Equipment state monitoring system based on power plant production real-time data

Abstract

The invention discloses an equipment state monitoring system based on power plant production real-time data, which relates to the technical field of power plant equipment state monitoring, and comprises the steps of constructing an equipment topological graph module, collecting and preprocessing real-time operation parameters, generating a standardized real-time data sequence, identifying a steady-state working condition time period of equipment based on the standardized real-time data sequence, and monitoring the state of the equipment according to the steady-state working condition time period. Constructing an equipment topological graph according to the power plant process diagram; and a form difference degree module for extracting an operation instruction process parameter change sequence from the standardized real-time data sequence to obtain a to-be-evaluated operation sequence when the load regulation operation is identified, calling a reference operation sequence recorded by the equipment in a healthy state, and adopting a dynamic time warping algorithm to carry out evaluation on the to-be-evaluated operation sequence. And calculating the morphological difference degree between the to-be-evaluated operation sequence and the reference operation sequence. According to the invention, the accuracy and foresight of monitoring the state of the power plant equipment are improved by realizing the crossing from passive alarm to active prediction and from phenomenon description to essential diagnosis.

Description

Technical Field

[0001] This invention relates to the field of power plant equipment status monitoring technology, and in particular to an equipment status monitoring system based on real-time power plant production data. Background Technology

[0002] In the process of intelligentization in the power industry, power plant equipment condition monitoring technology is the core of ensuring safe and economical operation. Existing technologies are mainly based on real-time data from plant-level monitoring information systems and distributed control systems. They trigger alarms by setting parameter thresholds and calculate performance indicators when the equipment is running stably, comparing and analyzing them with design values ​​or historical benchmarks. This threshold alarm + steady-state performance analysis model constitutes the main technical foundation of current equipment condition monitoring.

[0003] However, existing technologies focus on monitoring the stable operating status of equipment, and are insufficient in assessing the health status of dynamic operation processes such as start-up, shutdown, and load adjustment. Due to the complex changes in transient process parameters, traditional methods are unable to effectively capture early abnormal characteristics contained in the operation sequence, resulting in delayed warnings and limiting the improvement of predictive maintenance capabilities. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an equipment status monitoring system based on real-time power plant production data, which solves the problem of delayed early warning caused by insufficient early anomaly capture capability of existing technologies in transient processes.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an equipment status monitoring system based on real-time power plant production data, comprising: a module for constructing an equipment topology map; collecting and preprocessing real-time operating parameters; generating standardized real-time data sequences; identifying steady-state operating time periods of equipment based on the standardized real-time data sequences; and constructing an equipment topology map based on the power plant process diagram.

[0008] The morphological difference module, when a load adjustment operation is detected, extracts the sequence of changes in operation command process parameters from the standardized real-time data sequence to obtain the operation sequence to be evaluated, calls the baseline operation sequence recorded by the equipment in a healthy state, and uses a dynamic time warping algorithm to calculate the morphological difference between the operation sequence to be evaluated and the baseline operation sequence.

[0009] The operation sequence abnormal signal module inputs the aligned operation sequence into a one-dimensional convolutional neural network, extracts the local morphological features of the sequence in the time dimension through the convolutional kernel, and outputs a quantified operation health score through a fully connected layer. The operation health score is evaluated, and an operation sequence abnormal signal is triggered based on the evaluation result.

[0010] The comparison module calculates the current operating efficiency of the equipment based on standardized real-time data sequences during the steady-state operating period of the equipment, compares the current operating efficiency with the historical baseline efficiency of the equipment's health, and generates a performance deviation warning signal based on the comparison result.

[0011] The causal network graph module receives deviation warning signals and, based on the constraints of the connection relationships between nodes in the equipment topology graph, runs a constraint-based causal discovery algorithm on normal operating data to automatically learn and construct a causal network graph describing the causal relationships between process parameters.

[0012] The structured diagnostic report module uses abnormal parameters corresponding to abnormal signals in the operation sequence as result nodes in the causal network diagram. It locates the root cause parameters through reverse path probability analysis and generates a structured diagnostic report.

[0013] As a preferred embodiment of the equipment status monitoring system based on real-time power plant production data described in this invention, the process of collecting and preprocessing real-time operating parameters to generate a standardized real-time data sequence includes the following steps:

[0014] Real-time operating parameters are collected from distributed control units and plant-level monitoring information units, and null values ​​are removed from the collected real-time operating parameters.

[0015] The real-time operating parameters that have completed null value removal are subjected to outlier correction processing, and the real-time operating parameters that have completed outlier correction are subjected to timestamp alignment processing to generate a standardized real-time data sequence.

[0016] As a preferred embodiment of the equipment status monitoring system based on real-time power plant production data according to the present invention, the system includes the following steps: identifying the steady-state operating time period of the equipment based on standardized real-time data sequences, and constructing an equipment topology map based on the power plant process diagram:

[0017] The first-order difference method is used to calculate the rate of change of power signal in a continuous time window in a standardized real-time data sequence. Based on the statistical analysis results of the rate of change of power signal in a known stable operation history of the unit, a steady-state judgment threshold is set. The rate of change of power signal is compared with the steady-state judgment threshold to obtain the candidate steady-state operating condition time period.

[0018] The volatility within the candidate steady-state operating condition time period is verified to obtain the final steady-state operating condition time period;

[0019] The process diagram of the power plant is analyzed to identify the pumps, fans, and heat exchangers shown in the diagram as nodes in the equipment topology diagram. The process pipelines are also analyzed to identify the directed edges in the equipment topology diagram. Based on the connection relationship between the equipment and pipelines in the power plant process diagram, the nodes and directed edges are associated to construct the equipment topology diagram.

[0020] As a preferred embodiment of the equipment status monitoring system based on real-time power plant production data described in this invention, the system includes the following steps: When a load regulation operation is detected, the system extracts the sequence of changes in operation command process parameters from a standardized real-time data sequence to obtain the operation sequence to be evaluated, and calls the baseline operation sequence recorded by the equipment in a healthy state.

[0021] By monitoring the load setpoint commands issued by the distributed control unit, the start and end times of the load adjustment operation can be identified.

[0022] Based on the start and end times of the identified load regulation operations, process parameter data related to the operation are extracted from the standardized real-time data sequence within the corresponding time period to form a process parameter change sequence.

[0023] The extracted sequence of process parameter changes is defined as the sequence of operations to be analyzed and evaluated in this study.

[0024] Based on the type and magnitude of the current load regulation operation, the baseline operation sequence is obtained by retrieving and calling the standard process parameter change sequence recorded under the same operation type and magnitude from the database that stores historical operating data of the equipment in its healthy state.

[0025] As a preferred embodiment of the equipment status monitoring system based on real-time power plant production data described in this invention, the following steps are included: A dynamic time warping algorithm is used to calculate the morphological difference between the operation sequence to be evaluated and the baseline operation sequence.

[0026] A dynamic time warping algorithm is used to non-linearly align the operation sequence to be evaluated with the benchmark operation sequence on the time axis to construct a cumulative distance matrix;

[0027] In the cumulative distance matrix, find the path with the minimum cumulative distance from the starting point to the ending point, and use the point-by-point accumulation method to obtain the sum of the Euclidean distances of the corresponding data points on the optimal curved path, thus obtaining the preliminary sequence morphological difference degree.

[0028] As a preferred embodiment of the equipment status monitoring system based on real-time power plant production data described in this invention, the system includes the following steps: The aligned operation sequence is input into a one-dimensional convolutional neural network; local morphological features of the sequence in the time dimension are extracted through convolutional kernels; a quantified operation health score is output through a fully connected layer; the operation health score is evaluated; and an operation sequence abnormality signal is triggered based on the evaluation result.

[0029] Subtract the mean of the process parameter in the baseline operation sequence from the data of each process parameter in the aligned operation sequence, and then divide by the standard deviation to generate the normalized aligned operation sequence.

[0030] The normalized alignment operation sequence is fed into a pre-trained one-dimensional convolutional neural network. The input layer dimension of the one-dimensional convolutional neural network is matched with the time step and feature dimension of the normalized alignment operation sequence.

[0031] The first convolutional layer of a one-dimensional convolutional neural network uses a convolutional kernel to slide across the time dimension of the normalized alignment operation sequence to extract local morphological features from the sequence and output a feature map.

[0032] By inputting the normalized alignment operation sequence into a one-dimensional convolutional neural network, and processing it layer by layer through convolutional and pooling layers, a feature vector is obtained. The feature vector is then input into a fully connected layer with a sigmoid activation function, and the fully connected layer outputs a scalar value.

[0033] The mean and standard deviation of the health score for normal operations in recent history are calculated using the sliding window method. The mean minus twice the standard deviation is set as the dynamic threshold. The current operation health score is compared with the dynamic threshold.

[0034] When the operation health score falls below the dynamic threshold, an operation sequence abnormality signal is triggered.

[0035] As a preferred embodiment of the equipment status monitoring system based on real-time power plant production data described in this invention, the following steps are included: During a period when the equipment is in a steady-state operating condition, the current operating efficiency of the equipment is calculated based on a standardized real-time data sequence:

[0036] Based on the start and end timestamps of the steady-state operating period, process parameter data points are extracted from the standardized real-time data sequence;

[0037] Substitute the process parameter data points into the thermodynamic efficiency calculation formula to calculate the instantaneous efficiency;

[0038] The instantaneous efficiency values ​​are averaged to obtain the current operating efficiency.

[0039] As a preferred embodiment of the equipment status monitoring system based on real-time power plant production data described in this invention, the following steps are included: comparing the current operating efficiency with the historical baseline efficiency of the equipment health, and generating a performance deviation early warning signal based on the comparison result:

[0040] The system retrieves the historical baseline efficiency of the equipment health from the historical database, obtains the absolute difference between the current operating efficiency and the historical baseline efficiency of the equipment health, and thus obtains the efficiency deviation value.

[0041] Based on the normal fluctuation range of efficiency obtained from long-term operation of equipment in a healthy state, a deviation threshold is set. The efficiency deviation value is compared with the deviation threshold. When the efficiency deviation value exceeds the deviation threshold, a performance deviation warning signal is generated.

[0042] As a preferred embodiment of the equipment status monitoring system based on real-time power plant production data described in this invention, the following steps are included: receiving deviation warning signals based on constraints on the connection relationships between nodes in the equipment topology diagram, running a constraint-based causal discovery algorithm on the normal operation data, and automatically learning and constructing a causal network diagram describing the causal relationships between process parameters:

[0043] Based on the abnormal device identifier in the performance deviation warning signal, starting from the abnormal device identifier, the upstream and downstream devices of the neighboring devices are found in the device topology map to form a local device set;

[0044] Extract time series data of process parameters of all devices in a local device set under recent normal operating conditions from the historical database to form a target data subset;

[0045] Based on the connection relationships of nodes in the device topology graph, a set of constraint conditions for conditional independence and causal direction prior is generated.

[0046] The PC causal discovery algorithm is run on a subset of the target data. A fully connected undirected graph is initialized on the subset of the target data. Based on the conditional independence constraints in the constraint set, the conditional independence of the edges in the undirected graph is checked to obtain a partially directed acyclic graph.

[0047] By orienting edges in a portion of a directed acyclic graph, a causal network graph describing the causal relationships between process parameters is generated.

[0048] As a preferred embodiment of the equipment status monitoring system based on real-time power plant production data described in this invention, the following steps are included: In the causal network diagram, taking the abnormal parameters corresponding to the abnormal signals of the operation sequence as the result nodes, the root cause parameters are located through reverse path probability analysis, and a structured diagnostic report is generated:

[0049] In the causal network diagram of causal relationships, the abnormal parameters corresponding to the abnormal signals of the operation sequence are marked as result nodes;

[0050] Starting from the marked fruit node, traverse backwards along all directed edges in the causal network graph that point to this fruit node, and record each complete directed path from the cause node to the fruit node.

[0051] For each directed path obtained through traversal, based on Bayes' theorem, historical data is used to calculate the conditional probability that each cause node on the path causes the effect node to be abnormal. The conditional probabilities on the directed path are compared, and the cause node with the largest conditional probability value is determined as the root cause parameter.

[0052] By integrating root cause parameters, directed paths, and conditional probabilities, a structured diagnostic report is generated.

[0053] The beneficial effects of this invention are as follows: By constructing a collaborative module for equipment topology diagrams, morphological difference, abnormal operation sequence signals, comparison, causal network diagrams, and structured diagnostic reports, a closed-loop process from data acquisition to intelligent diagnosis is achieved. Real-time operating parameters are standardized, steady-state conditions are identified, and a topology diagram reflecting equipment connectivity is constructed. Furthermore, dynamic time warping algorithms and one-dimensional convolutional neural networks are used to analyze the morphological characteristics of transient operation sequences such as load regulation, quantifying health status and triggering abnormal signals. Efficiency deviation warnings are provided during the steady-state phase. Based on equipment topology constraints and causal discovery algorithms, root cause parameters are located, generating structured diagnostic reports. This deeply integrates temporal morphological analysis, causal reasoning, and equipment topology, solving the pain points of traditional monitoring methods' insufficient early anomaly detection and difficulty in root cause localization during transient processes. It achieves a leap from passive alarm to proactive prediction, and from phenomenon description to essential diagnosis, improving the accuracy and foresight of power plant equipment status monitoring. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the 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.

[0055] Figure 1 This is a schematic diagram of an equipment status monitoring system based on real-time power plant production data.

[0056] Figure 2 A flowchart for identifying load regulation operations. Detailed Implementation

[0057] 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.

[0058] 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.

[0059] 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.

[0060] Reference Figure 1 and Figure 2 This is one embodiment of the present invention, which provides an equipment status monitoring system based on real-time power plant production data, comprising the following steps:

[0061] The module for constructing equipment topology maps collects and preprocesses real-time operating parameters, generates standardized real-time data sequences, identifies the steady-state operating time periods of equipment based on the standardized real-time data sequences, and constructs equipment topology maps according to the power plant process diagram.

[0062] Collect and preprocess real-time operating parameters to generate standardized real-time data sequences.

[0063] Real-time operating parameters are collected from distributed control units and plant-level monitoring information units, and null values ​​are removed from the collected real-time operating parameters.

[0064] Furthermore, by deploying a data interface unit at the boundary of the power plant control network, real-time operating parameters of key equipment are read in parallel from the real-time databases of the process controllers of the distributed control unit and the plant-level monitoring information unit at a specific sampling frequency. The types of real-time operating parameters collected include analog measurement values ​​and switch status values. Based on the real-time operating parameter data stream, null data points caused by communication interruption or signal loss are identified. For the identified null data points, they are directly removed from the data stream to form a real-time operating parameter sequence with null values ​​removed.

[0065] The real-time operating parameters that have completed null value removal are subjected to outlier correction processing, and the real-time operating parameters that have completed outlier correction are subjected to timestamp alignment processing to generate a standardized real-time data sequence.

[0066] Furthermore, based on the collected real-time operating parameter data stream, null data points caused by communication interruptions or signal loss are identified. These null data points are directly removed from the data stream, forming a real-time operating parameter sequence with null value removal. Since the real-time operating parameters from the distributed control unit and the plant-level monitoring information unit have microsecond-level differences in acquisition time, timestamp alignment is performed using a unified clock source as a reference. An alignment time window is defined for each parameter channel, and the values ​​of all parameter data points falling within the same time window are calculated using linear interpolation to obtain their values ​​at the center of the window. The set of values ​​for all parameters under a unified timestamp is then generated as the final standardized real-time data sequence.

[0067] Based on standardized real-time data sequence identification of steady-state operating time periods of equipment, an equipment topology diagram is constructed according to the power plant process diagram.

[0068] The first-order difference method is used to calculate the rate of change of power signal in a standardized real-time data sequence within a continuous time window. Based on the statistical analysis results of the rate of change of power signal in a known stable operating history of the unit, a steady-state judgment threshold is set. The rate of change of power signal is compared with the steady-state judgment threshold to obtain the candidate steady-state operating condition time period.

[0069] Furthermore, power signal data points are extracted from the standardized real-time data sequence. Two consecutive data points are taken in chronological order, and the value of the latter data point is subtracted from the value of the former data point, then divided by the time interval between the two data points. The result is the instantaneous power signal change rate corresponding to that time point. The power signal change rate sequence is obtained by calculating all continuous data points applied to the entire continuous time window. The power signal change rate data of the unit in the known stable operation history recorded in the historical database is queried. The absolute value of the historical change rate data is calculated, and the high percentile value of its statistical distribution is taken as the steady-state judgment threshold. Each instantaneous change rate value in the power signal change rate sequence is compared with the steady-state judgment threshold set in step two. When the absolute value of the power signal change rate is lower than the steady-state judgment threshold for a continuous period of time, the time period is marked as a candidate steady-state operating condition time period.

[0070] It should be noted that for the power signal in the standardized real-time data sequence, two consecutive data points are taken in chronological order. The value of the latter data point is subtracted from the value of the former data point, and then divided by the time interval between the two data points. The result is the instantaneous rate of change corresponding to that time point. Applying this calculation to all consecutive data points within the entire time window will yield a rate of change sequence that reflects the trend of the power signal over the time period.

[0071] The volatility of the candidate steady-state operating condition time period is verified to obtain the final steady-state operating condition time period.

[0072] Furthermore, within the candidate steady-state operating condition time period, check whether the rate of change of other key process parameters in the standardized real-time data sequence also remains at a low level. If the verification is successful, the candidate steady-state operating condition time period is confirmed as the final steady-state operating condition time period.

[0073] The process diagram of the power plant is analyzed to identify the pumps, fans, and heat exchangers shown in the diagram as nodes in the equipment topology diagram. The process pipelines are also analyzed to identify the directed edges in the equipment topology diagram. Based on the connection relationship between the equipment and pipelines in the power plant process diagram, the nodes and directed edges are associated to construct the equipment topology diagram.

[0074] Furthermore, the system reads the element information of the power plant process diagram, identifies specific graphic symbols representing pumps, fans, and heat exchangers, and marks equipment identifiers. Each equipment identifier is treated as a node in the equipment topology diagram. The system also reads the connection information of the power plant process diagram, identifies the lines representing process pipelines, and determines the direction of each pipeline based on the arrow markings or process flow direction. Each directed pipeline is treated as a directed edge in the equipment topology diagram. Based on the connection endpoint relationship between equipment symbols and pipeline lines in the power plant process diagram, the system establishes the connection relationship between nodes in the equipment topology diagram and directed edges.

[0075] The morphological difference module, when a load adjustment operation is detected, extracts the sequence of changes in operation command process parameters from the standardized real-time data sequence to obtain the operation sequence to be evaluated, calls the baseline operation sequence recorded by the equipment in a healthy state, and uses a dynamic time warping algorithm to calculate the morphological difference between the operation sequence to be evaluated and the baseline operation sequence.

[0076] When a load adjustment operation is detected, the sequence of changes in the operation command process parameters is extracted from the standardized real-time data sequence to obtain the operation sequence to be evaluated, and the baseline operation sequence recorded by the equipment in a healthy state is called up.

[0077] By monitoring the load setpoint commands issued by the distributed control unit, the start and end times of the load adjustment operation can be identified.

[0078] Furthermore, the load setpoint command signal sent by the distributed control unit to the unit is continuously monitored; when the value of the load setpoint command changes by more than 2 percent of the rated load per minute, this moment is recorded as the start time of the load adjustment operation. The load setpoint command is continuously monitored, and when the value stabilizes near the new setpoint within a predetermined time window and the fluctuation is less than the threshold, this stabilization moment is recorded as the end time of the load adjustment operation.

[0079] Based on the start and end times of the identified load regulation operations, data on process parameters related to the operation within the corresponding time period are extracted from the standardized real-time data sequence to form a process parameter change sequence.

[0080] Furthermore, based on the start and end times of the load regulation operation, data within the time period from the start to the end time is extracted from the standardized real-time data sequence; process parameters that are strongly correlated with the current load regulation operation are selected from the data within this time period, and then combined in chronological order to form a process parameter change sequence.

[0081] The extracted sequence of process parameter changes is defined as the sequence of operations to be analyzed and evaluated in this study.

[0082] Furthermore, the process parameter change sequence is formally labeled and defined as the operation sequence to be evaluated, which will be used for subsequent comparative analysis with the benchmark operation sequence.

[0083] Based on the type and magnitude of the current load regulation operation, the baseline operation sequence is obtained by retrieving and calling the standard process parameter change sequence recorded under the same operation type and magnitude from the database that stores historical operating data of the equipment in its healthy state.

[0084] Furthermore, based on the characteristics of the current load regulation operation and the specific load change range, the query conditions are used to search the database containing historical operating data recorded when the equipment is in a healthy state. The database is searched for the standard process parameter change sequence recorded when the equipment was confirmed to be in a healthy state under the same operation category and range in history. The retrieved results are then used as the baseline operation sequence for this comparison.

[0085] The dynamic time warping algorithm is used to calculate the morphological difference between the operation sequence to be evaluated and the benchmark operation sequence.

[0086] A dynamic time warping algorithm is used to nonlinearly align the operation sequence to be evaluated with the benchmark operation sequence on the time axis to construct a cumulative distance matrix.

[0087] Furthermore, initialize a matrix of size equal to the length of the operation sequence to be evaluated multiplied by the length of the reference operation sequence, called the cumulative distance matrix; calculate the Euclidean distance between each pair of data points in the operation sequence to be evaluated and the reference operation sequence, as the base distance matrix; apply the recursive rules of the dynamic time warping algorithm, starting from the starting point of the cumulative distance matrix, the cumulative distance value of each position is the base distance of the current point pair plus the minimum value of the cumulative distances of the three adjacent positions to the left, above, and upper left, and gradually fill the entire cumulative distance matrix.

[0088] It should be noted that the process involves iterating through each data point in the operation sequence to be evaluated and pairing it with each data point in the benchmark operation sequence to form multiple pairs of data points. For any pair of data points, one point comes from the operation sequence to be evaluated and the other comes from the benchmark operation sequence. The values ​​of these two data points are extracted in all feature dimensions. For each feature dimension, the difference between the value of the data point in the operation sequence to be evaluated and the value of the data point in the benchmark operation sequence is calculated. The difference in each feature dimension is squared, and the squared differences in all feature dimensions are summed. The square root of the sum is then taken to obtain the Euclidean distance between the data points.

[0089] In the cumulative distance matrix, find the path with the minimum cumulative distance from the starting point to the ending point, and use the point-by-point accumulation method to obtain the sum of the Euclidean distances of the corresponding data points on the optimal curved path, thus obtaining the preliminary sequence morphological difference degree.

[0090] The preliminary expression for sequence morphological dissimilarity is: ;

[0091] in, For preliminary sequence morphological differences, The total number of matching pairs. For the optimal curved path, the first Euclidean distance between matching point pairs To evaluate the first operation in the sequence The matching point at the th ... Values ​​in each feature dimension For the first operation in the baseline sequence The matching point at the th ... Values ​​in each feature dimension The total number of feature dimensions. Indexed by feature dimensions, For the index of the matching point, This refers to the temporal index position of the data point in the operation sequence to be evaluated.

[0092] Furthermore, by employing a dynamic time warping algorithm to find the optimal bending path and calculate the preliminary sequence morphological difference, nonlinear elastic matching of the device operation sequence on the time axis is achieved. This effectively eliminates the stretching and distortion effects of the sequence in the time dimension caused by natural fluctuations in operating speed, thus focusing the analysis on the differences in the operation morphology itself. Compared to directly comparing the original sequence, the advantage lies in its ability to more accurately capture subtle morphological anomalies related to time misalignment during operation, such as anomalies in the acceleration curve during startup or small oscillations during stabilization. This provides cleaner and more comparable input features for subsequent deep learning-based health assessments, significantly improving the sensitivity and accuracy of early anomaly detection.

[0093] It should be noted that existing technologies for analyzing time-series data typically employ fixed-time-window truncation and comparison methods, or simple linear time warping. Essentially, these methods assume that the sequences being compared have a strict alignment relationship or a fixed scaling ratio on the time axis. This assumption often fails in the dynamic industrial operating environment, leading to minor time axis shifts being misjudged as morphological anomalies, or true morphological differences being masked by time misalignment. The core difference between the steps described in this invention and existing technologies lies in the introduction of dynamic time warping, a non-linear, elastic matching algorithm. This algorithm does not pre-determine any time transformation model but instead autonomously finds an optimal curvature path with the minimum cumulative distance through dynamic programming. This achieves precise compensation for complex time deformations between sequences. This difference allows the method to adapt to actual changes in operating speed, fundamentally overcoming the false positives and false negatives caused by the rigid time alignment assumptions of existing technologies. It represents a technological leap from comparison under coarse alignment to morphological analysis after precise deformation compensation.

[0094] The operation sequence abnormality signal module inputs the aligned operation sequence into a one-dimensional convolutional neural network, extracts the local morphological features of the sequence in the time dimension through the convolutional kernel, and outputs a quantified operation health score through a fully connected layer. The operation health score is evaluated, and the operation sequence abnormality signal is triggered based on the evaluation result.

[0095] The aligned operation sequence is input into a one-dimensional convolutional neural network. The local morphological features of the sequence in the time dimension are extracted by the convolutional kernel, and the quantified operation health score is output through a fully connected layer. The operation health score is evaluated, and an operation sequence abnormal signal is triggered based on the evaluation result.

[0096] The normalized aligned operation sequence is generated by subtracting the mean of the process parameter in the baseline operation sequence from the data of each process parameter in the aligned operation sequence, and then dividing by the standard deviation.

[0097] Furthermore, by subtracting the mean of each process parameter in the baseline operation sequence from the data in the aligned operation sequence and dividing by the standard deviation, a normalized aligned operation sequence is generated. This achieves dimensional unification and distribution standardization of multi-source heterogeneous data, eliminating the differences in dimensions and numerical ranges of different process parameters.

[0098] The normalized alignment operation sequence is fed into a pre-trained one-dimensional convolutional neural network, where the input layer dimension of the one-dimensional convolutional neural network matches the time step and feature dimension of the normalized alignment operation sequence.

[0099] Furthermore, the normalized alignment operation sequence is used as input data and fed into a pre-trained one-dimensional convolutional neural network. The input layer of the one-dimensional convolutional neural network is designed to receive a two-dimensional array, where the number of rows in the array is equal to the number of time steps contained in the normalized alignment operation sequence, and the number of columns in the array is equal to the number of process parameter types contained in the normalized alignment operation sequence.

[0100] It should be noted that the dataset based on historical operation sequences consists of records of normal operation under the device's health status, each sequence of which has been labeled as a healthy sample. The normalized historical operation sequences are used as input features, and their corresponding health status labels are used as training targets. Backpropagation and gradient descent optimizers (e.g., the Adam optimizer) are used to iteratively update parameters such as kernel weights and bias terms in the one-dimensional convolutional neural network, aiming to minimize the binary cross-entropy loss between the predicted output and the true label. Through multiple iterations of training, the one-dimensional convolutional neural network learns to extract representative features from the healthy operation sequences. When the loss function of the one-dimensional convolutional neural network converges on the validation set and its accuracy stabilizes, the trained one-dimensional convolutional neural network is obtained.

[0101] The first convolutional layer of a one-dimensional convolutional neural network uses a convolutional kernel to slide along the time dimension of the normalized alignment operation sequence to extract local morphological features from the sequence and output a feature map.

[0102] Furthermore, the first convolutional layer of a one-dimensional convolutional neural network contains multiple convolutional kernels with fixed widths (e.g., a kernel width of 5 time steps). Each kernel slides along the time dimension of the normalized alignment operation sequence, covering a local time window at each slide. At each window position, the kernel performs a convolution operation with the covered local sequence segment and generates an activation value through the ReLU activation function. After the kernel traverses the entire sequence, it outputs a feature map, which records the local morphological features in the sequence that match the kernel pattern.

[0103] By inputting the normalized alignment operation sequence into a one-dimensional convolutional neural network, and processing it layer by layer through convolutional and pooling layers, a feature vector is obtained. The feature vector is then input into a fully connected layer with a sigmoid activation function, and the fully connected layer outputs a scalar value.

[0104] Furthermore, the normalized alignment operation sequence is passed sequentially through multiple convolutional and pooling layers in a one-dimensional convolutional neural network; the convolutional layers gradually extract deeper features, while the pooling layers reduce the dimensionality of the feature map; after processing by all layers, it is finally compressed into a fixed-length feature vector; this feature vector is then input into a fully connected layer with a Sigmoid activation function, which maps the feature vector to a scalar value between zero and one.

[0105] The mean and standard deviation of health scores for normal operations in recent history are calculated using a sliding window method. The mean minus twice the standard deviation is set as a dynamic threshold, and the current operation health score is compared with the dynamic threshold.

[0106] Furthermore, a sliding window containing recent occurrences is set; all health scores marked as normal operations within this window are obtained from the history; the arithmetic mean and standard deviation of the scores are calculated; the mean minus twice the standard deviation is set as the dynamic threshold; and the current operation health score obtained in step four is compared with this dynamic threshold.

[0107] It should be noted that the arithmetic mean of the health scores is obtained by summing the health scores of all normal operations within the sliding window and dividing the sum by the total number of health scores. The difference between each health score and the arithmetic mean is then calculated, and each difference is squared. All squared differences are summed, and the variance is obtained by dividing the sum by the total number of health scores. The standard deviation of the health score is then calculated by taking the square root of the variance.

[0108] When the operation health score falls below the dynamic threshold, an operation sequence abnormality signal is triggered.

[0109] Furthermore, if the current operational health score is lower than the dynamic threshold set in step five, an operational sequence abnormality signal is immediately triggered. This operational sequence abnormality signal includes the timestamp of the abnormal operation, the associated device identifier, and the specific operational health score value.

[0110] The comparison module calculates the current operating efficiency of the equipment based on standardized real-time data sequences during the period when the equipment is in steady-state operation, compares the current operating efficiency with the historical baseline efficiency of the equipment's health, and generates a performance deviation warning signal based on the comparison result.

[0111] During the period when the equipment is in steady-state operation, the current operating efficiency of the equipment is calculated based on standardized real-time data sequences.

[0112] Based on the start and end timestamps of the steady-state operating period, process parameter data points are extracted from the standardized real-time data sequence.

[0113] Furthermore, based on the start and end timestamps of the identified steady-state operating period, all data points within that period are extracted from the standardized real-time data sequence; process parameters required for calculating efficiency are then selected from the data points. For example, for a water pump, data points for parameters such as motor power, inlet pressure, outlet pressure, and flow rate are selected to form a set of process parameter data points.

[0114] Substitute the process parameter data points into the thermodynamic efficiency calculation formula to calculate the instantaneous efficiency value.

[0115] The instantaneous efficiency value is expressed as: ;

[0116] in, This is the instantaneous efficiency value. The mechanical transmission efficiency of the pump. The density of water, It is the acceleration due to gravity. For traffic, Due to export pressure, Due to import pressure, This refers to the motor power.

[0117] Furthermore, by precisely extracting data based on steady-state operating time periods and substituting it into the thermodynamic efficiency calculation formula to calculate instantaneous efficiency values ​​point by point, this method achieves a refined and real-time assessment of equipment operating efficiency. It abandons the traditional coarse algorithm that averages the entire operating time, instead capturing the true energy efficiency performance of the equipment at every instant during stable operation. This fine-grained analysis can reveal minute fluctuations and trend changes in efficiency, thus enabling the keen detection of early signs of equipment performance degradation before an overall decline in efficiency occurs.

[0118] It should be noted that existing technologies typically employ a macroscopic and lagging approach when calculating equipment operating efficiency: that is, after a relatively long reporting period, the cumulative output and cumulative input (such as total power generation and total coal consumption) within that period are aggregated for overall efficiency calculation. This approach cannot distinguish between steady-state and transient operating conditions, conflating the low efficiency of dynamic processes with the true efficiency of steady-state operation. Furthermore, the results are severely lagging and cannot be used for real-time monitoring and early warning. The fundamental difference between the steps described in this invention and existing technologies lies in that they are strictly limited to the identified steady-state operating condition time period and real-time efficiency calculations are performed based on instantaneous data points. This difference allows efficiency assessment to truly reflect the performance level of the equipment under steady-state conditions, eliminating interference from transient processes such as start-up, shutdown, and load changes, and elevating efficiency monitoring from post-event statistics to the level of real-time perception.

[0119] The instantaneous efficiency values ​​are averaged to obtain the current operating efficiency.

[0120] Furthermore, all instantaneous efficiency values ​​are summed, and the sum is divided by the total number of instantaneous efficiency values. The quotient is the current operating efficiency value of the equipment during the steady-state operating period.

[0121] The current operating efficiency is compared with the historical baseline efficiency of the equipment's health, and a performance deviation warning signal is generated based on the comparison results.

[0122] The system retrieves the historical baseline efficiency of the equipment's health from the historical database, obtains the absolute difference between the current operating efficiency and the historical baseline efficiency of the equipment's health, and thus obtains the efficiency deviation value.

[0123] Furthermore, from the historical database storing long-term operating data of the device in a healthy state, the efficiency benchmark value obtained from the statistics of the device in a healthy state is queried and retrieved as the historical benchmark efficiency of the device health. The current operating efficiency value is subtracted from the historical benchmark efficiency value of the device health, and the absolute value of the result is taken to obtain the efficiency deviation value.

[0124] Based on the normal fluctuation range of efficiency obtained from long-term operation of equipment in a healthy state, a deviation threshold is set. The efficiency deviation value is compared with the deviation threshold. When the efficiency deviation value exceeds the deviation threshold, a performance deviation warning signal is generated.

[0125] Furthermore, the system analyzes a large number of efficiency values ​​recorded by the equipment during long-term operation in a healthy state in the historical database, and statistically analyzes the normal fluctuation range of efficiency values. The system sets the deviation threshold by subtracting twice the standard deviation from the historical baseline efficiency of the equipment. The system compares the efficiency deviation value with the deviation threshold. If the efficiency deviation value is greater than the deviation threshold, the system determines that the equipment performance has deteriorated and generates a performance deviation warning signal, which includes the equipment identifier of the deviation, the current operating efficiency value, the historical baseline efficiency value of the equipment, and the efficiency deviation value information.

[0126] The causal network graph module receives deviation warning signals and, based on the constraints of the connection relationships between nodes in the equipment topology graph, runs a constraint-based causal discovery algorithm on normal operating data to automatically learn and construct a causal network graph describing the causal relationships between process parameters.

[0127] The received deviation warning signal is based on the constraints of the connection relationship between nodes in the equipment topology diagram. The constraint-based causal discovery algorithm is run on the normal operation data to automatically learn and construct a causal network diagram describing the causal relationship between process parameters.

[0128] Based on the abnormal device identifier in the performance deviation warning signal, starting from the abnormal device identifier, the upstream and downstream devices of the neighboring devices are found in the device topology map to form a local device set.

[0129] Furthermore, the performance deviation warning signal includes the identifier of the abnormal device; starting from the abnormal device identifier, all upstream and downstream device nodes that are directly connected to the abnormal device identifier through directed edges are found in the device topology graph; the upstream device nodes, downstream device nodes, and the abnormal device identifier itself together constitute a local device set.

[0130] Extract time series data of process parameters of all devices in a local device set under recent normal operating conditions from the historical database to form a target data subset.

[0131] Furthermore, the system queries the historical database for time-series data of all process parameters of each device within the local device set that were marked as operating under normal conditions over a recent continuous period; these data are then extracted and integrated to form the target data subset.

[0132] Based on the connection relationships of nodes in the device topology graph, a set of constraints for conditional independence and causal direction prior is generated.

[0133] Furthermore, the connection relationships between nodes in the equipment topology graph are examined. For two nodes in the equipment topology graph that are not directly connected by a directed edge, a conditional independence constraint is generated, indicating that the parameters corresponding to these two nodes should be independent of each other given the parameters of their common neighbor nodes. Based on the process flow direction indicated by the directed edge in the equipment topology graph, a mandatory causal direction prior is added to the corresponding node pair. All generated conditional independence constraints and causal direction priors together constitute a constraint set.

[0134] The PC causal discovery algorithm is run on a subset of the target data. A fully connected undirected graph is initialized on the subset of the target data. Based on the conditional independence constraints in the constraint set, the conditional independence of the edges in the undirected graph is checked to obtain a partially directed acyclic graph.

[0135] Furthermore, the PC causal discovery algorithm is run on a subset of the target data. At the beginning of the algorithm, a fully connected undirected graph is initialized, where each node represents a process parameter in a local device set. Then, based on the conditional independence constraints defined in the constraint set, conditional independence is tested on each edge in the undirected graph. If the test result shows that two nodes are independent under a certain condition set, the edge connecting these two nodes in the undirected graph is deleted. This process is repeated until no more edges can be deleted, resulting in a partially directed acyclic graph.

[0136] By orienting edges in a portion of a directed acyclic graph, a causal network graph describing the causal relationships between process parameters is generated.

[0137] Furthermore, in partially directed acyclic graphs, V-shaped structures (i.e., two non-adjacent nodes pointing to an adjacent node) are identified, and the direction of the edges is determined according to the rules of V-shaped structures. For the remaining edges with undetermined directions, the causal direction priors defined in the constraint condition set are applied to determine their directions. Finally, a complete directed acyclic graph is obtained, which is a causal network graph describing the causal relationship between process parameters.

[0138] The structured diagnostic report module uses abnormal parameters corresponding to abnormal signals in the operation sequence as result nodes in the causal network diagram. It locates the root cause parameters through reverse path probability analysis and generates a structured diagnostic report.

[0139] In the causal network diagram, the abnormal parameters corresponding to the abnormal signals of the operation sequence are used as the result nodes. The root cause parameters are located through reverse path probability analysis, and a structured diagnostic report is generated.

[0140] In a causal network diagram of causal relationships, the abnormal parameters corresponding to the abnormal signals of the operation sequence are marked as result nodes.

[0141] Furthermore, in the causal network diagram describing the causal relationship between process parameters, find the node with the exact same name as the abnormal parameter recorded in the abnormal signal of the operation sequence, and explicitly mark the node as the result node of this root cause analysis.

[0142] Starting from the marked result node, traverse backwards along all directed edges in the causal network graph that point to this result node, and record each complete directed path from the result node to the result node.

[0143] Furthermore, starting from the marked result node, traverse backwards along all directed edges in the causal network graph that point to that result node, recording each complete path that connects to the result node from the outermost cause node through a series of directed edges.

[0144] For each directed path obtained through traversal, the conditional probability of each cause node causing the effect node to become abnormal is calculated using historical data based on Bayes' theorem. The conditional probabilities on the directed path are compared, and the cause node with the largest conditional probability value is determined as the root cause parameter.

[0145] The conditional probability expression for an anomaly is: ;

[0146] in, Cause node Under abnormal conditions, the result node The conditional probability of anomalies For the result node Under conditions where an anomaly has already occurred, the cause node The conditional probability of an anomaly occurring. For the result node The prior probability, Cause node The probability of evidence, As the cause node, This is the result node.

[0147] Furthermore, by calculating conditional probabilities based on Bayes' theorem and using these probabilities to locate root cause parameters, this method achieves a leap from qualitative correlation to quantitative causal inference in fault diagnosis. It transforms the complex connections in causal network diagrams into quantifiable probability values, thus providing objective and accurate data support for root cause identification. This not only effectively avoids the subjectivity and ambiguity that may arise from rule-based matching in traditional expert systems but also overcomes the vulnerability of simple correlation analysis to spurious correlations. By comparing the conditional probability values ​​on different paths, the most likely root cause of the anomaly can be automatically and reliably identified, improving the accuracy and credibility of diagnostic conclusions.

[0148] It should be noted that existing technologies for fault diagnosis typically employ rule-based reasoning or fault tree analysis, heavily relying on pre-defined rules or tree structures. Their diagnostic logic is static and fixed, making it difficult to adapt to dynamic changes and complex coupling relationships, and they cannot provide probabilistic confidence measurements of the diagnostic results. The fundamental difference between the steps described in this invention and existing technologies lies in abandoning the static rule base and instead dynamically calculating causal probabilities based on historical data. It utilizes Bayes' theorem to combine the causal topology provided by the causal network diagram with actual operational data, calculating the conditional probability of an abnormal result node given an abnormal cause node. This transforms the diagnostic process from experience-based logical matching to data-driven probabilistic reasoning, achieving adaptability of the diagnostic method and measurability of the conclusions. This represents a significant evolution in fault diagnosis technology from experience-driven to data-driven approaches.

[0149] By integrating root cause parameters, directed paths, and conditional probabilities, a structured diagnostic report is generated.

[0150] Furthermore, the root cause parameters, the directed path, and the conditional probability values ​​obtained from the determination are integrated and filled into the report template to generate a structured diagnostic report that includes the abnormal phenomenon, the root cause, the chain of evidence, and the confidence level.

[0151] In summary, this invention achieves a closed-loop process from data acquisition to intelligent diagnosis by constructing a device topology diagram module, a morphological difference module, an operation sequence anomaly signal module, a comparison module, a causal network diagram module, and a structured diagnostic report module. It standardizes real-time operating parameters, identifies steady-state conditions, and constructs a topology diagram reflecting device connectivity. Furthermore, it analyzes the morphological characteristics of transient operation sequences such as load regulation using dynamic time warping algorithms and one-dimensional convolutional neural networks, quantifies health status, and triggers anomaly signals. It provides efficiency deviation warnings during the steady-state phase. Based on device topology constraints and causal discovery algorithms, it locates root cause parameters and generates a structured diagnostic report. This deep integration of time-series morphological analysis, causal reasoning, and device topology solves the pain points of traditional monitoring methods, such as insufficient early anomaly detection and difficulty in root cause localization during transient processes. It achieves a leap from passive alarm to proactive prediction, and from phenomenon description to essential diagnosis, improving the accuracy and foresight of power plant equipment status monitoring.

[0152] 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. An equipment status monitoring system based on real-time power plant production data, characterized in that: This includes a module for constructing equipment topology diagrams, collecting and preprocessing real-time operating parameters, generating standardized real-time data sequences, identifying steady-state operating time periods of equipment based on standardized real-time data sequences, and constructing equipment topology diagrams based on power plant process diagrams; The morphological difference module, when a load adjustment operation is detected, extracts the sequence of changes in operation command process parameters from the standardized real-time data sequence to obtain the operation sequence to be evaluated, calls the baseline operation sequence recorded by the equipment in a healthy state, and uses a dynamic time warping algorithm to calculate the morphological difference between the operation sequence to be evaluated and the baseline operation sequence. The operation sequence abnormal signal module inputs the aligned operation sequence into a one-dimensional convolutional neural network, extracts the local morphological features of the sequence in the time dimension through the convolutional kernel, and outputs a quantified operation health score through a fully connected layer. The operation health score is evaluated, and an operation sequence abnormal signal is triggered based on the evaluation result. The comparison module calculates the current operating efficiency of the equipment based on standardized real-time data sequences during the steady-state operating period of the equipment, compares the current operating efficiency with the historical baseline efficiency of the equipment's health, and generates a performance deviation warning signal based on the comparison result. The causal network graph module receives deviation warning signals and, based on the constraints of the connection relationships between nodes in the equipment topology graph, runs a constraint-based causal discovery algorithm on normal operating data to automatically learn and construct a causal network graph describing the causal relationships between process parameters. The structured diagnostic report module uses abnormal parameters corresponding to abnormal signals in the operation sequence as result nodes in the causal network diagram. It locates the root cause parameters through reverse path probability analysis and generates a structured diagnostic report.

2. The equipment status monitoring system based on real-time power plant production data as described in claim 1, characterized in that: Collect and preprocess real-time operating parameters to generate standardized real-time data sequences, including the following steps: Real-time operating parameters are collected from distributed control units and plant-level monitoring information units, and null values ​​are removed from the collected real-time operating parameters. The real-time operating parameters that have completed null value removal are subjected to outlier correction processing, and the real-time operating parameters that have completed outlier correction are subjected to timestamp alignment processing to generate a standardized real-time data sequence.

3. The equipment status monitoring system based on real-time power plant production data as described in claim 2, characterized in that: Based on the standardized real-time data sequence identification equipment's steady-state operating time period, and according to the power plant process diagram, an equipment topology diagram is constructed, including the following steps: The first-order difference method is used to calculate the rate of change of power signal in a continuous time window in a standardized real-time data sequence. Based on the statistical analysis results of the rate of change of power signal in a known stable operation history of the unit, a steady-state judgment threshold is set. The rate of change of power signal is compared with the steady-state judgment threshold to obtain the candidate steady-state operating condition time period. The volatility within the candidate steady-state operating condition time period is verified to obtain the final steady-state operating condition time period; The process diagram of the power plant is analyzed to identify the pumps, fans, and heat exchangers shown in the diagram as nodes in the equipment topology diagram. The process pipelines are also analyzed to identify the directed edges in the equipment topology diagram. Based on the connection relationship between the equipment and pipelines in the power plant process diagram, the nodes and directed edges are associated to construct the equipment topology diagram.

4. The equipment status monitoring system based on real-time power plant production data as described in claim 3, characterized in that: When a load regulation operation is detected, the sequence of changes in the operation command process parameters is extracted from the standardized real-time data sequence to obtain the operation sequence to be evaluated. The baseline operation sequence recorded by the equipment in a healthy state is then invoked, including the following steps: By monitoring the load setpoint commands issued by the distributed control unit, the start and end times of the load adjustment operation can be identified. Based on the start and end times of the identified load regulation operations, process parameter data related to the operation are extracted from the standardized real-time data sequence within the corresponding time period to form a process parameter change sequence. The extracted sequence of process parameter changes is defined as the sequence of operations to be analyzed and evaluated in this study. Based on the type and magnitude of the current load regulation operation, the baseline operation sequence is obtained by retrieving and calling the standard process parameter change sequence recorded under the same operation type and magnitude from the database that stores historical operating data of the equipment in its healthy state.

5. The equipment status monitoring system based on real-time power plant production data as described in claim 4, characterized in that: The dynamic time warping algorithm is used to calculate the morphological difference between the operation sequence to be evaluated and the benchmark operation sequence, including the following steps: A dynamic time warping algorithm is used to non-linearly align the operation sequence to be evaluated with the benchmark operation sequence on the time axis to construct a cumulative distance matrix; In the cumulative distance matrix, find the path with the minimum cumulative distance from the starting point to the ending point, and use the point-by-point accumulation method to obtain the sum of the Euclidean distances of the corresponding data points on the optimal curved path, thus obtaining the preliminary sequence morphological difference degree.

6. The equipment status monitoring system based on real-time power plant production data as described in claim 5, characterized in that: The aligned operation sequence is input into a one-dimensional convolutional neural network. Local morphological features of the sequence in the time dimension are extracted through convolutional kernels, and a quantified operation health score is output through a fully connected layer. The operation health score is evaluated, and an operation sequence abnormality signal is triggered based on the evaluation result, including the following steps: Subtract the mean of the process parameter in the baseline operation sequence from the data of each process parameter in the aligned operation sequence, and then divide by the standard deviation to generate the normalized aligned operation sequence. The normalized alignment operation sequence is fed into a pre-trained one-dimensional convolutional neural network. The input layer dimension of the one-dimensional convolutional neural network is matched with the time step and feature dimension of the normalized alignment operation sequence. The first convolutional layer of a one-dimensional convolutional neural network uses a convolutional kernel to slide across the time dimension of the normalized alignment operation sequence to extract local morphological features from the sequence and output a feature map. By inputting the normalized alignment operation sequence into a one-dimensional convolutional neural network, and processing it layer by layer through convolutional and pooling layers, a feature vector is obtained. The feature vector is then input into a fully connected layer with a sigmoid activation function, and the fully connected layer outputs a scalar value. The mean and standard deviation of the health score for normal operations in recent history are calculated using the sliding window method. The mean minus twice the standard deviation is set as the dynamic threshold. The current operation health score is compared with the dynamic threshold. When the operation health score falls below the dynamic threshold, an operation sequence abnormality signal is triggered.

7. The equipment status monitoring system based on real-time power plant production data as described in claim 6, characterized in that: During the period when the equipment is in steady-state operation, the current operating efficiency of the equipment is calculated based on standardized real-time data sequences, including the following steps: Based on the start and end timestamps of the steady-state operating period, process parameter data points are extracted from the standardized real-time data sequence; Substitute the process parameter data points into the thermodynamic efficiency calculation formula to calculate the instantaneous efficiency value.

8. The equipment status monitoring system based on real-time power plant production data as described in claim 7, characterized in that: The current operating efficiency is compared with the historical baseline efficiency of the equipment's health, and a performance deviation warning signal is generated based on the comparison results, including the following steps: The system retrieves the historical baseline efficiency of the equipment health from the historical database, obtains the absolute difference between the current operating efficiency and the historical baseline efficiency of the equipment health, and thus obtains the efficiency deviation value. Based on the normal fluctuation range of efficiency obtained from long-term operation of equipment in a healthy state, a deviation threshold is set. The efficiency deviation value is compared with the deviation threshold. When the efficiency deviation value exceeds the deviation threshold, a performance deviation warning signal is generated.

9. The equipment status monitoring system based on real-time power plant production data as described in claim 8, characterized in that: Receiving deviation warning signals involves constraining the connections between nodes based on the equipment topology diagram, running a constraint-based causal discovery algorithm on normal operating data, automatically learning and constructing a causal network diagram describing the causal relationships between process parameters, including the following steps: Based on the abnormal device identifier in the performance deviation warning signal, starting from the abnormal device identifier, the upstream and downstream devices of the neighboring devices are found in the device topology map to form a local device set; Extract time series data of process parameters of all devices in a local device set under recent normal operating conditions from the historical database to form a target data subset; Based on the connection relationships of nodes in the device topology graph, a set of constraint conditions for conditional independence and causal direction prior is generated. The PC causal discovery algorithm is run on a subset of the target data. A fully connected undirected graph is initialized on the subset of the target data. Based on the conditional independence constraints in the constraint set, the conditional independence of the edges in the undirected graph is checked to obtain a partially directed acyclic graph. By orienting edges in a portion of a directed acyclic graph, a causal network graph describing the causal relationships between process parameters is generated.

10. The equipment status monitoring system based on real-time power plant production data as described in claim 9, characterized in that: In the causal network diagram, the abnormal parameters corresponding to the abnormal signals of the operation sequence are taken as the result nodes. The root cause parameters are located through reverse path probability analysis, and a structured diagnostic report is generated, including the following steps: In the causal network diagram of causal relationships, the abnormal parameters corresponding to the abnormal signals of the operation sequence are marked as result nodes; Starting from the marked fruit node, traverse backwards along all directed edges in the causal network graph that point to this fruit node, and record each complete directed path from the cause node to the fruit node. For each directed path obtained through traversal, based on Bayes' theorem, historical data is used to calculate the conditional probability that each cause node on the path causes the effect node to be abnormal. The conditional probabilities on the directed path are compared, and the cause node with the largest conditional probability value is determined as the root cause parameter. By integrating root cause parameters, directed paths, and conditional probabilities, a structured diagnostic report is generated.

Images