimization components enable the model to automatically optimize parameters according to actual operation data, make full use of historical experience and current information, enhance the adaptability to equipment status changes and prediction reliability, and reduce the need for manual intervention. In addition, the Bayesian network and counterfactual reasoning technology in the fault diagnosis and causal inference subsystem can accurately determine the cause of the fault based on probability relationships and hypothetical scenario analysis, especially in the case of multi-factor interaction in complex systems, effectively avoiding misdiagnosis. In terms of intelligent decision-making and collaborative operation, group intelligent collaborative diagnosis realizes global collaborative diagnosis through the collaborative mechanism between intelligent agents, improving diagnostic efficiency and accuracy; emotional computing-assisted decision-making takes into account the status of the operator, reduces the adverse effects of human factors on decision-making by adjusting the alarm display and providing prompt information, and ensures the accuracy of decision-making. Finally, good human-computer interaction functions realize efficient interaction between operators and systems, further assist operators in their work, improve overall work efficiency and system practicality, and provide more reliable guarantees for the safe and stable operation of power plant equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a flow chart of the intelligent alarm system for monitoring power plant equipment of the present invention; Figure 2 It is a flow chart of data collection and data fusion of the present invention; Figure 3It is the intelligent learning and adaptive optimization flow chart of the present invention; Figure 4 It is a flow chart of the intelligent decision-making and collaborative operation module of the present invention. DETAILED DESCRIPTION

[0016] Example 1 like Figure 1-4 As shown, an intelligent alarm system for power plant equipment monitoring includes a data acquisition and environment modeling module, a data fusion and deep analysis module, an intelligent learning and adaptive optimization component, and a fault diagnosis and causal inference subsystem; The data acquisition and environment modeling module collects the operating parameters of the equipment, scans and fuses the data through laser radar and high-definition cameras, and generates a holographic image containing static and dynamic information of the equipment; The data fusion and deep analysis module combines time series data and spatial information to construct a multi-dimensional spatiotemporal map, analyze the spatiotemporal evolution of the equipment operation status, and use topological data analysis methods to explore the data topology structure and identify hidden patterns and anomalies. The intelligent learning and adaptive optimization components include an evolutionary strategy optimization module and a memory-enhanced neural network prediction module; The fault diagnosis and causal inference subsystem includes a Bayesian network fault assessment module and a counterfactual reasoning root cause analysis module.

[0017] In the preferred solution, in the intelligent learning and adaptive optimization components: the evolution strategy optimization module introduces the evolution strategy ES algorithm to build an intelligent learning framework, and compares the model parameters to the genes of individual organisms; By simulating biological evolution mechanisms such as natural selection and crossover mutation, the model can automatically select and optimize parameter combinations based on actual operating data feedback, so that the model can adapt to changes in power plant operating conditions and improve adaptability to changes in equipment status and prediction accuracy; Adopt MANN and equip it with an external memory module to store a large amount of historical operation data and fault cases. MANN is a memory-enhanced neural network. When predicting the future operating status of the equipment, the current input data is combined with similar historical patterns in the memory module to make judgments, enhance the ability to capture potential failure trends, and improve the reliability and foresight of predictions.

[0018] In the preferred scheme, in the fault diagnosis and causal inference subsystem: a Bayesian network fault diagnosis model is constructed based on historical data and expert experience to determine the probabilistic dependency relationship between equipment operating parameters, environmental factors and various types of faults, and to assign conditional probabilities to the connections between network node variables; After accessing the real-time monitoring data, the Bayesian inference engine is used to calculate the posterior probability of different faults based on the probability relationship; Among them, the possibility of various faults when the equipment temperature rises is evaluated by combining the changes of multiple parameters, providing a quantitative decision-making basis for fault diagnosis; Use counterfactual reasoning techniques to simulate hypothetical scenarios in equipment operation, including simulation of hypothetical scenarios in standby operation in the event of abnormalities in key equipment or sensor measurement deviations; By comparing the differences in system states between actual and hypothetical scenarios, we can gain a deep understanding of the fault causal mechanism and accurately determine the root cause of the fault. This can be used to diagnose faults caused by the interaction of multiple factors in complex systems and avoid misdiagnosis.

[0019] In the preferred solution, the data acquisition of the data acquisition and environmental modeling module uses sensors to collect temperature, pressure, and vibration physical quantity data, with a sampling frequency of 10-60ms once, to obtain high-precision raw data and monitor subtle changes in the operating status of the equipment in real time; Use laser radar, high-definition cameras or 3D scanning equipment to conduct a comprehensive scan of the power plant's internal structure layout, equipment appearance and operating status from multiple angles; By fusing the point cloud data and image data obtained through scanning through complex algorithms, a holographic image containing the equipment's geometric shape, positional relationship and real-time operation dynamic information is generated in real time, providing an intuitive visual basis for subsequent analysis.

[0020] In the preferred solution, in the data fusion and in-depth analysis module: With the help of space-time network technology, the time series data of equipment operating parameters are combined with information such as the spatial location and connection relationship of the equipment in the power plant to construct a multi-dimensional space-time map; Analyze the connection relationship and data propagation path between nodes in the space-time graph, observe the evolution of the equipment operation status in the space-time dimension, predict the impact of the fault on adjacent equipment in space, its propagation trend over time, and the potential impact range; Use topological data analysis methods to calculate the topological invariants of data points in high-dimensional space to explore the topological structure of data; By identifying hidden patterns in the topological structure, anomalies and abnormal patterns can be effectively discovered in high-dimensional nonlinear data sets, and faults can be located in a timely manner based on changes in the topological distribution of data points when equipment fails.

[0021] The preferred solution also includes an intelligent decision-making and collaborative operation module and a human-computer interaction and user experience improvement module; The intelligent decision-making and collaborative operation module includes a group intelligent collaborative diagnosis module and an emotional computing auxiliary decision-making module; The AI ​​computing predictive maintenance module provides natural language dialogue services to assist operators and improve user experience.

[0022] In the preferred solution, in the intelligent decision-making and collaborative operation module: Swarm intelligence collaborative diagnosis module: Multiple intelligent agents with autonomous decision-making and local information processing capabilities are deployed in the system to monitor and preliminarily diagnose the equipment or areas they are responsible for. Global collaborative diagnosis is achieved through collaborative mechanisms such as pheromone transmission and sharing of local optimal solutions. When equipment is abnormal, intelligent agents quickly gather together for joint analysis and make comprehensive judgments to draw more accurate and comprehensive fault diagnosis conclusions, thereby improving diagnostic efficiency and accuracy. Emotional computing decision-making support module: uses high-definition cameras, microphones and physiological signal monitoring equipment to collect multimodal data on operators’ facial expressions, voice intonation, and heart rate, and analyzes their emotional state and workload; The operator status is taken into account during fault diagnosis and decision-making to assist in determining the impact of human factors on operational errors or misjudgments, such as adjusting alarm displays or providing prompt information based on stressful conditions to help operators make accurate decisions.

[0023] 1. Data Collection and Environmental Modeling: Data Collection: Install high-precision sensors on core equipment such as boilers, turbines, and generators to collect data on physical quantities such as temperature, pressure, and vibration at a sampling frequency of 10-60ms. For example, in the superheater part of the boiler, the sensor collects temperature data every 30ms to monitor its subtle changes in real time. Through continuous monitoring, it is possible to capture temperature fluctuations of 0.1°C in a short period of time, providing high-precision raw data for subsequent analysis.

[0024] Environmental modeling: Multiple laser radars, high-definition cameras and 3D scanning equipment are used to scan the interior of the power plant from different angles. These devices are distributed in key locations in the plant, such as the top of the boiler and on both sides of the steam turbine. The point cloud data and image data obtained by scanning are fused through complex algorithms to generate a holographic image containing the geometric shape, positional relationship and real-time operation dynamic information of the equipment. In the holographic image, the rotation of the turbine rotor and the opening and closing status of the valve can be clearly seen, providing an intuitive visual basis for subsequent analysis.

[0025] 2. Data fusion and in-depth analysis: Constructing a multi-dimensional space-time map: With the help of space-time network technology, the collected time series data of equipment operation parameters are combined with the spatial position and connection relationship information of the equipment in the power plant. For example, for the flow data of the boiler feed pump, a multi-dimensional space-time map is constructed in combination with its position in the pipeline system and its connection relationship with other equipment. By defining the node connection strength function, the spatiotemporal characteristics of the equipment operation data are accurately reflected. During the analysis process, it was found that when the flow of the feed pump fluctuated abnormally within a certain period of time, the connection strength between it and the adjacent pipeline pressure node was significantly enhanced, providing important clues for fault prediction.

[0026] Topological data analysis: Use topological data analysis methods to calculate the topological invariants of data points in high-dimensional space. Taking the voltage, current and other data point sets of the generator as an example, the topological structure of the data is explored by defining a topological similarity measurement function. Before a fault occurs, by identifying hidden patterns in the topological structure, it is found that the distribution of voltage and current data in high-dimensional space has abnormal aggregation, and abnormal points and abnormal patterns are effectively discovered, so as to locate the possible fault site in time, providing key information for subsequent fault diagnosis.

[0027] 3. Intelligent learning and adaptive optimization: Evolutionary strategy optimization: The evolutionary strategy ES algorithm is introduced to build an intelligent learning framework, and the model parameters are compared to the genes of individual organisms. In the process of optimizing the turbine efficiency prediction model, according to the actual operation data feedback, the optimization parameter combination is automatically screened by simulating biological evolution mechanisms such as natural selection and crossover mutation. After a period of operation, the model has improved the efficiency prediction accuracy of the turbine under different load conditions by 15%, effectively adapted to the changes in the power plant operating conditions, and improved the adaptability and prediction accuracy of equipment status changes.

[0028] Memory-enhanced neural network prediction: MANN is used and equipped with an external memory module to store a large amount of historical operation data and fault cases. When predicting the combustion stability of the boiler, the current input data such as fuel flow and air flow are comprehensively judged with similar historical patterns in the memory module. For example, when the fuel quality changes, the system can quickly retrieve the combustion state change pattern under similar circumstances from the memory module, enhancing the ability to capture potential fault trends, and increasing the advance warning time of combustion faults from the original 10 minutes to 30 minutes, improving the reliability and foresight of predictions.

[0029] 4. Fault diagnosis and causal inference: Bayesian network fault assessment: Based on the power plant's many years of historical data and expert experience, a Bayesian network fault diagnosis model was constructed to determine the probabilistic dependencies between equipment operating parameters, environmental factors, and various types of faults. For generator fault diagnosis, parameters such as temperature, humidity, winding insulation resistance, and surrounding environmental factors are considered to assign conditional probabilities to connections between network node variables. When connected to real-time monitoring data, the Bayesian inference engine is used to accurately calculate the posterior probabilities of different faults. In an actual fault, by combining multiple parameter changes, it was quickly assessed that the fault was caused by a winding short circuit, providing a quantitative decision-making basis for fault diagnosis and greatly shortening the troubleshooting time.

[0030] Counterfactual reasoning root cause analysis: Use counterfactual reasoning technology to simulate hypothetical scenarios in equipment operation. For example, assume that a certain control valve of a steam turbine is abnormally closed. By comparing the difference in system status between the actual and hypothetical scenarios, the cause and effect mechanism of the fault is deeply analyzed. In a speed control system fault troubleshooting, by simulating the hypothetical scenario of sensor measurement deviation, it is accurately determined that the erroneous speed control command is caused by sensor failure, thereby accurately determining the root cause of the fault, effectively avoiding misdiagnosis in the case of complex system multi-factor interaction, and improving the accuracy and reliability of fault diagnosis.

[0031] 5. Intelligent decision-making and collaborative operation: Group intelligent collaborative diagnosis: 20 intelligent agents with autonomous decision-making and local information processing capabilities are deployed in the system, responsible for monitoring and preliminary diagnosis of equipment in different areas. For example, 10 intelligent agents are responsible for the boiler area, and the other 10 intelligent agents are responsible for the turbine and generator areas. When an abnormality occurs in a burner of the boiler, the intelligent agent responsible for the area quickly detects the abnormality and exchanges information with other intelligent agents through collaborative mechanisms such as pheromone transmission and sharing of local optimal solutions. Under the guidance of pheromone concentration, the intelligent agents collaborate efficiently. After 3 iterations, the comprehensive judgment draws an accurate and comprehensive fault diagnosis conclusion, shortening the fault diagnosis time from the original 30 minutes to 10 minutes, improving the diagnostic efficiency and accuracy.

[0032] Emotional computing aids decision-making: Use high-definition cameras, microphones, and physiological signal monitoring equipment to collect multimodal data such as facial expressions, voice intonation, and heart rate of operators. During an emergency fault handling process, the system detected that the operator's heart rate was significantly accelerated and his voice tone became tense. Through the analysis of the emotional state evaluation function and the workload evaluation function, it was determined that the operator was in a highly nervous state and had a heavy workload. At this time, the system automatically adjusts the alarm display intensity according to the decision-making assistance coefficient, enlarges the font of the alarm information by 50%, and displays it in red flashing. At the same time, it provides detailed operation prompt information, such as checking the status of specific valves and the steps to adjust equipment parameters, to help operators make accurate decisions, effectively reducing the impact of human factors on operational errors or misjudgments.

[0033] 6. Improved human-computer interaction and user experience: Operators can interact with the system through natural language.

[0034] example: When the operator asks "What is the current operating efficiency of the boiler?", the system understands the question through natural language processing technology, retrieves relevant data from the database, and answers the operator in clear voice and chart form, such as "The current operating efficiency of the boiler is 85%, which is 3% lower than yesterday. The main reason is that the fuel quality has slightly decreased." At the same time, the system automatically recommends relevant equipment maintenance suggestions and operating precautions based on the operator's operating habits and historical query records, such as "It is recommended to focus on checking the wear of the burner during the next maintenance", which improves the user experience, realizes efficient interaction between operators and the system, and assists operators to better complete their work tasks.

[0035] Through practical application in the power plant, the intelligent alarm system and method effectively improved the accuracy and timeliness of equipment monitoring, reduced the equipment failure rate, ensured the safe and stable operation of the power plant, and provided strong support for the efficient production of the power plant.

[0036] Example 2 Further illustrate with reference to Example 1, Figure 1-4 As shown, the method includes: S1, data acquisition and environmental modeling: using sensors to collect temperature, pressure, and vibration physical quantity data of power plant equipment at a sampling frequency of 1060ms, obtaining high-precision raw data, and monitoring subtle changes in the operating status of the equipment in real time; Use laser radar, high-definition cameras or 3D scanning equipment to conduct a comprehensive scan of the power plant's internal structure layout, equipment appearance and operating status from multiple angles; The point cloud data and image data obtained by scanning are fused through algorithms to generate holographic images containing the geometric shape, position relationship and real-time operation dynamic information of the equipment, providing intuitive visual basis for subsequent analysis; High-precision sensors are carefully deployed at each important equipment node of the power plant. These sensors collect data strictly at a frequency of 10-60ms. Take a large steam turbine generator set as an example. The temperature sensor at the bearing part records the temperature data every 20ms. The pressure sensor and vibration sensor also collect data at a high frequency synchronously, so as to keenly capture the subtle fluctuations of the equipment's operating status and obtain extremely accurate raw data, laying a solid foundation for subsequent analysis.

[0037] At the same time, laser radars, high-definition cameras and 3D scanning equipment are rationally arranged around the entire power plant building. These devices conduct all-round and blind-angle scanning of the internal structure layout, equipment appearance and operating status of the power plant from different directions and angles. Through advanced algorithms, the collected point cloud data and image data are deeply integrated to successfully generate detailed and vivid holographic images. In this holographic image, not only can the overall outline of the steam turbine generator set and the precise position relationship of each component be clearly seen, but also the high-speed rotation of the rotor, the precise opening and closing of the valve and other dynamic information can be observed in real time, which provides an intuitive and comprehensive visual basis for subsequent in-depth analysis and greatly enhances the ability to control the overall status of the equipment.

[0038] S2, Data Fusion and In-depth Analysis Module Steps: With the help of spatiotemporal network technology, the time series data of equipment operation parameters are combined with the spatial location and connection relationship information of the equipment in the power plant to construct a multi-dimensional spatiotemporal map, where the construction of the spatiotemporal map should accurately reflect the spatiotemporal characteristics of the equipment operation data; Analyze the constructed spatiotemporal graph, including studying the connection relationship between nodes and the data propagation path, to observe the evolution of the equipment operation status in the spatiotemporal dimension, and then accurately predict the impact of the fault on adjacent equipment in space, the propagation trend over time, and the potential impact range; Use topological data analysis methods to calculate the topological invariants of data points in high-dimensional space to explore the topological structure of data. By identifying hidden patterns in the topological structure, abnormal points and abnormal patterns can be effectively found in high-dimensional nonlinear data sets. The fault can be located in time according to the changes in the topological distribution of data points when the equipment fails. The application of topological data analysis methods should have high accuracy and reliability. With the help of cutting-edge space-time network technology, the time series data of the operating parameters of equipment such as steam turbine generator sets, boilers, transformers, etc. are closely combined with their location information in the spatial layout of the power plant and the connection relationship between them, to construct a complex and sophisticated multi-dimensional space-time map. For example, for the steam generation and transportation system of the boiler, by accurately analyzing the connection strength function between each node (such as different heating surface parts, steam pipeline nodes, etc.), the relationship between the changes in key parameters such as steam flow and pressure in time and space is fully considered to ensure that the space-time map can accurately reflect the space-time characteristics of the equipment operation data.

[0039] When conducting in-depth analysis of the constructed spatiotemporal graph, the focus is on studying the connection relationship between nodes and the propagation path of data in them. Through long-term data tracking and analysis, it is found that when a burner of a boiler has unstable combustion, it can be clearly observed in the spatiotemporal graph that the connection strength between its related nodes (such as fuel supply nodes, air delivery nodes, steam production nodes, etc.) changes significantly, thereby accurately predicting the impact that the fault may have on adjacent equipment (such as superheaters, reheaters, etc.) in space, as well as the propagation trend over time and the potential impact range.

[0040] Using advanced topological data analysis methods, we calculated the topological invariants of massive equipment operation data points in high-dimensional space. Taking the electrical parameter data of the generator as an example, we deeply explored the topological structure of the data through a carefully defined topological similarity measurement function. Before a local short circuit fault occurred in the stator winding of a generator, we successfully discovered the abnormal distribution pattern of current and voltage data in high-dimensional space by keenly identifying the hidden patterns in the topological structure, and effectively located the potential fault points. This fully demonstrated the high accuracy and reliability of the topological data analysis method in discovering abnormal points and abnormal patterns, and provided key clues for taking preventive measures in a timely manner.

[0041] S3, Intelligent learning and adaptive optimization component steps: Evolution strategy optimization step, that is, introducing the evolution strategy ES algorithm to build an intelligent learning framework, analogizing the model parameters to the genes of individual organisms, simulating biological evolution mechanisms such as natural selection and crossover mutation, and automatically screening and optimizing parameter combinations based on actual operating data feedback, so that the model can adapt to changes in power plant operating conditions and improve the adaptability and prediction accuracy of equipment status changes. The implementation process of the evolution strategy should ensure the effectiveness and stability of parameter optimization; The memory-enhanced neural network prediction step uses MANN and is equipped with an external memory module to store a large amount of historical operation data and fault cases. When predicting the future operation status of the equipment, it combines the current input data with similar historical patterns in the memory module to make judgments, enhance the ability to capture potential fault trends, and improve the reliability and foresight of predictions. The prediction process of the memory-enhanced neural network should make full use of historical data and current information to ensure the reliability of the prediction results; In the process of evolutionary strategy optimization, the innovative evolutionary strategy ES algorithm is introduced to build an intelligent learning framework, and the various parameters in the model are cleverly compared to the genes of individual organisms. Taking the energy consumption prediction model of the power system of a power plant as an example, through a large amount of actual operation data feedback, the biological evolution process such as the survival of the fittest and the survival of the fittest in natural selection, as well as cross-mutation, is simulated to automatically screen and optimize the model parameters. After a period of continuous optimization, the prediction accuracy of the energy consumption of the power system under different working conditions has increased by more than 20%, effectively adapting to the complex and changeable working conditions of the power plant, significantly improving the adaptability to equipment status changes and prediction accuracy, and strictly ensuring the effectiveness and stability of parameter optimization during the implementation of the entire evolutionary strategy.

[0042] In terms of memory-enhanced neural network prediction, MANN is used and equipped with a large-capacity external memory module to store historical operating data and various fault cases for several years. When predicting the combustion efficiency change trend of the boiler, the current input data such as fuel type, coal supply, air supply volume and similar historical patterns in the memory module are comprehensively judged. For example, when the fuel supply channel changes and causes the fuel characteristics to change, the system can quickly retrieve the combustion efficiency change pattern and related countermeasures under similar fuel conditions in the past from the memory module, greatly enhancing the ability to capture potential fault trends. Through practical application verification, this prediction method has extended the advance warning time of abnormal combustion efficiency by an average of 40%, which has greatly improved the reliability and foresight of the prediction, and fully utilized the rich historical data and current real-time information in the prediction process to ensure the high reliability of the prediction results.

[0043] S4, Fault diagnosis and causal inference module steps: Construct a Bayesian network fault diagnosis model based on historical data and expert experience, determine the probability dependency between equipment operating parameters, environmental factors and various types of faults, and assign conditional probabilities to the connections between network node variables. The construction of the Bayesian network model should be based on sufficient historical data and accurate expert experience; After accessing the real-time monitoring data, the Bayesian reasoning engine is used to calculate the posterior probability of different faults based on the probability relationship, and the possibility of various faults when the equipment temperature rises is evaluated by combining the changes in multiple parameters, providing a quantitative decision-making basis for fault diagnosis. The calculation process of the Bayesian reasoning engine should be accurate and efficient; Use counterfactual reasoning technology to simulate hypothetical scenarios in equipment operation, key equipment anomalies or sensor measurement deviations, etc. By comparing the differences in system states between actual and hypothetical scenarios, we can deeply understand the fault causal mechanism and accurately determine the root cause of the fault. This can be used to diagnose faults caused by the interaction of multiple factors in complex systems to avoid misdiagnosis. The application of counterfactual reasoning technology should be able to effectively reveal the causal relationship of faults; Based on the massive historical data accumulated by the power plant and the rich experience of senior experts, a Bayesian network fault diagnosis model was carefully constructed to comprehensively determine the complex probabilistic dependencies between equipment operating parameters (such as temperature, pressure, flow, electrical parameters, etc.), environmental factors (such as humidity, dust concentration, air temperature, etc.) and various types of faults, and accurately assign conditional probabilities to the connections between network node variables. Taking transformer fault diagnosis as an example, operating parameters such as oil temperature, winding insulation resistance, load current, as well as environmental humidity, surrounding electromagnetic interference and other factors were considered to construct a detailed Bayesian network model.

[0044] After accessing the real-time monitoring data, the efficient Bayesian inference engine is used to quickly calculate the posterior probability of different faults strictly based on the pre-set probability relationship. In the case of abnormal increase in transformer oil temperature, by combining the real-time changes of multiple parameters, it is accurately assessed that the fault caused by winding insulation aging is the most likely, providing an accurate quantitative decision-making basis for fault diagnosis, greatly shortening the time and scope of fault investigation.

[0045] At the same time, advanced counterfactual reasoning technology is used to simulate various hypothetical scenarios in equipment operation, such as assuming that a component of a key equipment fails abnormally or there is a deviation in sensor measurement. In a turbine speed control system fault analysis, by simulating the scenario of the speed control valve opening sensor measurement deviation, the difference in system status between the actual and hypothetical scenarios was compared, and the fault cause mechanism was deeply analyzed. It was accurately determined that the root cause of the fault was the zero drift problem of the sensor, which effectively avoided misdiagnosis in the case of complex system multi-factor interaction and significantly improved the accuracy and reliability of fault diagnosis.

[0046] S5. Intelligent decision-making and collaborative operation steps: Group intelligent collaborative diagnosis steps, deploy multiple intelligent agents with autonomous decision-making and local information processing capabilities in the system, responsible for monitoring and preliminary diagnosis of the equipment or areas they are responsible for, and use collaborative mechanisms such as pheromone transmission and sharing of local optimal solutions to achieve global collaborative diagnosis. When the equipment is abnormal, the intelligent agents quickly gather together for joint analysis, and make comprehensive judgments to draw more accurate and comprehensive fault diagnosis conclusions, thereby improving diagnostic efficiency and accuracy. The process of group intelligent collaborative diagnosis should ensure smooth collaboration between intelligent agents and reliable diagnostic results; The emotional computing-assisted decision-making step uses high-definition cameras, microphones, and physiological signal monitoring equipment to collect multimodal data on operators’ facial expressions, voice intonation, and heart rate, analyze their emotional state and workload, consider the operator’s state during fault diagnosis and decision-making, and assist in determining the impact of human factors on operational errors or misjudgments; Adjust the alarm display or provide prompt information according to the stress state to help operators make accurate decisions. The process of emotional computing to assist decision-making should be able to accurately analyze the operator's state and provide effective auxiliary information; In terms of group intelligent collaborative diagnosis, 30 agents with strong autonomous decision-making and efficient local information processing capabilities are reasonably deployed in the entire power plant system, responsible for close monitoring and preliminary diagnosis of different types of equipment or specific areas. For example, 10 agents focus on monitoring boilers and auxiliary equipment, 10 agents are responsible for monitoring the operating status of steam turbine generator sets, and the other 10 agents monitor the electrical system in real time. When the induced draft fan of the boiler vibrates abnormally, the agent responsible for the area immediately detects the abnormality and sends a signal to the surrounding agents through the pheromone transmission mechanism, while sharing local preliminary diagnostic information. Under the synergistic effect of the dynamic guidance of pheromone concentration and the sharing mechanism of local optimal solutions, the agents quickly gather and carry out comprehensive and in-depth analysis. After multiple iterations of optimization, comprehensive judgments are made to obtain accurate and comprehensive fault diagnosis conclusions, shortening the fault diagnosis time that may have taken several hours to less than 1 hour, greatly improving the diagnostic efficiency and accuracy, and ensuring smooth collaboration between agents and highly reliable diagnostic results throughout the group intelligent collaborative diagnosis process.

[0047] In the emotional computing-assisted decision-making process, high-definition cameras, high-sensitivity microphones and advanced physiological signal monitoring equipment are used to collect multimodal data such as facial expressions, voice intonation, heart rate, etc. of operators during work. During the emergency handling process of a power plant, the system analyzed the collected data in real time and found that the operator's facial expression was tense, his voice intonation was rapid, and his heart rate was significantly accelerated. After comprehensive analysis by the emotional state evaluation function and the workload evaluation function, it was determined that the operator was in a state of high tension and workload overload. At this time, the system automatically adjusts the alarm display mode according to the decision-making assistance coefficient, adjusts the display color of the alarm information from the conventional yellow to the eye-catching red, and highlights it in a flashing manner. At the same time, the volume of the voice prompt is increased by 50%, and detailed operation prompt information is provided according to the predefined rule base, such as checking whether a specific safety valve is closed, the operation steps of starting the backup equipment, etc., which effectively assists the operator to make accurate decisions and significantly reduces the impact of human factors on operational errors or misjudgments.

[0048] S6. Based on the human-computer interaction function of the system, efficient interaction between the operator and the system is achieved to assist the operator in his work.

[0049] The intelligent alarm system is equipped with a powerful human-computer interaction function, and operators can interact with the system efficiently through natural and smooth voice or concise and clear text input. For example, when an operator wants to understand the recent operating trend of a certain equipment, he only needs to ask the system a simple question: "Please show the operating efficiency trend chart of the No. 3 steam turbine generator unit in the past week." The system can quickly understand the operator's intention, retrieve relevant data from the massive database in a short time, and present it on the operation terminal screen in the form of intuitive and clear charts, while using voice to explain in detail the key information and changing trends in the chart. In addition, the system can also automatically recommend relevant equipment maintenance suggestions, operating skills, and safety precautions based on the operator's historical operation records and personal preferences, such as "Based on your previous operating habits and current equipment operating conditions, it is recommended that you focus on checking the bearing lubrication of the No. 3 steam turbine generator unit during the next inspection." It greatly improves the user experience, realizes in-depth interaction and efficient collaboration between operators and the system, provides strong auxiliary support for the daily work of operators, and effectively improves the overall operation and management efficiency of the power plant.

[0050] Through practical application in this power plant, it has been fully demonstrated that this intelligent alarm system and method have significant advantages and excellent effects in improving equipment monitoring accuracy, enhancing fault diagnosis capabilities, optimizing decision-making processes, and improving human-computer interaction experience, providing a solid technical guarantee for the safe, stable and efficient operation of the power plant.

[0051] Example 3 Further illustrate with reference to Example 2, Figure 1-4 As shown, steps S2-S4 are specifically as follows: S2.1. Data fusion and deep analysis module steps: With the help of spatiotemporal network technology, the time series data of equipment operating parameters are combined with the spatial location and connection relationship information of the equipment in the power plant to construct a multi-dimensional spatiotemporal map. ,in is a node set, is the edge set, is the time dimension information; define the node connection strength function for: ; in, and Node and In time The key operating parameters of is the time window size; this function is used to construct a spatiotemporal graph to accurately reflect the spatiotemporal characteristics of the equipment operation data; Analyze the constructed spatiotemporal graph and define the factors affecting fault propagation for: ; in, For Node In time The parameter change, To analyze the time window; through research To observe the evolution of the equipment's operating status in the time and space dimensions, and then accurately predict the impact of the fault on adjacent equipment in space, as well as its propagation trend over time and potential impact range; The topological data analysis method is used to calculate the topological invariant TI of data points in high-dimensional space. The data point set is , define the topological similarity metric function for: ; in, For data points In the The value under the feature map, is the adjustment coefficient, is the number of feature maps; by calculating TI, the topological structure of the data is explored, and by identifying the hidden patterns in the topological structure, abnormal points and abnormal patterns are effectively found in high-dimensional nonlinear data sets, and the fault is located in time according to the change in the topological distribution of data points when the equipment fails. The application of topological data analysis methods should have high accuracy and reliability; S3.1, Intelligent learning and adaptive optimization component steps: Evolution strategy optimization step, introduce the evolution strategy ES algorithm to build an intelligent learning framework, set the model parameter vector , define the fitness function for:

[0052] in, is the actual device operation status value, is the model prediction value, is the sample size, is the regularization coefficient; by simulating biological evolution mechanisms such as natural selection and crossover mutation, the optimization parameter combination is automatically screened based on actual operation data feedback, so that the model can adapt to the changes in power plant operating conditions and improve the adaptability and prediction accuracy of equipment status changes. The implementation process of the evolutionary strategy should ensure the effectiveness and stability of parameter optimization; The memory-enhanced neural network prediction step uses MANN and is equipped with an external memory module to store a large amount of historical operation data and fault cases; the current input data is , the historical pattern stored in the memory module is ( ), define the pattern matching function for: ; When predicting the future operating status of the device, the current input data is combined with similar historical patterns in the memory module to make a judgment; let the prediction value be , the calculation formula is: ; in, , Based on historical model The prediction value of the memory-enhanced neural network should be fully utilized to capture the potential fault trend and improve the prediction reliability and foresight. S4.1, Fault diagnosis and causal inference module steps: Based on historical data and expert experience, a Bayesian network fault diagnosis model is constructed to determine the probabilistic dependency between equipment operating parameters, environmental factors and various types of faults, and to assign conditional probabilities to the connections between network node variables; the equipment operating parameter vector is , the fault variable is , the environmental factor vector is , determine the conditional probability distribution based on a large amount of historical data and expert knowledge ; After accessing the real-time monitoring data, the Bayesian inference engine is used to calculate the posterior probability of different faults according to the probability relationship; assuming that the real-time monitoring data is and , according to Bayes' theorem, the fault The posterior probability The calculation formula is: ; in, The number of fault types; Combined with the changes in multiple parameters, the possibility of various faults when the equipment temperature rises is evaluated to provide a quantitative decision-making basis for fault diagnosis. The calculation process of the Bayesian reasoning engine should be accurate and efficient; The counterfactual reasoning technique is used to simulate the hypothetical scenario of equipment operation. The change of key equipment parameters is assumed to be , the sensor measurement deviation is , define the system state difference measurement function for: ; in, , and is the weight coefficient, It is the fault prediction value under normal circumstances; by comparing the differences in system states under actual and hypothetical scenarios, we can deeply understand the fault causal mechanism and accurately determine the root cause of the fault. It is used for fault diagnosis caused by the interaction of multiple factors in complex systems to avoid misdiagnosis. The application of counterfactual reasoning technology should be able to effectively reveal the fault causal relationship.

[0053] Example 4 Further illustrate with reference to Examples 2-3, Figure 1-4 As shown, step S5 is specifically as follows: Group intelligence collaborative diagnosis step: the system is deployed with Agents, each agent The data obtained from monitoring the equipment or area for which it is responsible is , the fault probability vector obtained after local preliminary diagnosis is ,in is the number of fault types; Defining Agents and Agent The pheromone concentration between , and its update formula is: ; in, is the pheromone volatility coefficient, ranging from arrive between, For the agent in this iteration and Agent The increment of pheromone between Based on the local diagnosis results, it is considered that When the diagnostic results are highly correlated, ,in Increase the strength factor for pheromones. and is the adjustment parameter; this formula is used to measure the degree of information correlation between agents and guide the collaboration between agents by updating the pheromone concentration; When the intelligent Based on the local diagnosis results, it is considered that The correlation of the diagnosis results is mainly calculated by calculating the local fault probability vector and the agent The similarity of the fault possibility vector is assumed to be ,when , is the set similarity threshold; Agent Update your own diagnosis results according to the pheromone concentration and the fault probability vector of other agents. The update formula is: ; in, , for the intelligent agent For Agents After multiple iterations, until the fault probability vectors of all agents converge, the results of all agents are finally combined to obtain an accurate and comprehensive fault diagnosis conclusion. During the whole process, the communication between agents is stable and the data sharing is timely, which improves the efficiency and accuracy of diagnosis. Emotional computing aided decision-making steps: Assume that the facial expression feature vector of the operator collected by high-definition camera, microphone and physiological signal monitoring equipment is , the speech intonation feature vector is , the heart rate characteristic value is HR; Define the emotional state evaluation function as , the calculation formula is: ; in, , and is the weight coefficient of different modal data, is the average heart rate of the operator during normal work. is the standard deviation of heart rate; this function is used to comprehensively evaluate the emotional state of the operator, and its value range is arrive The closer It means the mood is more tense or unstable; Assume that the operator's workload evaluation function is ,in is the amount of equipment data and task complexity index that the current operator needs to process, and the calculation formula is: ; in, For the Quantitative indicators of data or tasks, and is the adjustment index, is the total number of data or tasks; this function is used to evaluate the workload of operators, and its value range is arrive The closer It means the workload is greater; The decision support coefficient is defined as ,when When the system believes that the operator may be affected by emotions or workload, the decision risk increases. is the set threshold; at this time, the system adjusts the alarm display intensity according to the following formula : ; in, Displays the intensity of the original alarm. is the intensity adjustment coefficient; at the same time, the system provides prompt information according to the predefined rule base; When DAC is in a certain range, it prompts the operator to check specific equipment or review the operation steps to help the operator make accurate decisions. The process of emotional computing-assisted decision-making should be able to accurately analyze the operator's status and provide effective auxiliary information.

[0054] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. An intelligent alarm system for monitoring power plant equipment, characterized by: It includes data acquisition and environment modeling module, data fusion and deep analysis module, intelligent learning and adaptive optimization component and fault diagnosis and causal inference subsystem; The data acquisition and environment modeling module collects the operating parameters of the equipment, scans and fuses the data through laser radar and high-definition cameras, and generates a holographic image containing static and dynamic information of the equipment; The data fusion and deep analysis module combines time series data and spatial information to construct a multi-dimensional spatiotemporal map, analyze the spatiotemporal evolution of the equipment operation status, and use topological data analysis methods to explore the data topology structure and identify hidden patterns and anomalies. The intelligent learning and adaptive optimization components include an evolutionary strategy optimization module and a memory-enhanced neural network prediction module; The fault diagnosis and causal inference subsystem includes a Bayesian network fault assessment module and a counterfactual reasoning root cause analysis module.

2. According to claim 1, an intelligent alarm system for monitoring power plant equipment is characterized by: In the intelligent learning and adaptive optimization components: the evolution strategy optimization module introduces the evolution strategy ES algorithm to build an intelligent learning framework, and compares the model parameters to the genes of individual organisms; By simulating biological evolution mechanisms such as natural selection and crossover mutation, the model can automatically select and optimize parameter combinations based on actual operating data feedback, so that the model can adapt to changes in power plant operating conditions and improve adaptability to changes in equipment status and prediction accuracy; Adopt MANN and equip it with an external memory module to store a large amount of historical operation data and fault cases. MANN is a memory-enhanced neural network. When predicting the future operating status of the equipment, the current input data is combined with similar historical patterns in the memory module to make judgments, enhance the ability to capture potential failure trends, and improve the reliability and foresight of predictions.

3. The intelligent alarm system for monitoring power plant equipment according to claim 1 is characterized by: In the fault diagnosis and causal inference subsystem: a Bayesian network fault diagnosis model is constructed based on historical data and expert experience to determine the probabilistic dependency between equipment operating parameters, environmental factors and various types of faults, and to assign conditional probabilities to the connections between network node variables; After accessing the real-time monitoring data, the Bayesian inference engine is used to calculate the posterior probability of different faults based on the probability relationship; Among them, the possibility of various faults when the equipment temperature rises is evaluated by combining the changes of multiple parameters, providing a quantitative decision-making basis for fault diagnosis; Use counterfactual reasoning techniques to simulate hypothetical scenarios in equipment operation, including simulation of hypothetical scenarios in standby operation in the event of abnormalities in key equipment or sensor measurement deviations; By comparing the differences in system states between actual and hypothetical scenarios, we can gain a deep understanding of the fault causal mechanism and accurately determine the root cause of the fault. This can be used to diagnose faults caused by the interaction of multiple factors in complex systems and avoid misdiagnosis.

4. The intelligent alarm system for monitoring power plant equipment according to claim 1 is characterized in that: The data acquisition and environmental modeling module uses sensors to collect temperature, pressure, and vibration data at a sampling frequency of 10-60ms to obtain high-precision raw data and monitor subtle changes in the operating status of the equipment in real time; Use laser radar, high-definition cameras or 3D scanning equipment to conduct a comprehensive scan of the power plant's internal structure layout, equipment appearance and operating status from multiple angles; By fusing the point cloud data and image data obtained through scanning through complex algorithms, a holographic image containing the equipment's geometric shape, positional relationship and real-time operation dynamic information is generated in real time, providing an intuitive visual basis for subsequent analysis.

5. The intelligent alarm system for monitoring power plant equipment according to claim 1 is characterized in that: In the data fusion and deep analysis module: With the help of space-time network technology, the time series data of equipment operating parameters are combined with information such as the spatial location and connection relationship of the equipment in the power plant to construct a multi-dimensional space-time map; Analyze the connection relationship and data propagation path between nodes in the space-time graph, observe the evolution of the equipment operation status in the space-time dimension, predict the impact of the fault on adjacent equipment in space, its propagation trend over time, and the potential impact range; Use topological data analysis methods to calculate the topological invariants of data points in high-dimensional space to explore the topological structure of data; By identifying hidden patterns in the topological structure, anomalies and abnormal patterns can be effectively discovered in high-dimensional nonlinear data sets, and faults can be located in a timely manner based on changes in the topological distribution of data points when equipment fails.

6. The intelligent alarm system for monitoring power plant equipment according to claim 1 is characterized by: It also includes intelligent decision-making and collaborative operation modules and human-computer interaction and user experience improvement modules; The intelligent decision-making and collaborative operation module includes a group intelligent collaborative diagnosis module and an emotional computing auxiliary decision-making module; The AI ​​computing predictive maintenance module provides natural language dialogue services to assist operators and improve user experience.

7. According to claim 1, the intelligent alarm system for monitoring power plant equipment is characterized in that: include: In the intelligent decision-making and collaborative operation module: Swarm intelligence collaborative diagnosis module: Multiple intelligent agents with autonomous decision-making and local information processing capabilities are deployed in the system to monitor and preliminarily diagnose the equipment or areas they are responsible for. Global collaborative diagnosis is achieved through collaborative mechanisms such as pheromone transmission and sharing of local optimal solutions. When equipment is abnormal, intelligent agents quickly gather together for joint analysis and make comprehensive judgments to draw more accurate and comprehensive fault diagnosis conclusions, thereby improving diagnostic efficiency and accuracy. Emotional computing decision-making support module: uses high-definition cameras, microphones and physiological signal monitoring equipment to collect multimodal data on operators’ facial expressions, voice intonation, and heart rate, and analyzes their emotional state and workload; The operator status is taken into account during fault diagnosis and decision-making to assist in determining the impact of human factors on operational errors or misjudgments, such as adjusting alarm displays or providing prompt information based on stressful conditions to help operators make accurate decisions.

8. The alarm method of the intelligent alarm system for power plant equipment monitoring according to any one of claims 1 to 7, characterized in that: S1. Data collection and environmental modeling: Use sensors to collect temperature, pressure, and vibration data of power plant equipment at a sampling frequency of 1060ms to obtain high-precision raw data and monitor subtle changes in equipment operating status in real time; Use laser radar, high-definition cameras or 3D scanning equipment to conduct a comprehensive scan of the power plant's internal structure layout, equipment appearance and operating status from multiple angles; The point cloud data and image data obtained by scanning are fused through algorithms to generate holographic images containing the geometric shape, position relationship and real-time operation dynamic information of the equipment, providing intuitive visual basis for subsequent analysis; S2, Data Fusion and In-depth Analysis Module Steps: With the help of spatiotemporal network technology, the time series data of equipment operation parameters are combined with the spatial location and connection relationship information of the equipment in the power plant to construct a multi-dimensional spatiotemporal map, where the construction of the spatiotemporal map should accurately reflect the spatiotemporal characteristics of the equipment operation data; Analyze the constructed spatiotemporal graph, including studying the connection relationship between nodes and the data propagation path, to observe the evolution of the equipment operation status in the spatiotemporal dimension, and then accurately predict the impact of the fault on adjacent equipment in space, the propagation trend over time, and the potential impact range; Use topological data analysis methods to calculate the topological invariants of data points in high-dimensional space to explore the topological structure of data. By identifying hidden patterns in the topological structure, abnormal points and abnormal patterns can be effectively found in high-dimensional nonlinear data sets. The fault can be located in time according to the changes in the topological distribution of data points when the equipment fails. The application of topological data analysis methods should have high accuracy and reliability. S3, Intelligent learning and adaptive optimization component steps: Evolution strategy optimization step, that is, introducing the evolution strategy ES algorithm to build an intelligent learning framework, analogizing the model parameters to the genes of individual organisms, simulating biological evolution mechanisms such as natural selection and crossover mutation, and automatically screening and optimizing parameter combinations based on actual operating data feedback, so that the model can adapt to changes in power plant operating conditions and improve the adaptability and prediction accuracy of equipment status changes. The implementation process of the evolution strategy should ensure the effectiveness and stability of parameter optimization; The memory-enhanced neural network prediction step uses MANN and is equipped with an external memory module to store a large amount of historical operation data and fault cases. When predicting the future operation status of the equipment, it combines the current input data with similar historical patterns in the memory module to make judgments, enhance the ability to capture potential fault trends, and improve the reliability and foresight of predictions. The prediction process of the memory-enhanced neural network should make full use of historical data and current information to ensure the reliability of the prediction results; S4, Fault diagnosis and causal inference module steps: Construct a Bayesian network fault diagnosis model based on historical data and expert experience, determine the probability dependency between equipment operating parameters, environmental factors and various types of faults, and assign conditional probabilities to the connections between network node variables. The construction of the Bayesian network model should be based on sufficient historical data and accurate expert experience; After accessing the real-time monitoring data, the Bayesian reasoning engine is used to calculate the posterior probability of different faults based on the probability relationship, and the possibility of various faults when the equipment temperature rises is evaluated by combining the changes in multiple parameters, providing a quantitative decision-making basis for fault diagnosis. The calculation process of the Bayesian reasoning engine should be accurate and efficient; Use counterfactual reasoning technology to simulate hypothetical scenarios in equipment operation, key equipment anomalies or sensor measurement deviations, etc. By comparing the differences in system states between actual and hypothetical scenarios, we can deeply understand the fault causal mechanism and accurately determine the root cause of the fault. This can be used to diagnose faults caused by the interaction of multiple factors in complex systems to avoid misdiagnosis. The application of counterfactual reasoning technology should be able to effectively reveal the causal relationship of faults; S5. Intelligent decision-making and collaborative operation steps: Group intelligent collaborative diagnosis steps, deploy multiple intelligent agents with autonomous decision-making and local information processing capabilities in the system, responsible for monitoring and preliminary diagnosis of the equipment or areas they are responsible for, and use collaborative mechanisms such as pheromone transmission and sharing of local optimal solutions to achieve global collaborative diagnosis. When the equipment is abnormal, the intelligent agents quickly gather together for joint analysis, and make comprehensive judgments to draw more accurate and comprehensive fault diagnosis conclusions, thereby improving diagnostic efficiency and accuracy. The process of group intelligent collaborative diagnosis should ensure smooth collaboration between intelligent agents and reliable diagnostic results; The emotional computing-assisted decision-making step uses high-definition cameras, microphones, and physiological signal monitoring equipment to collect multimodal data on operators’ facial expressions, voice intonation, and heart rate, analyze their emotional state and workload, consider the operator’s state during fault diagnosis and decision-making, and assist in determining the impact of human factors on operational errors or misjudgments; Adjust the alarm display or provide prompt information according to the stress state to help operators make accurate decisions. The process of emotional computing to assist decision-making should be able to accurately analyze the operator's state and provide effective auxiliary information; S6. Based on the human-computer interaction function of the system, efficient interaction between the operator and the system is achieved to assist the operator in his work.

9. The alarm method of the intelligent alarm system for power plant equipment monitoring according to claim 8 is characterized by: Steps S2-S4 are specifically as follows: S2.

1. Data fusion and deep analysis module steps: With the help of spatiotemporal network technology, the time series data of equipment operating parameters are combined with the spatial location and connection relationship information of the equipment in the power plant to construct a multi-dimensional spatiotemporal map. ,in is a node set, is the edge set, is the time dimension information; define the node connection strength function for: ； in, and Node and In time The key operating parameters of is the time window size; this function is used to construct a spatiotemporal graph to accurately reflect the spatiotemporal characteristics of the equipment operation data; Analyze the constructed spatiotemporal graph and define the factors affecting fault propagation for: ； in, For Node In time The parameter change, To analyze the time window; through research To observe the evolution of the equipment's operating status in the time and space dimensions, and then accurately predict the impact of the fault on adjacent equipment in space, as well as its propagation trend over time and potential impact range; The topological data analysis method is used to calculate the topological invariant TI of data points in high-dimensional space. The data point set is , define the topological similarity metric function for: ； in, For data points In the The value under the feature map, is the adjustment coefficient, is the number of feature maps; by calculating TI, the topological structure of the data is explored, and by identifying the hidden patterns in the topological structure, abnormal points and abnormal patterns are effectively found in high-dimensional nonlinear data sets, and the fault is located in time according to the change in the topological distribution of data points when the equipment fails. The application of topological data analysis methods should have high accuracy and reliability; S3.1, Intelligent learning and adaptive optimization component steps: Evolution strategy optimization step, introduce the evolution strategy ES algorithm to build an intelligent learning framework, set the model parameter vector , define the fitness function for: in, is the actual device operation status value, is the model prediction value, is the sample size, is the regularization coefficient; by simulating biological evolution mechanisms such as natural selection and crossover mutation, the optimization parameter combination is automatically screened based on actual operation data feedback, so that the model can adapt to the changes in power plant operating conditions and improve the adaptability and prediction accuracy of equipment status changes. The implementation process of the evolutionary strategy should ensure the effectiveness and stability of parameter optimization; The memory-enhanced neural network prediction step uses MANN and is equipped with an external memory module to store a large amount of historical operation data and fault cases; the current input data is , the historical pattern stored in the memory module is ( ), define the pattern matching function for: ； When predicting the future operating status of the device, the current input data is combined with similar historical patterns in the memory module to make a judgment; let the prediction value be , the calculation formula is: ； in, , Based on historical model The prediction value of the memory-enhanced neural network should be fully utilized to capture the potential fault trend and improve the prediction reliability and foresight. S4.1, Fault diagnosis and causal inference module steps: Based on historical data and expert experience, a Bayesian network fault diagnosis model is constructed to determine the probabilistic dependency between equipment operating parameters, environmental factors and various types of faults, and to assign conditional probabilities to the connections between network node variables; the equipment operating parameter vector is , the fault variable is , the environmental factor vector is , determine the conditional probability distribution based on a large amount of historical data and expert knowledge ; After accessing the real-time monitoring data, the Bayesian inference engine is used to calculate the posterior probability of different faults according to the probability relationship; assuming that the real-time monitoring data is and , according to Bayes' theorem, the fault The posterior probability The calculation formula is: ； in, The number of fault types; Combined with the changes in multiple parameters, the possibility of various faults when the equipment temperature rises is evaluated to provide a quantitative decision-making basis for fault diagnosis. The calculation process of the Bayesian reasoning engine should be accurate and efficient; The counterfactual reasoning technique is used to simulate the hypothetical scenario of equipment operation. The change of key equipment parameters is assumed to be , the sensor measurement deviation is , define the system state difference measurement function for: ； in, , and is the weight coefficient, It is the fault prediction value under normal circumstances; by comparing the differences in system states under actual and hypothetical scenarios, we can deeply understand the fault causal mechanism and accurately determine the root cause of the fault. It is used for fault diagnosis caused by the interaction of multiple factors in complex systems to avoid misdiagnosis. The application of counterfactual reasoning technology should be able to effectively reveal the fault causal relationship.

10. The alarm method of the intelligent alarm system for power plant equipment monitoring according to claim 8, characterized in that: Step S5 is specifically as follows: Swarm intelligence collaborative diagnosis steps: The system is deployed with Agents, each agent The data obtained from monitoring the equipment or area for which it is responsible is , the fault probability vector obtained after local preliminary diagnosis is ,in is the number of fault types; Defining Agents and Agent The pheromone concentration between , and its update formula is: ； in, is the pheromone volatility coefficient, ranging from arrive between, For the agent in this iteration and Agent The increment of pheromone between Based on the local diagnosis results, it is considered that When the diagnostic results are highly correlated, ,in Increase the strength factor for pheromones. and is the adjustment parameter; this formula is used to measure the degree of information correlation between agents and guide the collaboration between agents by updating the pheromone concentration; When the intelligent Based on the local diagnosis results, it is considered that The correlation of the diagnosis results is mainly calculated by calculating the local fault probability vector and the agent The similarity of the fault possibility vector is assumed to be ,when , is the set similarity threshold; Agent Update your own diagnosis results according to the pheromone concentration and the fault probability vector of other agents. The update formula is: ； in, , for the intelligent agent For Agents After multiple iterations, until the fault probability vectors of all agents converge, the results of all agents are finally combined to obtain an accurate and comprehensive fault diagnosis conclusion. During the whole process, the communication between agents is stable and the data sharing is timely, which improves the efficiency and accuracy of diagnosis. Emotional computing aided decision-making steps: Assume that the facial expression feature vector of the operator collected by high-definition camera, microphone and physiological signal monitoring equipment is , the speech intonation feature vector is , the heart rate characteristic value is HR; Define the emotional state evaluation function as , the calculation formula is: ； in, , and is the weight coefficient of different modal data, is the average heart rate of the operator during normal work. is the standard deviation of heart rate; this function is used to comprehensively evaluate the emotional state of the operator, and its value range is arrive The closer It means the mood is more tense or unstable; Assume that the operator's workload evaluation function is ,in is the amount of equipment data and task complexity index that the current operator needs to process, and the calculation formula is: ； in, For the Quantitative indicators of data or tasks, and is the adjustment index, is the total number of data or tasks; this function is used to evaluate the workload of operators, and its value range is arrive The closer It means the workload is greater; The decision support coefficient is defined as ,when When the system believes that the operator may be affected by emotions or workload, the decision risk increases. is the set threshold; at this time, the system adjusts the alarm display intensity according to the following formula : ； in, Displays the intensity of the original alarm. is the intensity adjustment coefficient; at the same time, the system provides prompt information according to the predefined rule base; When DAC is in a certain range, it prompts the operator to check specific equipment or review the operation steps to help the operator make accurate decisions. The process of emotional computing-assisted decision-making should be able to accurately analyze the operator's status and provide effective auxiliary information.

Images