Safety control application method and system based on big data acquisition and analysis

Abstract

The invention relates to the technical field of intelligent safety management and control of a thermal power plant, and discloses a safety management and control application method and system based on big data acquisition and analysis. The method comprises the following steps: constructing an equipment operation situation map by fusing real-time data streams, and realizing dynamic panoramic perception of a system operation state; key nodes are identified based on the atlas, a real-time safety margin index is calculated by using a safety evaluation model, a dynamic instruction set with priority is generated, and the inspection robot is driven to carry out self-adaptive accurate data acquisition; and the central processing unit updates the atlas by comparing and analyzing real-time and historical data, so as to realize risk re-assessment and early warning. According to the method, equipment monitoring is improved from isolated parameter analysis to global association cognition, a fixed inspection mode is converted into predictive intervention based on real-time risks, and the state perception depth, the early warning timeliness and the resource allocation efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent safety management and control technology for thermal power plants, specifically to a safety management and control application method and system based on big data collection and analysis. Background Technology

[0002] Thermal power plant production systems are complex, containing numerous high-parameter operating devices such as boilers, steam turbines, and generators, posing significant challenges to their safe and stable operation. Existing technologies generally rely on decentralized control systems to independently monitor specific parameters. Monitoring data is typically presented as independent numerical values ​​or simple trend charts, lacking effective correlation analysis between data points. Assessing equipment operating status largely depends on the experience of operators, making it difficult to grasp the interrelationships between devices and the overall operational status of the system from a holistic perspective.

[0003] In terms of equipment inspection and maintenance, the current common practice is to use periodic inspections or automated inspections along fixed routes. Inspection personnel conduct routine checks on equipment according to preset cycles and routes, or inspection robots collect data along fixed tracks. This model is inherently static and preventative, unable to dynamically adjust the focus and frequency of inspections based on the real-time health status of the equipment. When a piece of equipment exhibits abnormal symptoms, a static inspection plan struggles to respond promptly, potentially delaying the inspection of high-risk equipment. Furthermore, the even distribution of inspection resources also results in low resource utilization, hindering precise intervention in the early stages of an accident.

[0004] The existing technological system suffers from two shortcomings: its perception of equipment operating status is isolated and one-sided, and multi-source heterogeneous data has not been effectively integrated to form a deep understanding of the overall system security posture; its operation and maintenance strategies lack adaptability, and its proactive inspection and monitoring are insufficient, making it difficult to achieve precise resource scheduling based on real-time risk prediction. How to achieve dynamic and panoramic perception of the operating status of key equipment in thermal power plants, and based on this, establish an adaptive and precise proactive safety management mechanism, is a technical challenge that needs to be addressed in current industrial practice. Summary of the Invention

[0005] The purpose of this invention is to provide a security management and control application method and system based on big data collection and analysis to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a security management and control application method based on big data collection and analysis, the method comprising:

[0007] Extract equipment operation status sequences from the real-time data stream of the thermal power plant, perform multi-source data fusion on the equipment operation status sequences, and generate an equipment operation status map;

[0008] Based on the equipment operation status map, key equipment nodes are identified and a dynamic topology network is constructed. Temperature field distribution data and stress fluctuation data are collected by a sensor cluster deployed on the key equipment nodes.

[0009] The temperature field distribution data and stress fluctuation data are input into a pre-built safety assessment model, which outputs a real-time safety margin index for each key equipment node.

[0010] Based on the real-time safety margin index, a dynamic task instruction set for the inspection robot is generated, the dynamic task instruction set including the target device node coordinates and priority weights;

[0011] The inspection robot responds to the dynamic task instruction set, moves along the preset basic path, and performs multi-dimensional data collection after arriving at the target device node, and sends the collected real-time status data back to the central processing unit.

[0012] The central processing unit compares and analyzes the real-time status data with historical operating data, calculates the trend of equipment health decline, and updates the equipment operating status map.

[0013] Based on the updated equipment operation status map, the risk level of key equipment nodes is reassessed. If the risk level exceeds the preset threshold, an early warning signal is triggered and an emergency response plan is generated.

[0014] Preferably, the method further includes:

[0015] The emergency response plan was distributed to the on-site operation terminal, and the inspection route of the inspection robot was adjusted to concentrate on high-risk equipment node areas.

[0016] During emergency response, continuously monitor changes in equipment status, dynamically revise the emergency response plan, and synchronize the revised emergency response plan to all relevant terminals;

[0017] After completing the emergency response, summarize all the data from the entire incident, update the parameters of the safety assessment model, and generate a list of recommended equipment maintenance.

[0018] Preferably, extracting equipment operating status sequences from the real-time data stream of a thermal power plant, and performing multi-source data fusion on the equipment operating status sequences to generate an equipment operating status map specifically includes:

[0019] Receive real-time data streams from thermal power plants from distributed data acquisition devices, and perform timestamp alignment and outlier filtering on the real-time data streams from thermal power plants;

[0020] Extract parameter sequences related to equipment operation from the processed real-time data stream of the thermal power plant, including vibration frequency sequences, temperature change sequences, and pressure fluctuation sequences;

[0021] The parameter sequence is segmented using a sliding time window, and correlation analysis is performed on the multi-parameter data within each time window to establish a correlation matrix between parameters;

[0022] Based on the correlation matrix, devices with strong correlations are grouped into the same device group, and the core monitoring parameters for each device group are determined.

[0023] Based on the spatial distribution information of equipment groups and the correlation of operating parameters, a multi-layered equipment operation status map is constructed, in which different layers represent different types of operating parameters.

[0024] Preferably, based on the device operation status map, identifying key device nodes and constructing a dynamic topology network specifically includes:

[0025] Analyze the connectivity and centrality indices of each device node in the device operation status map, and screen out key device nodes with connectivity higher than the threshold.

[0026] Obtain the physical location information of key equipment nodes, establish connection relationships based on the process correlation between equipment, and form an initial topology network;

[0027] Based on the initial topology network, device operating status parameters are introduced as edge weights to construct a weighted dynamic topology network;

[0028] Real-time monitoring of the operational status changes of key equipment nodes; when an abnormal status is detected, dynamic adjustment of edge weights and node importance scores in the topology network.

[0029] Preferably, inputting the temperature field distribution data and stress fluctuation data into the pre-built safety assessment model specifically includes:

[0030] Collect temperature field distribution data and stress fluctuation data of key equipment nodes under normal operating conditions as training samples;

[0031] Deep learning networks are used to extract features from training samples and establish a mapping relationship between device status and security level.

[0032] The real-time collected temperature field distribution data and stress fluctuation data are input into the trained safety assessment model to obtain the safety status score of each key equipment node.

[0033] The real-time safety margin index is calculated based on the degree of deviation between the safety status score and the preset safety threshold.

[0034] Preferably, generating a dynamic task instruction set for the inspection robot based on the real-time safety margin index specifically includes:

[0035] Establish a mapping table between the locations of key equipment nodes and inspection path points;

[0036] Calculate the urgency coefficient of each critical device node based on the rate of change of the real-time safety margin index;

[0037] The optimal inspection sequence is generated by combining the spatial distribution density and urgency coefficient of key equipment nodes with a task scheduling algorithm;

[0038] Based on the optimal inspection sequence, a dynamic task instruction set is generated, which includes the target device node coordinates, inspection content, and priority weights.

[0039] Preferably, the central processing unit compares and analyzes the real-time status data with historical operating data to calculate the trend of equipment health decline, specifically including:

[0040] Extract status data sequences of similar devices under the same operating conditions from historical databases;

[0041] A predictive model for the change of equipment status parameters over time is established using time series analysis.

[0042] The deviation of the state parameters is calculated by comparing the real-time state data with the output of the prediction model.

[0043] Based on the trend of deviation, a regression analysis algorithm is used to predict the rate of decline in equipment health.

[0044] Preferably, based on the updated equipment operation status map, the risk level of key equipment nodes is reassessed, specifically including:

[0045] Extract the node status feature vectors from the updated equipment operation status map;

[0046] Input the node state feature vector into the risk classification model and output the risk probability of each critical device node;

[0047] Risk levels are classified according to the probability of risk, and corresponding risk thresholds are set.

[0048] When the risk level exceeds the risk threshold, the early warning signal generation process is initiated.

[0049] Triggering the early warning signal and generating the emergency response plan specifically includes:

[0050] The warning level is determined based on the risk level, and a warning signal of the corresponding level is generated;

[0051] Analyze the failure modes of high-risk equipment nodes and formulate targeted emergency response measures;

[0052] Optimize the execution sequence and timing of emergency response measures based on on-site environmental conditions and available resources;

[0053] The optimized emergency response measures will be packaged into a standardized emergency response plan.

[0054] Preferably, during emergency response, continuously monitoring changes in equipment status and dynamically revising the emergency response plan specifically includes:

[0055] During emergency response, real-time collection of equipment status parameters and environmental impact data is conducted.

[0056] Establish emergency response effectiveness evaluation indicators to monitor the effectiveness of emergency response measures;

[0057] When the response deviates from expectations, the reasons should be analyzed and the response measures adjusted.

[0058] Based on the adjusted response measures, update the operational procedures and parameter settings in the emergency response plan;

[0059] After completing the emergency response, summarize all the data from the entire incident and update the parameters of the safety assessment model, specifically including:

[0060] Collect all monitoring data and operation records from the triggering of the early warning to the end of the emergency response;

[0061] The collected data is cleaned and labeled to build a model training sample set;

[0062] An incremental learning algorithm is used to update the weight parameters of the security assessment model using a new training sample set;

[0063] Based on the updated safety assessment model, the operational risk characteristics of the equipment are recalculated, and the equipment maintenance strategy is optimized.

[0064] Preferably, the present invention also includes a security management and control application system based on big data collection and analysis. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the security management and control application method based on big data collection and analysis as described above.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] By extracting device operating status sequences from real-time data streams and fusing multi-source data, a dynamic device operating status map is generated, integrating previously isolated device parameters into a relational network model reflecting the overall system operating status. This map intuitively reveals the interdependencies and influence paths between devices, overcoming the limitations of isolated parameter displays in traditional monitoring interfaces. Identifying key nodes and constructing a dynamic topology network based on the map allows data acquisition activities to focus on devices that play a pivotal role in system security, enhancing the depth and relevance of status perception. This map-based status generation method transforms the monitoring of complex system operating status from a local, static perspective to a global, dynamic one, enhancing the predictability of potential risk transmission paths.

[0067] By using the real-time safety margin indicators output by the safety assessment model as the direct basis for generating dynamic task instruction sets for inspection robots, and assigning priority weights to these instruction sets, the allocation of inspection resources shifts from a fixed mode to a response mode based on real-time risks. Inspection robots no longer mechanically execute predetermined routes, but instead prioritize handling critical equipment nodes with lower safety margins based on the risk distribution assessed by the system in real time. This mechanism upgrades inspection activities from preventative periodic checks to predictive and precise interventions, enabling the automatic focus of maintenance resources on the weakest links. The inspection mode driven by dynamic task instructions effectively shortens the anomaly response time of high-risk equipment, improves the efficiency and targeting of inspection work without changing the total amount of resources, and constructs a proactive defense system that allocates resources on demand. Attached Figure Description

[0068] Figure 1 This is a schematic diagram illustrating the working principle of the security management and control application method based on big data collection and analysis described in this invention.

[0069] Figure 2 Flowchart for emergency response and contingency plan updates;

[0070] Figure 3 A flowchart for identifying key equipment nodes and constructing a dynamic topology network;

[0071] Figure 4 A graph showing the decline in the health of core equipment over operating time;

[0072] Figure 5 This is a statistical chart showing the risk level distribution of critical equipment. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Please see Figure 1 This invention provides a safety management and control application method and system based on big data acquisition and analysis. The method includes: extracting equipment operating status sequences from real-time data streams of thermal power plants; fusing multi-source data to generate an equipment operating status map; identifying key equipment nodes and constructing a dynamic topology network based on the equipment operating status map; collecting temperature field distribution data and stress fluctuation data through sensor clusters deployed on the key equipment nodes; inputting the temperature field distribution data and stress fluctuation data into a pre-constructed safety assessment model, which outputs a real-time safety margin index for each key equipment node; and generating a safety assessment system based on the real-time safety margin index. The inspection robot has a dynamic task instruction set, which includes the coordinates and priority weights of the target equipment node. Responding to the dynamic task instruction set, the inspection robot moves along a preset basic path and performs multi-dimensional data collection upon reaching the target equipment node, transmitting the collected real-time status data back to the central processing unit. The central processing unit compares and analyzes the real-time status data with historical operating data, calculates the equipment health degradation trend, and updates the equipment operation status map. Based on the updated equipment operation status map, the risk level of key equipment nodes is reassessed. If the risk level exceeds a preset threshold, an early warning signal is triggered and an emergency response plan is generated.

[0075] Example 1: In specific implementation, real-time data streams from thermal power plants are received from distributed data acquisition devices. The data streams undergo timestamp alignment and outlier filtering. Timestamp alignment sorts data points from different sources according to a unified time base. Outlier filtering uses statistical methods to identify and remove data points that deviate from the normal range, thus ensuring the integrity and consistency of the real-time data streams. From the processed real-time data streams, parameter sequences related to equipment operation are extracted. These parameter sequences include vibration frequency sequences, temperature change sequences, and pressure fluctuation sequences. The vibration frequency sequence captures the vibration characteristics of rotating machinery, the temperature change sequence reflects the evolution of the equipment's thermal state, and the pressure fluctuation sequence indicates the operating conditions of the fluid system. These parameter sequences serve as the basic input for multi-source data fusion.

[0076] A sliding time window is used to segment the parameter sequence. The sliding time window moves forward at fixed time intervals, and correlation analysis is performed on the multi-parameter data within each time window to establish a correlation matrix between the parameters. The correlation matrix can be calculated using the following formula:

[0077]

[0078] in: Indicates parameters and parameters The correlation coefficient within the time window, and They represent parameters respectively and parameters At the point of time The value, and It is a parameter and parameters The average value within the time window. This refers to the size of the time window. This formula quantifies the degree of linear correlation between parameters; a correlation coefficient close to 1 or -1 indicates a strong correlation, while a coefficient close to 0 indicates a weak correlation. Based on the correlation matrix, devices with strong correlations are grouped into the same device group, and the core monitoring parameters for each group are determined. These core monitoring parameters are those that exhibit high correlation within the group and can represent the overall operating status of the device group.

[0079] In some embodiments, a multi-layered equipment operation status map is constructed based on the spatial distribution information of equipment groups and the correlation of operating parameters. The spatial distribution information originates from the physical layout data of the equipment within the thermal power plant, and the correlation of operating parameters is derived from a correlation matrix. Different layers of the equipment operation status map represent different types of operating parameters, such as vibration parameter layers, temperature parameter layers, and pressure parameter layers. Each layer graphically displays the correlation between parameters within the equipment group. It can be understood that the multi-layered structure allows for independent analysis of the changing trends of different parameter types, while achieving a comprehensive status assessment through layer overlay. Optionally, the construction of the equipment operation status map can employ graph database technology for storage and management, where nodes represent equipment or parameters, and edges represent correlations, thereby supporting efficient query and update operations.

[0080] In practice, the specific parameters for timestamp alignment and outlier filtering can be adjusted according to the actual operating environment of the thermal power plant, such as the time window size. The sampling frequency can be set according to the data sampling frequency, with common values ​​being 60 seconds or 300 seconds to balance real-time performance and analytical accuracy. Outlier filtering can employ the Z-score method, setting a threshold such as 3 times the standard deviation to automatically identify and remove outlier data points. The segmented approach of the sliding time window ensures the timeliness of data analysis and avoids delays caused by historical data backlog. The establishment of the correlation matrix relies on numerical computing libraries, such as Python's Pandas or NumPy libraries for efficient matrix operations. In some embodiments, device grouping employs clustering algorithms, such as hierarchical clustering based on correlation coefficients, grouping devices with correlation coefficients higher than a preset threshold, such as 0.8, into the same group. The determination of core monitoring parameters is achieved by evaluating the centrality of parameters within the group, for example, selecting the parameter with the highest average correlation coefficient with other parameters as the core monitoring parameter.

[0081] In practice, the multi-layered structure of the equipment operation status map is achieved through visualization tools, such as using D3.js or Tableau to create interactive maps, facilitating operators' monitoring of equipment operating status. Operating parameters in different layers can be updated independently; for example, the temperature parameter layer refreshes every few seconds, while the vibration parameter layer updates more frequently to adapt to the dynamic characteristics of parameter changes. Optionally, a weighting mechanism can be introduced during map construction to allocate layer display priority based on parameter importance, thereby highlighting key monitoring areas.

[0082] Example 2: See Figure 3 In practical implementation, the connectivity and centrality indices of each equipment node in the equipment operation status map are analyzed to screen out key equipment nodes with connectivity exceeding a threshold. The connectivity index measures the number of edges directly connected to a device node in the topology, while the centrality index assesses the importance of a node as an information or energy flow hub in the network. A connectivity threshold is set to filter out key equipment nodes that have a significant impact on the overall operational stability of the network. The physical location information of key equipment nodes is obtained from the plant's equipment layout drawings or geographic information system coordinates. Connections are established based on the process correlations between equipment. Process correlations refer to the upstream and downstream relationships or functional coupling relationships of equipment in the thermal power plant's production process, forming an initial topology network. In the initial topology network, nodes represent equipment and edges represent process connections. Based on the initial topology network, equipment operating status parameters are introduced as edge weights to construct a weighted dynamic topology network. Equipment operating status parameters include equipment load rate, efficiency, or failure frequency, and edge weights are numerically represented to indicate connection strength or degree of influence. Real-time monitoring of the operational status changes of key equipment nodes; when an abnormal status is detected, dynamic adjustment of edge weights and node importance scores in the topology network. The node importance score is obtained by calculating the eigenvector centrality or betweenness centrality of the node in the network. An abnormal status refers to an event in which the operating parameters deviate from the normal range.

[0083] Temperature field distribution data and stress fluctuation data of key equipment nodes under normal operating conditions are collected as training samples. Normal operating conditions are defined by data from historical operation records showing no fault alarms. Temperature field distribution data is acquired using an infrared thermal imager array or distributed temperature sensors, while stress fluctuation data is collected using strain gauges or vibration acceleration sensors. A deep learning network (e.g., convolutional neural network or long short-term memory network) is used to extract features from the training samples, establishing a mapping relationship between equipment status and safety level. The safety level is labeled based on historical maintenance records and expert knowledge. The real-time collected temperature field distribution data and stress fluctuation data are input into the trained safety assessment model to obtain a safety status score for each key equipment node. The safety status score is a value between 0 and 1, with higher values ​​indicating better safety. Based on the deviation of the safety status score from a preset safety threshold, a real-time safety margin index is calculated. This index reflects the remaining buffer space between the current equipment state and the critical safety state.

[0084] In some embodiments, the edge weights of the weighted dynamic topology network are adjusted according to specific rules, such as proportionally reducing the weight value of the corresponding edge when any device in the connected network experiences a sudden load drop. The node importance score is dynamically updated using the following formula:

[0085]

[0086] in: This represents the node importance score of node v. It is the degree centrality of node v. It is the betweenness centrality of node v. It is the average weight of all edges connected to node v. It is the maximum value of all edge weights in the network. , , These are weighting coefficients used to balance the influence of different indicators.

[0087] In practical implementation, the safety assessment model is trained using supervised learning. The training samples must include a sufficient number of data points under normal conditions and various known fault states. The model minimizes the loss function between the predicted safety level and the true label using a backpropagation algorithm. It is understood that after training, the model needs to be validated on an independent test set to avoid overfitting. The real-time safety margin index can be calculated using the relative difference method: (safety status score - preset safety threshold) / preset safety threshold. A negative result indicates that the safety level is below the safety threshold. Optionally, the preset safety threshold can be differentiated based on equipment type, operating years, and importance; for example, a higher safety threshold can be set for core main equipment.

[0088] In some embodiments, the feature extraction layer of the deep learning network can be designed as a multi-branch structure to process the spatial features of temperature field distribution data and the time-series features of stress fluctuation data, respectively. Finally, feature fusion and safety level classification are performed through a fully connected layer. Real-time acquired data needs to undergo standardized preprocessing consistent with the training samples before being input into the safety assessment model to eliminate the influence of dimensions. Optionally, the output of the safety status score can be subdivided into multiple dimensions, such as structural integrity score and thermal stability score, and then a comprehensive safety status score is obtained through weighted summation.

[0089] Example 3: In specific implementation, a mapping table between the locations of critical equipment nodes and inspection path points is established. The locations of critical equipment nodes are determined in three-dimensional coordinates using the factory coordinate system. Inspection path points are a set of geographic coordinate points reachable by the robot, pre-set on the factory's electronic map. The mapping table is stored in a relational database table, recording the identifier of each critical equipment node and its corresponding one or more nearest inspection path point numbers, as well as the Euclidean distance between them. Based on the rate of change of the real-time safety margin index, the urgency coefficient of each critical equipment node is calculated. The rate of change of the real-time safety margin index is obtained by calculating the numerical difference of the safety margin index within a unit time window. The calculation of the urgency coefficient requires normalization using a benchmark reference value. Combining the spatial distribution density of critical equipment nodes and the urgency coefficient, a task scheduling algorithm is used to generate the optimal inspection sequence. The spatial distribution density is obtained by calculating the number of nodes within a circular area centered on the critical equipment node and with a given radius. Based on the optimal inspection sequence, a dynamic task instruction set is generated, which includes the target device node coordinates, inspection content, and priority weights. The target device node coordinates are directly obtained from the spatial location database associated with the mapping relationship table. The inspection content specifies the data types and collection accuracy that the robot needs to collect. The priority weights are obtained through a linear transformation of the urgency coefficient.

[0090] State data sequences of similar equipment under identical operating conditions are extracted from historical databases. Similar equipment refers to equipment with the same model, specifications, manufacturing batch, and design parameters. Identical operating conditions are defined by load rate range, ambient temperature and humidity range, and maintenance history. State data sequences include time series such as vibration spectrum data, temperature gradient data, and pressure pulsation data. A predictive model for the changes of equipment state parameters over time is established using time series analysis methods. The time series analysis method employs a seasonal autoregressive integral moving average (SRM) model, which can output predicted values ​​of state parameters and their confidence intervals for specific future time periods. In practice, the application of the seasonal autoregressive integral moving average model begins with extracting state data sequences of similar equipment under identical operating conditions from historical databases. These sequences include vibration frequency sequences, temperature change sequences, and pressure fluctuation sequences. The model performs a stationarity test on the state data sequences, eliminating trend and seasonal components through differencing to ensure time series stationarity. Subsequently, the autoregressive order, differencing order, moving average order, and seasonal periodic parameters are identified, determined based on the analysis of data autocorrelation plots and partial autocorrelation plots. Model validation ensures goodness of fit by checking whether the residual sequence is white noise. Finally, the model outputs predicted values ​​of state parameters and their confidence intervals for a specific future time period. The confidence intervals are calculated based on the residual variance and the assumption of a normal distribution, providing a quantitative basis for analyzing the decline trend of equipment health. The deviation of state parameters is calculated by comparing real-time state data with the prediction model output; the deviation is the absolute value of the standardized residual between the real-time measured value and the model prediction value. Based on the trend of the deviation, a regression analysis algorithm is used to predict the rate of decline of equipment health. The regression analysis algorithm employs the weighted least squares method, and the rate of decline of equipment health is obtained by fitting the slope of the deviation data within a continuous time window.

[0091] In some embodiments, the urgency coefficient is calculated using the following improved formula:

[0092]

[0093] in: This represents the urgency coefficient of the j-th critical equipment node. This represents the change in the real-time safety margin indicator of the j-th critical device node within a specific time interval. It is a reference benchmark value for the change in real-time safety margin indicators. This represents the spatial distribution density of equipment in the region where the j-th critical equipment node is located. It is a reference value for spatial distribution density. and It is a dimensionless weighted coefficient used to balance the weights of two influencing factors.

[0094] In practical implementation, the task scheduling algorithm can employ an improved genetic algorithm. This algorithm iteratively optimizes the population by minimizing the total inspection cost using a fitness function. The total inspection cost is a weighted sum of path length, overall urgency, and task delay time. The output of the optimal inspection sequence is an ordered sequence of key device node identifiers, guiding the robot to sequentially access each target device node. The dynamic task instruction set is transmitted to the inspection robot's onboard control unit via an industrial wireless LAN, encoded and transmitted using Protocol Buffers serialization format. Optionally, the linear transformation of priority weights uses a minimum-maximum scaling method to map the urgency coefficient to a preset priority value range.

[0095] In some embodiments, the establishment of a device condition parameter prediction model requires the inclusion of historical data from at least one complete annual operating cycle to cover the device's operating characteristics in different seasons. After model training, an Ljung-Box test is performed to verify the white noise characteristics of the residuals. The deviation of the condition parameters is calculated using a studentized residual method, i.e., the residual is divided by an estimate of its standard error. It can be understood that the prediction of the device health degradation rate requires a deviation time series within a sliding time window, fitted with the slope of its trend line using linear regression.

[0096] See Figure 4 This line chart focuses on three core equipment categories in thermal power plants, showcasing their health changes over a 12-month operating cycle. Based on time series analysis and regression algorithms, the chart visualizes the rate of equipment health degradation, intuitively reflecting the dynamic deterioration patterns of core equipment health. It helps thermal power plants accurately identify equipment at high risk of degradation, providing data for adjusting maintenance cycles and optimizing inspection resource allocation. This facilitates early intervention in preventative maintenance of high-risk equipment, reducing the probability of downtime due to failures.

[0097] Example 4: In specific implementation, node status feature vectors are extracted from the updated equipment operation status map. Each node status feature vector is a multi-dimensional numerical array, whose elements include the original monitoring parameters directly extracted from the equipment operation status map and calculated statistical features. The statistical features include the mean, variance, and peak factor of the monitoring parameters within the sliding time window. The node status feature vectors are input into a risk classification model, which outputs the risk probability of each critical equipment node. The risk classification model uses a gradient boosting decision tree-based classifier, trained using historical equipment status data and corresponding fault records. Risk levels are divided according to the magnitude of the risk probability, and corresponding risk thresholds are set. Risk levels are classified into low, medium, and high, with risk thresholds determined based on industry safety standards and equipment criticality. When the risk level exceeds the risk threshold, a warning signal generation process is initiated, including signal encoding, priority determination, and transmission channel selection.

[0098] Early warning levels are determined based on risk levels, and corresponding early warning signals are generated, with a one-to-one correspondence between warning levels and risk levels. Failure modes of high-risk equipment nodes are analyzed, and targeted emergency response measures are formulated. Failure mode analysis is based on fault tree analysis and a historical failure case database. The execution sequence and timing of emergency response measures are optimized based on site environmental conditions and available resources, including equipment accessibility and environmental hazards. The optimized emergency response measures are then packaged into standardized emergency response plans, which use a unified scripting language to describe operational steps, judgment conditions, and expected results.

[0099] Table 1: Correspondence between Risk Level Classification and Early Warning Signals

[0100]

[0101] In some embodiments, the construction of node state feature vectors requires standardization of the original monitoring parameters to eliminate the influence of different parameter dimensions. The standardization process employs the Z-score method. The risk probability is calculated using the following logistic function formula:

[0102]

[0103] in: This represents the risk probability of critical equipment node v. This indicates that the risk score calculated for node v in the risk classification model is the output value after a nonlinear transformation of a linear combination of the node's state feature vectors.

[0104] In practice, the early warning signal generation process includes a signal verification step to prevent false alarms due to data transmission errors. Failure mode analysis requires a comprehensive judgment based on real-time equipment operating parameters and design margins. It can be understood that optimizing emergency response measures is a multi-objective programming problem, requiring simultaneous consideration of response effectiveness, resource consumption, and time costs. Optionally, standardized emergency response plans can be described using an Extensible Markup Language (EXPLAIN) format to enhance machine readability and interoperability.

[0105] In some embodiments, the risk threshold setting can incorporate a dynamic adjustment mechanism, periodically correcting it based on the equipment's operating age and load conditions. The representation of warning levels must be compatible with the factory's existing alarm system to ensure that alarm information can be correctly integrated and displayed. It is understood that the encapsulation of emergency response plans includes metadata information such as the plan version number, effective date, and issuer.

[0106] See Figure 5 This chart, a bar graph, visually presents the distribution of key equipment across low, medium, and high risk levels. It also visualizes the implementation of the technical solution: by extracting the node feature vectors from the equipment operation status map, inputting them into a risk classification model to obtain risk probabilities and classify levels, this chart is the output of that process. Its value lies in clearly showing the current risk structure of the equipment. The overall predominance of low-risk equipment indicates a relatively stable system operation, but high-risk equipment requires focused management. This high-risk equipment will trigger early warning signals and generate emergency plans, while simultaneously guiding inspection resources to converge on this area, providing an intuitive basis for proactive safety management decisions.

[0107] Example 5: See Figure 2In practice, emergency response plans are distributed to on-site operation terminals, and the inspection routes of patrol robots are adjusted to focus on high-risk equipment nodes. The emergency response plans are transmitted via the factory's internal network in the form of data packets. On-site operation terminals include fixed industrial control computers and mobile handheld devices. Adjustments to the inspection routes are achieved by updating the path point sequence in the robot navigation system. High-risk equipment nodes refer to areas where equipment with a risk probability exceeding 0.5 is concentrated. During emergency response, equipment status parameters and environmental impact data are collected in real time. Equipment status parameters include temperature, pressure, and vibration, while environmental impact data covers ambient temperature, humidity, and dust concentration. Emergency response effectiveness evaluation indicators are established to monitor the effectiveness of emergency response measures. These indicators include quantitative parameters such as parameter recovery rate, stabilization time, and deviation integral. When the response effect deviates from expectations, the causes are analyzed and the response measures are adjusted. Cause analysis involves comparing the expected results with the actual monitoring data, and adjustments may include modifying parameter settings or changing the response method. Based on the adjusted response measures, the operational procedures and parameter settings in the emergency response plan will be updated. The update process will be managed through a version control mechanism to ensure the timeliness and consistency of the plan.

[0108] After emergency response is completed, all monitoring data and operation records from the triggering of the warning to the end of the emergency response are collected. The monitoring data includes raw sensor data and calculated derived indicators, while the operation records detail all manual and automated actions performed. The collected data is cleaned and labeled to construct a model training sample set. Data cleaning includes removing outliers and filling in missing values. Data labeling is based on the response results, categorizing data as successful or requiring optimization. An incremental learning algorithm is used to update the weight parameters of the safety assessment model using the new training sample set. This algorithm integrates new knowledge without retraining the entire model. Based on the updated safety assessment model, the equipment operation risk characteristics are recalculated, and equipment maintenance strategies are optimized. These optimizations include adjusting maintenance cycles, replacing spare parts, or modifying inspection standards. A list of recommended equipment maintenance is generated, listing the equipment numbers requiring maintenance, maintenance types, and recommended execution times.

[0109] In some embodiments, the emergency response effectiveness evaluation index is calculated using the following formula:

[0110]

[0111] in: This represents the overall score of the emergency response effectiveness evaluation indicators. This indicates the percentage of key parameters that have returned to their normal range. This represents the preset target stabilization time. This indicates the actual time required to reach stability. This represents the integral value of the parameter deviation. Indicates the deviation integration threshold. , , These are the weighting coefficients for each sub-indicator.

[0112] In practice, the on-site operation terminal receives the emergency response plan, parses it, displays a flowchart of the operation steps, and simultaneously issues audio-visual prompts to notify the operators. The inspection robot employs a dynamic obstacle avoidance algorithm as it moves towards high-risk areas to prevent collisions with personnel or other mobile devices. It can be understood that monitoring the emergency response effectiveness evaluation indicators is a continuous process, with evaluation results displayed in real-time on the monitoring screen. Optionally, suggestions for adjusting response measures can be generated through an expert system, which matches and recommends based on a historical case database. The incremental learning algorithm can be an online sequential extreme learning machine or an incremental support vector machine; model updates are automatically performed during periods of low system load.

[0113] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A security management and control application method based on big data collection and analysis, characterized in that, The method includes: Extract equipment operation status sequences from the real-time data stream of the thermal power plant, perform multi-source data fusion on the equipment operation status sequences, and generate an equipment operation status map; Based on the equipment operation status map, key equipment nodes are identified and a dynamic topology network is constructed. Temperature field distribution data and stress fluctuation data are collected by a sensor cluster deployed on the key equipment nodes. The temperature field distribution data and stress fluctuation data are input into a pre-built safety assessment model, which outputs a real-time safety margin index for each key equipment node. Based on the real-time safety margin index, a dynamic task instruction set for the inspection robot is generated, the dynamic task instruction set including the target device node coordinates and priority weights; The inspection robot responds to the dynamic task instruction set, moves along the preset basic path, and performs multi-dimensional data collection after arriving at the target device node, and sends the collected real-time status data back to the central processing unit. The central processing unit compares and analyzes the real-time status data with historical operating data, calculates the trend of equipment health decline, and updates the equipment operating status map. Based on the updated equipment operation status map, the risk level of key equipment nodes is reassessed. If the risk level exceeds the preset threshold, an early warning signal is triggered and an emergency response plan is generated.

2. The security management and control application method based on big data collection and analysis as described in claim 1, characterized in that, The method further includes: The emergency response plan was distributed to the on-site operation terminal, and the inspection route of the inspection robot was adjusted to concentrate on high-risk equipment node areas. During emergency response, continuously monitor changes in equipment status, dynamically revise the emergency response plan, and synchronize the revised emergency response plan to all relevant terminals; After completing the emergency response, summarize all the data from the entire incident, update the parameters of the safety assessment model, and generate a list of recommended equipment maintenance.

3. The security management and control application method based on big data collection and analysis as described in claim 1, characterized in that, Extracting equipment operating status sequences from the real-time data stream of a thermal power plant, and performing multi-source data fusion on these sequences to generate an equipment operating status map, specifically includes: Receive real-time data streams from thermal power plants from distributed data acquisition devices, and perform timestamp alignment and outlier filtering on the real-time data streams from thermal power plants; Extract parameter sequences related to equipment operation from the processed real-time data stream of the thermal power plant, including vibration frequency sequences, temperature change sequences, and pressure fluctuation sequences; The parameter sequence is segmented using a sliding time window, and correlation analysis is performed on the multi-parameter data within each time window to establish a correlation matrix between parameters; Based on the correlation matrix, devices with strong correlations are grouped into the same device group, and the core monitoring parameters for each device group are determined. Based on the spatial distribution information of equipment groups and the correlation of operating parameters, a multi-layered equipment operation status map is constructed, in which different layers represent different types of operating parameters.

4. The security management and control application method based on big data collection and analysis as described in claim 3, characterized in that, Based on the aforementioned equipment operation status map, identifying key equipment nodes and constructing a dynamic topology network specifically includes: Analyze the connectivity and centrality indices of each device node in the device operation status map, and screen out key device nodes with connectivity higher than the threshold. Obtain the physical location information of key equipment nodes, establish connection relationships based on the process correlation between equipment, and form an initial topology network; Based on the initial topology network, device operating status parameters are introduced as edge weights to construct a weighted dynamic topology network; Real-time monitoring of the operational status changes of key equipment nodes; when an abnormal status is detected, dynamic adjustment of edge weights and node importance scores in the topology network.

5. The security management and control application method based on big data collection and analysis as described in claim 1, characterized in that, The input of the temperature field distribution data and stress fluctuation data into the pre-built safety assessment model specifically includes: Collect temperature field distribution data and stress fluctuation data of key equipment nodes under normal operating conditions as training samples; Deep learning networks are used to extract features from training samples and establish a mapping relationship between device status and security level. The real-time collected temperature field distribution data and stress fluctuation data are input into the trained safety assessment model to obtain the safety status score of each key equipment node. The real-time safety margin index is calculated based on the degree of deviation between the safety status score and the preset safety threshold.

6. The security management and control application method based on big data collection and analysis as described in claim 1, characterized in that, Based on the aforementioned real-time safety margin index, the generation of a dynamic task instruction set for the inspection robot specifically includes: Establish a mapping table between the locations of key equipment nodes and inspection path points; Calculate the urgency coefficient of each critical device node based on the rate of change of the real-time safety margin index; The optimal inspection sequence is generated by combining the spatial distribution density and urgency coefficient of key equipment nodes with a task scheduling algorithm; Based on the optimal inspection sequence, a dynamic task instruction set is generated, which includes the target device node coordinates, inspection content, and priority weights.

7. The security management and control application method based on big data collection and analysis as described in claim 1, characterized in that, The central processing unit compares and analyzes the real-time status data with historical operating data to calculate the trend of equipment health decline, specifically including: Extract status data sequences of similar devices under the same operating conditions from historical databases; A predictive model for the change of equipment status parameters over time is established using time series analysis. The deviation of the state parameters is calculated by comparing the real-time state data with the output of the prediction model. Based on the trend of deviation, a regression analysis algorithm is used to predict the rate of decline in equipment health.

8. The security management and control application method based on big data collection and analysis as described in claim 1, characterized in that, Based on the updated equipment operational status map, the risk level of key equipment nodes is reassessed, specifically including: Extract the node status feature vectors from the updated equipment operation status map; Input the node state feature vector into the risk classification model and output the risk probability of each critical device node; Risk levels are classified according to the probability of risk, and corresponding risk thresholds are set. When the risk level exceeds the risk threshold, the early warning signal generation process is initiated. Triggering the early warning signal and generating the emergency response plan specifically includes: The warning level is determined based on the risk level, and a warning signal of the corresponding level is generated; Analyze the failure modes of high-risk equipment nodes and formulate targeted emergency response measures; Optimize the execution sequence and timing of emergency response measures based on on-site environmental conditions and available resources; The optimized emergency response measures will be packaged into a standardized emergency response plan.

9. The security management and control application method based on big data collection and analysis as described in claim 2, characterized in that, During emergency response, continuous monitoring of equipment status changes and dynamic modification of the emergency response plan specifically include: During emergency response, real-time collection of equipment status parameters and environmental impact data is conducted. Establish emergency response effectiveness evaluation indicators to monitor the effectiveness of emergency response measures; When the response deviates from expectations, the reasons should be analyzed and the response measures adjusted. Based on the adjusted response measures, update the operational procedures and parameter settings in the emergency response plan; After completing the emergency response, summarize all the data from the entire incident and update the parameters of the safety assessment model, specifically including: Collect all monitoring data and operation records from the triggering of the early warning to the end of the emergency response; The collected data is cleaned and labeled to build a model training sample set; An incremental learning algorithm is used to update the weight parameters of the security assessment model using a new training sample set; Based on the updated safety assessment model, the operational risk characteristics of the equipment are recalculated, and the equipment maintenance strategy is optimized.

10. A security management and control application system based on big data collection and analysis, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the security management and control application method based on big data collection and analysis as described in any one of claims 1 to 9.

Images