Thermal power plant DCS production early warning method and system based on artificial intelligence

Abstract

The invention discloses a thermal power plant DCS production early warning method and system based on artificial intelligence, and relates to the technical field of thermal power plant intelligent operation and industrial process intelligent early warning. According to the method, deep fusion modeling is carried out on a historical DCS alarm time sequence event chain and a measuring point process topology network, a correlation mapping library is constructed, structural expression of a complex alarm evolution rule is achieved, early warning judgment is not limited to single-point threshold judgment any more, matching recognition is carried out based on a time sequence evolution mode under a real working condition, and the accuracy of early warning judgment is improved. The advancement and the accuracy of potential fault identification are obviously improved; through a time interval tolerance interval, a reliability index and a conditional probability mechanism, the false alarm rate is effectively reduced, and the credibility of an early warning result is enhanced; meanwhile, based on reverse tracing of the topological network and a historical case driven root cause probability calculation mechanism, a root cause node candidate list with priority ranking can be quickly generated, and the manual troubleshooting range and the diagnosis time are greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and intelligent early warning technology for thermal power plants and industrial processes, and in particular to an artificial intelligence-based DCS production early warning method and system for thermal power plants. Background Technology

[0002] With the deep integration of new-generation information technology and the energy industry, the production and operation of thermal power plants are gradually shifting from traditional experience-driven management to data-driven and intelligent decision-making management. Distributed Control Systems (DCS), as the core platform for process control in thermal power plants, undertake the tasks of real-time monitoring, control, and protection of boilers, turbines, generators, and their auxiliary systems. Their stability and reliability directly affect the safety, economy, and continuous operation capability of the unit. In recent years, with the development of sensing technology, the Industrial Internet, big data, and artificial intelligence, conducting production early warning, fault prediction, and intelligent operation and maintenance based on historical operating data and online monitoring data has become an important direction for the intelligent upgrading of thermal power plants.

[0003] In existing technologies, it is common to centrally collect DCS alarm information, operating parameters, and operation logs, and introduce expert rules, statistical analysis, or machine learning models to identify abnormal states and provide simple early warnings, which improves the timeliness of anomaly detection to some extent. However, due to the complex structure of power plant systems, the large number of measuring points, and the dense and complex coupling relationships of alarm information, how to accurately identify potential fault evolution links from massive, multi-source, and strongly correlated alarm and operation data, predict risk propagation paths in advance, and provide operators with targeted diagnostic and handling suggestions remains a core research issue facing current intelligent early warning technology for thermal power plants. Summary of the Invention

[0004] In view of the problems existing in the DCS production early warning technology of thermal power plants, this invention is proposed.

[0005] Therefore, the problem to be solved by this invention is how to effectively solve the problem of difficulty in timely detection and accurate location of potential faults under complex working conditions by constructing an association mapping between historical time-series event chains and process topology networks, introducing a reverse root cause probability tracing mechanism and generating fused early warning information.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an artificial intelligence-based DCS production early warning method for thermal power plants, comprising: S1: extracting the time-series event chain of historical alarm records and mapping it with the DCS measurement point process topology network to form an association mapping library; S2: monitoring DCS alarm data online and matching it with the association mapping library; when a starting pattern of the event chain is matched, an early warning is immediately triggered and the end node of the event chain is located as a potential fault point; S3: starting from the potential fault point, tracing back along the DCS measurement point process topology network, calculating the probability of each upstream node becoming the root cause based on historical cases, and generating a candidate list of root cause nodes; S4: generating fused early warning information containing phenomena, predictions, and suggested investigation objects based on the starting pattern, potential fault points, and candidate list of root cause nodes, and pushing it out.

[0007] As a preferred embodiment of the artificial intelligence-based DCS production early warning method for thermal power plants described in this invention, the generation of the time-series event chain includes: cleaning historical DCS alarm records to eliminate duplicate and transient jitter alarms; standardizing each alarm into a standardized alarm event containing a measurement point identifier, alarm type, and timestamp; defining two consecutive standardized alarm events within a preset first time window as a time-series event pair; and connecting multiple consecutive time-series event pairs in chronological order to form an alarm event chain sequence. The generation of the DCS measurement point process topology network includes: selecting multiple time periods characterizing stable unit operation from historical DCS data, extracting continuous time-series data of all DCS measurement points within the time periods to form a normal operating condition dataset; and, based on the normal operating condition dataset, performing the following steps to generate the process topology network: for any two different DCS measurement points... and symmetrically calculate multiple preset time offsets The mutual information values ​​are calculated, and the mutual information sequences are obtained. The maximum value of each sequence is defined as the measurement point pair. correlation strength value and measuring point pair correlation strength value And the time offset corresponding to reaching the maximum value is denoted as . arrive Optimal latency ;and arrive Optimal latency ; associate strength value All measuring points exceeding the intensity threshold Retain and form a candidate causal edge set; for each pair of measurement points in the candidate causal edge set The following rules apply: Rule a: If and If the delay is less than the upper limit of reasonable propagation time, then the causal direction is preliminarily determined to be... ;like and If the delay is less than the upper limit of reasonable propagation time, it is judged as... ;like and If neither value is greater than 0, or if both absolute values ​​exceed the reasonable propagation delay limit, the delay information is considered invalid, and proceed to rule b; Rule b: Query the pre-set process knowledge base. If the knowledge base contains... lie in The upstream process links are then used to determine the direction. If recorded lie in The upstream of, then determined as Otherwise, proceed to rule c; Rule c: When neither rule a nor rule b can provide a valid judgment, or when their judgments conflict, compare the strength of their association: if Then the direction of causality is determined as follows: ;in, The significance coefficient is the preset value; if Then the direction of causality is determined as follows: Otherwise, the assignment of directed edges to the corresponding measurement point pairs is abandoned, or they are marked as bidirectional associations; each DCS measurement point in the DCS measurement point process topology network is a network node.

[0008] As a preferred embodiment of the artificial intelligence-based DCS production early warning method for thermal power plants described in this invention, the formation of the association mapping library includes: determining the corresponding network node path in the DCS measurement point process topology network based on the measurement point identifier of each alarm event in the alarm event chain sequence; recording the frequency of each network node path in the historical record and calculating the confidence level; associating and storing the alarm event chain sequence with the corresponding network node path and the confidence level to form an association mapping library; the confidence level is the frequency of the network node path in the historical record divided by the sum of the frequencies of all different network node paths mapped from the same starting node in the alarm event chain.

[0009] As a preferred embodiment of the artificial intelligence-based DCS production early warning method for thermal power plants described in this invention, step S2 includes: receiving and parsing DCS alarm signals online to generate standard alarm events; storing the standard alarm events in a sliding window of a preset time length in chronological order to form a real-time event sequence; simultaneously calculating the time interval between adjacent events in the real-time event sequence to form a real-time interval sequence; comparing the measurement point identifiers and alarm type order of the real-time event sequence with the starting links of each event chain in the associated mapping library; for starting links with consistent order, performing the following operations: obtaining the set of time interval values ​​between all adjacent events statistically obtained from the historical records of the starting link; and setting the first time interval value set as the first time interval value set. Percentiles are used as the lower bound of the interval, and the th Percentiles serve as the upper bound of the interval, constituting the matching tolerance interval of the starting link; it is determined whether each interval value in the real-time interval sequence falls within the corresponding matching tolerance interval; if they all fall within it, the current real-time alarm event sequence is determined to be compatible with the starting link; for all starting links determined to be compatible, the following operations are performed: extract the total frequency of occurrence in the historical records from the association mapping library, and the conditional probability that the corresponding event chain will eventually be confirmed to cause a fault in the history; multiply the total frequency and conditional probability after normalization respectively to obtain a reliability index characterizing the matching reliability; associate the reliability index with the end node of the event chain corresponding to the starting link in the association mapping library to form a potential fault point-reliability index association set; select the record with the highest reliability index from the potential fault point-reliability index association set, and output the associated end node as the currently determined potential fault point.

[0010] As a preferred embodiment of the artificial intelligence-based DCS production early warning method for thermal power plants described in this invention, the reverse tracing includes: taking the potential fault point as the starting node, performing multi-level traversal in the DCS measurement point process topology network along the opposite direction of the directed edges; recording all reachable upper-level network nodes and the tracing path from the upper-level network node to the starting node.

[0011] As a preferred embodiment of the artificial intelligence-based DCS production early warning method for thermal power plants described in this invention, the generation of the root cause node candidate list includes: for each tracing path, querying the historical fault case database and counting the number of cases in which each superior network node is ultimately confirmed as the root cause; calculating a root cause probability value for each superior network node on each tracing path based on the number of cases and the time decay factor of the case occurrence time; and sorting all superior network nodes in descending order according to the root cause probability values ​​to generate and output the root cause node candidate list.

[0012] As a preferred embodiment of the artificial intelligence-based DCS production early warning method for thermal power plants described in this invention, the fused early warning information includes the initial alarm event that has occurred, the predicted potential fault points, and at least one root cause node with the highest probability that is recommended for investigation.

[0013] Secondly, the present invention provides an artificial intelligence-based DCS production early warning system for thermal power plants, comprising: The event mapping library construction module is used to extract the time-series event chain of historical alarm records and map and associate it with the DCS measurement point process topology network to form an associated mapping library; The alarm matching and early warning module is used to monitor DCS alarm data online and match it with the associated mapping library; when the starting pattern of the event chain is matched, an early warning is immediately triggered and the end node of the event chain is located as a potential fault point. The root cause probability tracing module is used to trace back along the DCS measurement point process topology network starting from the potential fault point, calculate the probability of each upstream node becoming the root cause by combining historical cases, and generate a candidate list of root cause nodes. The integrated early warning push module is used to generate integrated early warning information containing phenomena, predictions, and suggested investigation objects based on the starting mode, potential fault points, and root cause node candidate list, and then push it out.

[0014] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the artificial intelligence-based DCS production early warning method for thermal power plants as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the artificial intelligence-based DCS production early warning method for thermal power plants as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By deeply integrating and modeling the historical DCS alarm timing event chain with the measurement point process topology network, this invention constructs an association mapping library, realizing a structured expression of complex alarm evolution patterns. This enables early warning judgment to no longer be limited to single-point threshold judgment, but to match and identify based on the timing evolution pattern under real working conditions, significantly improving the advance detection and accuracy of potential fault identification.

[0017] By using time interval tolerance ranges, reliability indicators, and conditional probability mechanisms, the false alarm rate is effectively reduced and the credibility of the early warning results is enhanced. At the same time, the present invention, based on the reverse tracing of the topology network and the root cause probability calculation mechanism driven by historical cases, can quickly generate a candidate list of root cause nodes with priority ranking, which greatly reduces the scope of manual investigation and diagnosis time.

[0018] Furthermore, by integrating early warning information, abnormal phenomena, potential fault predictions, and suggested investigation targets are uniformly pushed to operators, providing them with decision-oriented handling basis and significantly improving the intelligence level, safety assurance capability, and operation and maintenance efficiency of DCS operation and maintenance in thermal power plants. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of an AI-based DCS production early warning method for thermal power plants.

[0021] Figure 2 This is a structural diagram of an AI-based DCS production early warning system for thermal power plants.

[0022] Figure 3 A schematic diagram of the structure of an electronic device for implementing the artificial intelligence-based DCS production early warning method for thermal power plants according to an embodiment of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] Figure 1 This is a flowchart of an artificial intelligence-based DCS production early warning method for thermal power plants according to an embodiment of the present invention. Figure 1 As shown, the AI-based DCS production early warning method for thermal power plants includes: S1: Extract the time-series event chain of historical alarm records and map it with the DCS measurement point process topology network to form an association mapping library.

[0027] S1.1: Generate a time-series event chain.

[0028] Clean up historical DCS alarm records to eliminate duplicate and momentary jitter alarms; standardize each alarm into a standardized alarm event that includes measurement point identifier, alarm type and timestamp.

[0029] Specifically, a repetitive alarm suppression time window is set. For multiple alarm records of the same measurement point identifier and alarm type that are generated consecutively within the repetitive alarm suppression time window, only the first record is retained, eliminating repetitive alarm entries caused by signal jitter or communication delay. Further, an instantaneous alarm filtering threshold is set (which can be set to 2 to 3 times the regular data scanning cycle of the DCS system). Alarm records with a duration less than the instantaneous alarm filtering threshold are judged as instantaneous jitter noise and are discarded. After cleaning, standardized operations are performed: the database fields of the original alarm records are parsed, and the measurement point identifier is uniformly mapped to a combination key of a globally unique device code and measurement point descriptor; the alarm type is normalized to elements in a preset finite state set, which includes at least "high-high alarm", "high alarm", "low alarm", "low-low alarm", and "poor quality alarm"; and the timestamps are uniformly converted to a millisecond-level time format based on Coordinated Universal Time (UTC).

[0030] Furthermore, within a preset first time window, two consecutive standardized alarm events are defined as a time-series event pair; multiple consecutive time-series event pairs are connected in chronological order to form an alarm event chain sequence.

[0031] Specifically, the chain is built based on the time interval between adjacent events: starting from the initial event, if the time interval between the next event and the current event is not greater than the first time window, then it is added to the current chain and the process continues to check the next event; otherwise, the current chain is terminated and the event is used as the starting point of a new chain.

[0032] Label each alarm event chain with a tag indicating whether it has caused an actual fault. The method is as follows: extract historical fault event records from the power plant operation and maintenance management system (such as work order system, equipment defect record, maintenance report). Each record should include at least: the time of the fault occurrence, the confirmed faulty equipment / measuring point, the fault description and handling measures.

[0033] The alarm event chain is time-correlated with the fault event record: if the last alarm event in an alarm event chain occurs within a preset correlation time window (e.g., within 30 minutes) before the occurrence time of a fault event, and the device / measuring point involved in the fault event is related to the end node of the event chain, then the event chain is marked as "caused a fault"; otherwise, it is marked as "did not cause a fault".

[0034] This association process can be manually verified to form a labeled historical alarm event chain dataset for subsequent conditional probability calculations.

[0035] S1.2: Generate the process topology network of DCS measurement points.

[0036] Multiple time periods characterizing stable unit operation were selected from historical DCS data, and continuous time-series data of all DCS measurement points within each time period were extracted to form a normal operating condition dataset.

[0037] Based on the aforementioned normal operating condition dataset, the following steps are performed to generate the process topology network: (1) For any two different DCS measurement points and symmetrically calculate multiple preset time offsets The mutual information values ​​are calculated, and the mutual information sequences are obtained. The maximum value of each sequence is defined as the measurement point pair. correlation strength value and measuring point pair correlation strength value And the time offset corresponding to reaching the maximum value is denoted as . arrive Optimal latency ;and arrive Optimal latency The preset time offset can be preset to [-60 seconds, +60 seconds], and the value is taken at intervals of the system's basic sampling period (e.g., 1 second).

[0038] The mutual information value can be estimated using methods based on kernel density estimation or histogram statistics.

[0039] (2) The correlation strength value All measurement point pairs exceeding the intensity threshold (which can be set based on the quantiles of the mutual information distribution). Retain them to form a set of candidate causal edges.

[0040] (3) For each pair of measurement points in the candidate causal edge set The following rules shall be used to determine the appropriate criteria: Rule a: Decision 1: If and If the propagation delay is less than the upper limit of a reasonable propagation time (set based on process physical limitations), then the causal direction is initially determined to be... ; Judgment 2: If and If the delay is less than the upper limit of reasonable propagation time, then the causal direction is preliminarily determined to be... If both judgment 1 and judgment 2 are satisfied, then proceed to rule b to resolve the conflict; if only one is satisfied, then directly determine the causal direction based on the initial judgment.

[0041] Rule b: Query the preset process knowledge base. If the knowledge base contains... lie in The upstream process links are then used to determine the direction. If recorded lie in The upstream of, then determined as Otherwise, proceed to rule c.

[0042] Rule c: When rule a conflicts with rule b and a valid decision cannot be made, compare the strength of their association: if... Then the direction of causality is determined as follows: ;in, The significance coefficient is the preset value; if Then the direction of causality is determined as follows: Otherwise, abandon the assignment of directed edges to the corresponding test point pairs, or mark them as bidirectional associations. It is usually set between 0.1 and 0.3.

[0043] It should be noted that in the DCS measurement point process topology network, each DCS measurement point is a network node, and the weight of each directed edge can be defined as the correlation strength in that direction.

[0044] Optionally, in rule c, when comparing association strengths to determine causal direction, a statistical significance test is added to prevent misjudgment when strength values ​​are close. The specific steps are as follows: First, calculate... Then, the significance of the difference was further calculated: By using a permutation test-based method, the null hypothesis ( and (Non-directional causality). Simultaneously, the time alignment of the two sequences is shuffled (e.g., random time shift), and the correlation strength between the shuffled sequences is recalculated. This process is repeated multiple times (e.g., 1000 times) to obtain... The empirical distribution under the null hypothesis. This involves repeating the above shuffling and calculation process many times (e.g., 1000 times) to obtain an empirical distribution of the differences under the null hypothesis (a set of...). (Value). If you want to test (i.e., the difference is greater than 0), then the p-value = (in the empirical distribution, Greater than or equal to (Number of repetitions) / (Total number of repetitions). The p-value represents: if and If there is truly no causal relationship (null hypothesis is true), then the observation of things like... is merely due to randomness. What is the probability of such a large (or even larger) difference?

[0045] If the difference is greater than 0 and the p-value is less than the significance level (e.g., 0.05), then it is determined that... It is statistically significant; conversely, if the difference is less than 0 and the p-value is less than the significance level, then it is judged as statistically significant. Otherwise, it is considered that the directionality is not significant, and the directed edge is not assigned.

[0046] As can be seen, the time delay estimation method based on maximizing mutual information in this invention can automatically learn the dynamic correlation and propagation delay between measurement points from the data, without relying on a prior fixed model, and is particularly suitable for complex and nonlinear industrial processes. The improved causal discrimination rule integrates time delay information, prior process knowledge, and statistical significance testing, forming a hierarchical and robust judgment system, which significantly improves the accuracy of the constructed process topology network. The generated network not only reflects physical connections but also reveals the dynamic influence paths at the data level, providing a reliable graphical model foundation for subsequent fault propagation analysis.

[0047] S1.3: Form an associated mapping library.

[0048] Based on the measurement point identifier of each alarm event in the alarm event chain sequence, determine the corresponding network node path in the DCS measurement point process topology network; record the frequency of each network node path in the historical record, and calculate the confidence level.

[0049] Specifically, determining the path of a network node is achieved through a graph matching algorithm.

[0050] The DCS measurement point topology network is considered as a directed graph G=(V, E), where V is the set of network nodes (measurement points) and E is the set of directed edges. For an alarm event chain, firstly, the measurement point identifier of each event is mapped to the corresponding network node in graph G. Then, a path is found in G such that for any consecutive node pair in the path... In graph G, there exists a path from... arrive There exists a directed edge e∈E, or there exists a path from E to E. arrive A directed path (along the direction of the directed edge) whose length does not exceed the preset maximum path length. If found, this path is recorded as the network node path corresponding to the alarm event chain.

[0051] For example, the preset maximum path length can be set to twice the average path length of the network, or a fixed value (such as 5 hops) can be set directly based on engineering experience.

[0052] If the same event chain corresponds to multiple directed paths, the path with the largest sum of edge weights (or the path with the highest priority according to the process knowledge base) is selected as the representative path.

[0053] Furthermore, the alarm event chain sequence is associated with and stored along with the corresponding network node paths and confidence levels to form an association mapping library.

[0054] The confidence level is the frequency of the network node path appearing in the historical record, divided by the sum of the frequencies of different network node paths mapped from all alarm event chains originating from the same starting node. Here, the starting node is the network node corresponding to the first event in the event chain.

[0055] The mapping association additional storage includes the following derived metrics: 1) Path support: the absolute frequency of the mapping in history. 2) Path confidence. 3) Average propagation time: the average actual time interval between adjacent alarm events within the event chain in history, which can be compared with the theoretical latency implicit in the topology network.

[0056] The physical storage structure of the association mapping library is implemented using a key-value database or a graph database. The "key" is the hash value or serialized string of the standardized alarm event chain, and the "value" is a structure containing fields such as the mapped network node path (represented by a sequence of node IDs), path support, path confidence, and average propagation time. An inverted index is also established to allow for quick querying of which historical alarm chains are associated through network nodes.

[0057] As can be seen, mapping the time-series alarm chain to a static process topology using a graph matching algorithm achieves the fusion of dynamic alarm sequences and static process structures, giving the alarm sequence a spatial (topological) interpretation. The stored multi-dimensional confidence and time indicators not only provide frequency information but also offer a rich data foundation for similarity assessment (such as time tolerance interval comparison) during the online matching phase. The structured storage design ensures efficient querying and scalability, making the association mapping library a core knowledge base containing historical experience, supporting rapid pattern matching and reasoning in the online phase.

[0058] S2: Monitor DCS alarm data online and match it with the associated mapping library; when the starting pattern of the event chain is matched, immediately trigger an early warning and locate the end node of the event chain as a potential fault point.

[0059] S2.1: Receive and parse DCS alarm signals online to generate standard alarm events; store the standard alarm events in a sliding window of a preset time length in chronological order to form a real-time event sequence; at the same time, calculate the time interval between adjacent events in the real-time event sequence to form a real-time interval sequence.

[0060] The preset time length of the sliding window is dynamically set according to the maximum length and average time span of the historical alarm event chain in the associated mapping library. For example, it is set to 1.5 times the maximum time span of the historical event chain to ensure that the window can accommodate a complete potential event chain pattern.

[0061] The specific operation is as follows: The system maintains a time-sorted queue as a sliding window. When a new standard alarm event is generated, it is first inserted into the tail of the queue according to its timestamp. Subsequently, the timestamp of the event at the head of the queue is checked. If the difference between the timestamp and the current time exceeds the preset time length of the sliding window, the event is removed from the queue, ensuring that only alarm events within the latest time range are retained within the window. The current state of this queue constitutes the real-time event sequence. Simultaneously, after each window update, all adjacent event pairs in the real-time event sequence are traversed, the time interval is calculated, and stored sequentially to form a real-time interval sequence corresponding to the event sequence. The above operations ensure the timeliness and continuity of the processed data.

[0062] It should be noted that when receiving DCS alarm signals online, an asynchronous listening and parsing thread is established. The parsing process includes format verification: checking whether the signal contains the necessary measurement point identifier, alarm status, and time information fields. If a field is missing or the format is incorrect, a parsing error log is recorded, and an attempt is made to repair it according to pre-configured rules (such as using the previous valid value or the default value). If repair is not possible, the signal is discarded to prevent erroneous data from polluting the real-time sequence. The format of the generated standard alarm events is strictly consistent with the standardized format defined in S1.1, ensuring comparability with the patterns stored in the historical association mapping library. This fault tolerance mechanism ensures the robustness of the system in the face of communication interference or data source anomalies.

[0063] S2.2: Compare the measurement point identifiers and alarm type order of the real-time event sequence with the starting links of each event chain in the associated mapping library; for starting links with the same order, perform the following operations: Obtain the set of time interval values ​​between all adjacent events statistically obtained from the historical records of the starting point; and set the first value of the time interval value set... Percentiles are used as the lower bound of the interval, and the th Percentiles serve as the upper bound of the interval, constituting the matching tolerance interval of the initial stage; it is determined whether each interval value in the real-time interval sequence falls within the corresponding matching tolerance interval; if all fall within it, the current real-time alarm event sequence is determined to be compatible with the initial stage; wherein, It is a preset constant that is greater than 0 and less than 50.

[0064] The specific implementation of comparing the measurement point identifiers and alarm type order of real-time event sequences with the starting links of each event chain in the association mapping library is achieved through efficient matching using a prefix tree or finite state machine. The starting links of all event chains in the association mapping library (e.g., the first k events, where k can be a fixed value) are used as a pattern library to construct an index. The real-time event sequence is used as the input stream and synchronously compared with the index. When the measurement point identifiers and alarm type order of consecutive events in the real-time event sequence completely match a pattern of a certain starting link, a deeper compatibility judgment of that candidate starting link is triggered. This method reduces the complexity of a full library scan to approximately linear, meeting the requirements for online real-time performance.

[0065] S2.3: For all starting links determined to be compatible, perform the following operations: extract the total frequency of occurrence in the historical records from the association mapping library, and the conditional probability that the corresponding event chain will eventually be confirmed to cause a failure in the history; multiply the total frequency after normalization and the conditional probability after normalization to obtain a reliability index characterizing the reliability of the match; associate the reliability index with the end node of the event chain corresponding to the starting link in the association mapping library to form a potential failure point-reliability index association set.

[0066] The conditional probability is the number of times a failure occurs after the event chain occurs, divided by the total number of times the event chain occurs.

[0067] The total frequency is normalized using the min-max normalization method, resulting in the set of frequencies for all compatible candidates. If all frequencies are the same, the normalized total frequency is 0.5.

[0068] S2.4: From the set of potential failure points and reliability indicators, select the record with the highest reliability indicator, and output the associated end node as the currently determined potential failure point.

[0069] S3: Starting from the potential fault point, trace back along the DCS measurement point process topology network, calculate the probability of each upstream node becoming the root cause by combining historical cases, and generate a candidate list of root cause nodes.

[0070] The reverse tracing includes: taking the potential fault point as the starting node, performing multi-level traversal in the DCS measurement point process topology network along the opposite direction of the directed edges; recording all reachable upper-level network nodes and the tracing path from the upper-level network node to the starting node.

[0071] It should be noted that a maximum tracing depth needs to be preset (e.g., 5), representing the maximum number of hops required to trace backwards along the incoming edges from the potential fault point (starting node). A depth-first search (DFS) or breadth-first search (BFS) algorithm is used to perform a backward traversal of the DCS measurement point process topology network (modeled as a directed graph G). During traversal, the unique path from the currently visited parent network node to the starting node is recorded.

[0072] Furthermore, "recording all reachable parent network nodes and tracing paths" requires a specific data structure. A record is created for each visited parent network node, containing at least: the node identifier, the shortest path hop count (depth) from that node to the starting node, and a list of all distinct paths from the parent network node to the starting node (paths are represented by node sequences). Since industrial process topologies may contain feedback loops, the graph may contain cycles. To prevent traversal from entering an infinite loop, the algorithm must include a loop detection mechanism: maintaining a visit stack or set during traversal to record nodes visited on the current path. When attempting to visit a node already present in the current path, it indicates a loop has been detected; further tracing of that branch is immediately terminated, and the currently formed loop path is recorded as a special "cyclic causal path" for analysis, but nodes within it are not included in the regular root cause candidate node calculation (or given a very low initial probability).

[0073] As can be seen, this invention, by setting a maximum depth and using confidence-based path pruning, focuses the tracing process on the most relevant and likely upstream links, significantly improving computational efficiency and the practicality of the results. The structured recording method provides a foundation for subsequent path-based statistical analysis. A dedicated loop handling mechanism enhances the algorithm's adaptability and robustness to complex industrial topologies, avoiding algorithm failure or misjudgment caused by logical loops.

[0074] Furthermore, the generation of the root cause node candidate list includes the following steps: First, for each tracing path, query the historical failure case database and count the number of cases in which each superior network node was ultimately identified as the root cause.

[0075] The historical fault case database is constructed by integrating data sources such as DCS historical alarm records, maintenance event work orders, and equipment maintenance reports. Each case entry includes at least: a root cause measurement point identifier confirmed by maintenance, related measurement points exhibiting the fault, typically including the potential fault point of this alert, the fault occurrence time, and an optional inference or confirmation path from the root cause to the manifested node. During a query, for a traced parent network node, the database searches for all cases that meet the following conditions: the root cause measurement point identifier confirmed by maintenance is equal to the parent network node, and the related measurement points exhibiting the fault include the potential fault point of the current alert or a node directly adjacent to it in the topology. The total number of cases meeting the conditions is counted as the original case occurrence count.

[0076] Secondly, based on the number of occurrences of the cases and the time decay factor of the occurrence time of the cases, a root cause probability value is assigned to each of the parent network nodes on each tracing path.

[0077] The calculation of the root cause probability value includes: For each associated case z, calculate the time decay factor. : ; in, This is the decay coefficient, which controls the rate at which the weight of older cases decreases (it can be set based on the time span of the case library or the validity period of historical data). For the current time, This refers to the time when the case occurred.

[0078] Calculate the upper-level network node Weighted case count : ; in, This represents the number of times the original case occurred, which makes recent cases contribute more to the probability.

[0079] For all traceable candidate parent nodes Calculate the root cause probability value : ; It should be noted that the denominator must be greater than 0. If it is equal to 0, it means that the weighted case count of all candidate nodes is 0 (i.e., there are no relevant historical cases). In this case, the above formula cannot be used directly (divided by 0). Instead, a default probability allocation strategy needs to be enabled, such as giving all candidate nodes an equal probability or allocating probabilities according to the reciprocal of the topological distance from the node to the failure point.

[0080] Finally, all the parent network nodes are sorted in descending order according to the root cause probability value to generate and output a candidate list of root cause nodes.

[0081] S4: Based on the starting pattern, potential fault points, and root cause node candidate list, generate fused early warning information containing phenomena, predictions, and suggested investigation objects, and push it out.

[0082] The integrated early warning information includes the initial alarm event that has occurred, the predicted potential fault points, and at least one root cause node with the highest probability that should be investigated. The integrated early warning information is then filled into a preset early warning information template to automatically generate a structured early warning information containing three fields: "Occurred Phenomenon", "Potential Risk Prediction", and "Primary Investigation Recommendation".

[0083] Based on the node corresponding to the primary investigation recommendation, query the historical operation case library, extract the summary of typical handling measures related to the node, add the summary as an operation guide attachment to the structured early warning information, and send the structured early warning information and operation guide attachment to the target monitoring terminal.

[0084] Furthermore, such as Figure 2 As shown, this embodiment also provides an artificial intelligence-based DCS production early warning system for thermal power plants, including: The event mapping library construction module is used to extract the time-series event chain of historical alarm records and map and associate it with the DCS measurement point process topology network to form an associated mapping library; The alarm matching and early warning module is used to monitor DCS alarm data online and match it with the associated mapping library; when the starting pattern of the event chain is matched, an early warning is immediately triggered and the end node of the event chain is located as a potential fault point. The root cause probability tracing module is used to trace back along the DCS measurement point process topology network starting from the potential fault point, calculate the probability of each upstream node becoming the root cause by combining historical cases, and generate a candidate list of root cause nodes. The integrated early warning push module is used to generate integrated early warning information containing phenomena, predictions, and suggested investigation objects based on the starting mode, potential fault points, and root cause node candidate list, and then push it out.

[0085] This embodiment also provides a computer device applicable to the case of an artificial intelligence-based DCS production early warning method for thermal power plants, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the artificial intelligence-based DCS production early warning method for thermal power plants as proposed in the above embodiment.

[0086] like Figure 3As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0087] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0088] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as an AI-based DCS production early warning method for thermal power plants.

[0089] In some embodiments, the AI-based DCS production early warning method for thermal power plants can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the AI-based DCS production early warning method for thermal power plants described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the AI-based DCS production early warning method for thermal power plants by any other suitable means (e.g., by means of firmware).

[0090] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0091] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the artificial intelligence-based DCS production early warning method for thermal power plants as proposed in the above embodiments.

[0092] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for early warning of production in a thermal power plant's DCS based on artificial intelligence, characterized in that: include: S1: Extract the time-series event chain of historical alarm records and map and associate it with the DCS measurement point process topology network to form an association mapping library; S2: Monitor DCS alarm data online and match it with the associated mapping library; when the starting pattern of the event chain is matched, immediately trigger an early warning and locate the end node of the event chain as a potential fault point; S3: Starting from the potential fault point, trace back along the DCS measurement point process topology network, calculate the probability of each upstream node becoming the root cause by combining historical cases, and generate a candidate list of root cause nodes. S4: Based on the starting pattern, potential fault points, and root cause node candidate list, generate fused early warning information containing phenomena, predictions, and suggested investigation objects, and push it out.

2. The artificial intelligence-based DCS production early warning method for thermal power plants as described in claim 1, characterized in that: The generation of the time-series event chain includes: Clean up historical DCS alarm records to eliminate duplicate and momentary jitter alarms; standardize each alarm into a standardized alarm event that includes measurement point identifier, alarm type and timestamp; Within a preset first time window, two standardized alarm events that occur consecutively in time are defined as a time-series event pair; multiple consecutive time-series event pairs are connected in chronological order to form an alarm event chain sequence. The generation of the DCS measurement point process topology network includes: Multiple time periods characterizing stable unit operation were selected from historical DCS data, and continuous time-series data of all DCS measurement points within the time periods were extracted to form a normal operating condition dataset. Based on the aforementioned normal operating condition dataset, the following steps are performed to generate the process topology network: For any two different DCS measurement points and symmetrically calculate multiple preset time offsets The mutual information values ​​are calculated, and the mutual information sequences are obtained. The maximum value of each sequence is defined as the measurement point pair. correlation strength value and measuring point pair correlation strength value And the time offset corresponding to reaching the maximum value is denoted as . arrive Optimal latency ;and arrive Optimal latency ; correlation strength value All measuring points exceeding the intensity threshold Retain them to form a set of candidate causal edges; For each pair of measurement points in the candidate causal edge set The following rules shall be used to determine the appropriate criteria: Rule a: If and If the delay is less than the upper limit of reasonable propagation time, then the causal direction is preliminarily determined to be... ;like and If the delay is less than the upper limit of reasonable propagation time, then the causal direction is preliminarily determined to be... If both conditions are met, proceed to rule b to resolve the conflict; if only one condition is met, proceed directly to the initial causal direction. Rule b: Query the preset process knowledge base. If the knowledge base contains... lie in The upstream process links are then used to determine the direction. If recorded lie in The upstream of, then determined as Otherwise, proceed to rule c; Rule c: When neither rule a nor rule b can provide a valid decision, or when their decisions conflict, compare the strength of their association. like Then the direction of causality is determined as follows: ;in, The significance coefficient is the preset value. like Then the direction of causality is determined as follows: ; Otherwise, abandon the assignment of directed edges to the corresponding measurement point pairs, or mark them as bidirectional associations; In the DCS measurement point process topology network, each DCS measurement point serves as a network node, and the weight of each directed edge can be defined as the correlation strength in the corresponding direction.

3. The artificial intelligence-based DCS production early warning method for thermal power plants as described in claim 2, characterized in that: The formation of the association mapping library includes: Based on the measurement point identifier of each alarm event in the alarm event chain sequence, determine the corresponding network node path in the DCS measurement point process topology network; record the frequency of each network node path in the historical record, and calculate the confidence level; The alarm event chain sequence is associated with and stored with the corresponding network node path and confidence level to form an association mapping library; The confidence level is the frequency of the network node path appearing in the historical record, divided by the sum of the frequencies of different network node paths mapped by all alarm event chains originating from the same starting node.

4. The artificial intelligence-based DCS production early warning method for thermal power plants as described in claim 3, characterized in that: S2 includes: The system receives and parses DCS alarm signals online to generate standard alarm events; it stores the standard alarm events in a sliding window of a preset time length in chronological order to form a real-time event sequence; and it calculates the time interval between adjacent events in the real-time event sequence to form a real-time interval sequence. The measurement point identifiers and alarm type order of the real-time event sequence are compared with the starting links of each event chain in the association mapping library; for starting links with the same order, the following operations are performed: Obtain the set of time interval values ​​between all adjacent events statistically obtained from the historical records of the starting point; and set the first value of the time interval value set... Percentiles are used as the lower bound of the interval, and the th Percentiles serve as the upper bound of the interval, constituting the matching tolerance interval of the starting step; it is determined whether each interval value in the real-time interval sequence falls within the corresponding matching tolerance interval; if they all fall within it, it is determined that the current real-time alarm event sequence is compatible with the starting step. For all starting links determined to be compatible, perform the following operations: extract the total frequency of occurrence in the historical records from the association mapping library, as well as the conditional probability that the corresponding event chain will eventually be confirmed to cause a failure in the history; multiply the normalized total frequency and the normalized conditional probability by the normalized total frequency to obtain a reliability index characterizing the reliability of the match; associate the reliability index with the end node of the event chain corresponding to the starting link in the association mapping library to form a potential failure point-reliability index association set; From the set of potential failure points and reliability indicators, select the record with the highest reliability indicator, and output the associated end node as the currently determined potential failure point.

5. The artificial intelligence-based DCS production early warning method for thermal power plants as described in claim 4, characterized in that: The reverse tracing includes: Starting from the potential fault point, a multi-level traversal is performed in the DCS measurement point process topology network in the opposite direction of the directed edges; all reachable upper-level network nodes and the tracing path from the upper-level network node to the starting node are recorded.

6. The artificial intelligence-based DCS production early warning method for thermal power plants as described in claim 5, characterized in that: The generation of the root cause node candidate list includes: For each tracing path, query the historical failure case database and count the number of cases in which each superior network node was ultimately identified as the root cause. Based on the number of occurrences of the cases and the time decay factor of the time of occurrence, a root cause probability value is assigned to each of the parent network nodes on each tracing path. All the parent network nodes are sorted in descending order according to the root cause probability value, and a candidate list of root cause nodes is generated and output.

7. The artificial intelligence-based DCS production early warning method for thermal power plants as described in claim 6, characterized in that: The fused early warning information includes the initial alarm event that has occurred, the predicted potential fault points, and at least one root cause node with the highest probability that is recommended for investigation.

8. An AI-based DCS production early warning system for thermal power plants, based on the AI-based DCS production early warning method for thermal power plants according to any one of claims 1 to 7, characterized in that: Also includes: The event mapping library construction module is used to extract the time-series event chain of historical alarm records and map and associate it with the DCS measurement point process topology network to form an associated mapping library; The alarm matching and early warning module is used to monitor DCS alarm data online and match it with the associated mapping library; when the starting pattern of the event chain is matched, an early warning is immediately triggered and the end node of the event chain is located as a potential fault point. The root cause probability tracing module is used to trace back along the DCS measurement point process topology network starting from the potential fault point, calculate the probability of each upstream node becoming the root cause by combining historical cases, and generate a candidate list of root cause nodes. The integrated early warning push module is used to generate integrated early warning information containing phenomena, predictions, and suggested investigation objects based on the starting mode, potential fault points, and root cause node candidate list, and then push it out.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the artificial intelligence-based DCS production early warning method for thermal power plants as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based DCS production early warning method for thermal power plants as described in any one of claims 1 to 7.

Images