Power plant heat exchanger early warning method, system and device based on time series modeling

Abstract

This disclosure provides a method, system, and equipment for early warning of power plant heat exchangers based on time series modeling, relating to the field of machine learning technology. The method includes: collecting historical multi-dimensional operating parameters of the power plant heat exchanger under healthy operating conditions to construct a time series set of historical parameters; performing time series extrapolation and residual analysis on the historical parameter time series set to obtain a historical residual time series of the power plant heat exchanger; vectorizing the time series trend of the historical residual time series to obtain a historical residual trend vector of the power plant heat exchanger; embedding the historical residual trend vector into the manifold space to obtain a reference manifold structure of the residual evolution trend under healthy conditions of the power plant heat exchanger; in the real-time monitoring stage, mapping the real-time operating parameter space of the power plant heat exchanger to the reference manifold structure to obtain the real-time mapped coordinates of the power plant heat exchanger; determining the degree of local alienation of the real-time mapped coordinates on the reference manifold structure to obtain an early warning signal of heat exchanger performance degradation, thereby improving the early warning efficiency of the heat exchanger.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a method, system and equipment for early warning of power plant heat exchangers based on time series modeling. Background Technology

[0002] As the core equipment in heat exchange systems in industries such as power and chemical engineering, the continuous and stable operation of heat exchangers is a crucial prerequisite for ensuring the safe and efficient operation of industrial production systems. With the popularization of the Industrial Internet of Things (IIoT) and online monitoring technologies, heat exchangers can collect multi-dimensional physical parameters such as temperature, pressure, flow rate, and heat exchange efficiency in real time during operation, forming massive amounts of high-dimensional time-series operational data. Currently, machine learning technology has been widely applied in the field of industrial equipment condition monitoring and performance evaluation, becoming a core technical support for data-driven equipment management. Machine learning-based analysis methods can achieve intelligent judgment of equipment operating status by extracting features, training models, and identifying states from historical operational time-series data.

[0003] Existing heat exchanger operating data often contains a large amount of non-steady-state operating condition information. Directly using the full historical data to build a benchmark model will introduce fluctuations and disturbances unrelated to the healthy steady state. This will cause the established benchmark model to fail to accurately depict the inherent operating patterns of the equipment in a healthy state, thereby reducing the accuracy of subsequent degradation status assessment. Furthermore, the performance degradation of heat exchangers is a gradual process, and its evolution trajectory in the high-dimensional monitoring parameter space is complex. Traditional methods are unable to effectively extract low-dimensional manifold features that can stably characterize the degradation trend of the equipment from massive, high-dimensional, and time-related operating data. This results in insufficient quantification of the difference between the real-time operating status and the healthy benchmark model, affecting the timeliness and reliability of degradation early warning. Summary of the Invention

[0004] This disclosure provides a method, system, and equipment for early warning of power plant heat exchangers based on time series modeling, in order to at least solve the above-mentioned technical problems existing in the prior art.

[0005] According to a first aspect of this disclosure, a power plant heat exchanger early warning method based on time series modeling is provided, comprising: S1: Collect historical multi-dimensional operating parameters of the power plant heat exchanger under healthy operating conditions to construct a time series set of historical parameters of the power plant heat exchanger; S2: Perform time series extrapolation and residual analysis on the historical parameter time series set to obtain the historical residual time series of the power plant heat exchanger; S3: Perform time-series trend vectorization on the historical residual time series to obtain the historical residual trend vector of the power plant heat exchanger; S4: Embed the historical residual trend vector into the manifold space to obtain the baseline manifold structure of the residual evolution trend of the power plant heat exchanger under healthy conditions. S5: During the real-time monitoring phase, the real-time operating parameters of the power plant heat exchanger are spatially mapped to the reference manifold structure to obtain the real-time mapped coordinates of the power plant heat exchanger. S6: Determine the outlier degree of the local alienation of the real-time mapped coordinates on the reference manifold structure to obtain the performance degradation early warning signal of the power plant heat exchanger.

[0006] In one possible implementation, S1 specifically includes: S101: Extract multi-dimensional measurement parameters of power plant heat exchangers under healthy operating conditions from the historical operation database of power plant heat exchangers; S102: Determine the steady-state operating condition of the multi-dimensional measuring point parameters to obtain the steady-state operating condition segment parameters of the power plant heat exchanger; S103: Time-domain alignment of steady-state operating condition segment parameters to obtain the synchronization time sequence of the power plant heat exchanger; S104: Merge and assemble the synchronous timing sequences to obtain the historical parameter timing set of the power plant heat exchanger.

[0007] In one possible implementation, S2 specifically includes: S201: Perform time-series dependency analysis on the parameter dimensions in the historical parameter time series set to obtain the autocorrelation evolution characteristics of the historical parameter time series set; S202: Based on autocorrelation evolution characteristics, the time-series evolution trajectory of the parameter dimension is trend-fitted to obtain the theoretical predicted value of the parameter dimension; S203: Based on theoretical predictions, the measured values ​​in the historical parameter time series set are compared hour by hour to obtain the degree of deviation between the measured values ​​and the theoretical predictions. S204: The numerical deviations are serialized and arranged in chronological order to obtain the historical residual time series of the power plant heat exchanger.

[0008] In one possible implementation, S3 specifically includes: S301: Perform sliding window segmentation on the historical residual time series to obtain local time series segments of the historical residual time series; S302: Extract trend features from local time segments to obtain the slope and amplitude of trend changes in local time segments; S303: Based on the slope of trend change and the amplitude of fluctuation, the evolution of the historical residual time series is mapped to obtain the instantaneous trend pointing vector of the historical residual time series. S304: Perform a time-series connection on the instantaneous trend pointing vector to obtain the historical residual trend vector of the power plant heat exchanger.

[0009] In one possible implementation, S4 specifically includes: S401: Perform local neighborhood detection on the historical residual trend vector to obtain the nearest neighbor index set of the historical residual trend vector; S402: Based on the nearest neighbor index set, perform local weight evaluation on the historical residual trend vector to obtain the local reconstruction weight of the historical residual trend vector; S403: Based on the local reconstruction weights, perform feature space mapping on the historical residual trend vector to obtain the low-dimensional manifold coordinates of the historical residual trend vector; S404: Tracing the evolution path of low-dimensional manifold coordinates to obtain the core evolution path of low-dimensional manifold coordinates; S405: Based on the core evolution path, the topological structure of the geometry of the low-dimensional manifold coordinates is characterized to obtain the reference manifold structure of the power plant heat exchanger.

[0010] In one possible implementation, S402 specifically includes: S4021: Calculate the spatial distance between the neighbor vectors indicated by the nearest neighbor index set and the historical residual trend vector to obtain the local neighborhood distance metric of the historical residual trend vector. S4022: Based on the local neighborhood distance metric, perform linear characterization analysis on the neighboring vectors to obtain the local reconstruction coefficients of the historical residual trend vector; S4023: Normalize and arrange the local reconstruction coefficients to obtain the local reconstruction weights of the historical residual trend vector.

[0011] In one possible implementation, S5 specifically includes: S501: Based on the residual evolution law corresponding to the benchmark manifold structure, the real-time multidimensional operating parameters of the power plant heat exchanger are extracted to obtain the real-time residual evolution time series of the power plant heat exchanger. S502: Based on the trend evolution law corresponding to the benchmark manifold structure, the real-time residual evolution time series is mapped with trend characteristics to obtain the real-time trend vector of the power plant heat exchanger. S503: Based on the embedding mapping rules of the reference manifold structure, the real-time trend vector is located in manifold space to obtain the real-time mapping coordinates of the power plant heat exchanger.

[0012] In one possible implementation, S6 specifically includes: S601: Based on the baseline manifold structure, the neighborhood range of the real-time mapped coordinates is defined to obtain the local neighborhood set of the real-time mapped coordinates; S602: Perform a separation evaluation between the reference coordinates within the local neighborhood set and the real-time mapped coordinates to obtain the local separation amplitude of the real-time mapped coordinates; S603: Based on the residual evolution characteristics of the power plant heat exchanger under healthy operating conditions, a threshold boundary is set for the local alienation amplitude to obtain the alienation judgment standard of the power plant heat exchanger. S604: Match and verify the local alienation amplitude with the alienation determination criteria to obtain the outlier status indication of the real-time mapped coordinates; S605: Based on the outlier status indication, when the local outlier amplitude exceeds the outlier determination standard, the power plant heat exchanger degradation warning is activated to generate a power plant heat exchanger performance degradation warning signal.

[0013] According to a second aspect of this disclosure, a power plant heat exchanger early warning system based on time series modeling is provided, comprising: The health time series parameter construction module is used to collect historical multi-dimensional operating parameters of the power plant heat exchanger under healthy operating conditions in order to construct a historical parameter time series set of the power plant heat exchanger; The historical residual sequence parsing module is used to perform time series extrapolation and residual parsing on the historical parameter time series set to obtain the historical residual time series of the power plant heat exchanger. The residual trend vectorization module is used to perform time-series trend vectorization on the historical residual time series to obtain the historical residual trend vector of the power plant heat exchanger. The benchmark manifold structure construction module is used to embed the historical residual trend vector into the manifold space to obtain the benchmark manifold structure of the residual evolution trend of the power plant heat exchanger under healthy conditions. The real-time manifold coordinate mapping module is used to map the real-time operating parameter space of the power plant heat exchanger to the reference manifold structure during the real-time monitoring phase, so as to obtain the real-time mapped coordinates of the power plant heat exchanger. The real-time coordinate manifold analysis module is used to determine the degree of local alienation of real-time mapped coordinates on the reference manifold structure, so as to obtain the performance degradation early warning signal of the power plant heat exchanger.

[0014] In one possible implementation, the health time series parameter construction module is specifically used for: Extract multi-dimensional measurement parameters of power plant heat exchangers under healthy operating conditions from the historical operation database of power plant heat exchangers; By performing steady-state condition discrimination on the parameters of multi-dimensional measuring points, the steady-state condition segment parameters of the power plant heat exchanger are obtained. By aligning the parameters of the steady-state operating condition segment in the time domain, the synchronous timing sequence of the power plant heat exchanger is obtained. The synchronous time sequence is fused and assembled to obtain the historical parameter time sequence set of the power plant heat exchanger.

[0015] In one possible implementation, the historical residual sequence parsing module is specifically used for: The time-series dependency characteristics of the parameter dimensions in the historical parameter time series set are analyzed to obtain the autocorrelation evolution characteristics of the historical parameter time series set; Based on the autocorrelation evolution characteristics, the time-series evolution trajectory of the parameter dimension is trend-fitted to obtain the theoretical predicted value of the parameter dimension; Based on theoretical predictions, the measured values ​​in the historical parameter time series are compared hour by hour to obtain the degree of deviation between the measured values ​​and the theoretical predictions. The numerical deviations are serialized and arranged in chronological order to obtain the historical residual time series of the power plant heat exchanger.

[0016] In one possible implementation, the residual trend vectorization module is specifically used for: Sliding window segmentation is performed on the historical residual time series to obtain local time series segments of the historical residual time series; Trend features are extracted from local time segments to obtain the slope and amplitude of trend changes in the local time segments; Based on the slope of trend change and the amplitude of fluctuation, the evolution of the historical residual time series is mapped to obtain the instantaneous trend pointing vector of the historical residual time series. By performing a time-series concatenation on the instantaneous trend pointing vector, the historical residual trend vector of the power plant heat exchanger is obtained.

[0017] In one possible implementation, the baseline manifold structure construction module is specifically used for: Local neighborhood detection is performed on the historical residual trend vector to obtain the nearest neighbor index set of the historical residual trend vector; Based on the nearest neighbor index set, local weight evaluation is performed on the historical residual trend vector to obtain the local reconstruction weight of the historical residual trend vector; Based on the local reconstruction weights, the historical residual trend vector is mapped to the feature space to obtain the low-dimensional manifold coordinates of the historical residual trend vector; By tracing the evolution path of low-dimensional manifold coordinates, the core evolution path of low-dimensional manifold coordinates can be obtained; Based on the core evolution path, the topological structure of the low-dimensional manifold coordinates is characterized to obtain the reference manifold structure of the power plant heat exchanger.

[0018] In one possible implementation, the baseline manifold structure construction module is further used for: Spatial distance is calculated between the neighbor vectors indicated by the nearest neighbor index set and the historical residual trend vector to obtain the local neighborhood distance metric of the historical residual trend vector. Based on the local neighborhood distance metric, the neighboring vectors are linearly represented and analyzed to obtain the local reconstruction coefficients of the historical residual trend vector; The local reconstruction coefficients are normalized and arranged to obtain the local reconstruction weights of the historical residual trend vector.

[0019] In one possible implementation, the real-time manifold coordinate mapping module is specifically used for: Based on the residual evolution law corresponding to the benchmark manifold structure, residual analysis is performed on the real-time multidimensional operating parameters of the power plant heat exchanger to obtain the real-time residual evolution time series of the power plant heat exchanger. Based on the trend evolution law corresponding to the baseline manifold structure, the real-time residual evolution time series is mapped with trend features to obtain the real-time trend vector of the power plant heat exchanger. Based on the embedding mapping rules of the reference manifold structure, the real-time trend vector is located in manifold space to obtain the real-time mapped coordinates of the power plant heat exchanger.

[0020] In one possible implementation, the real-time coordinate manifold analysis module is specifically used for: Based on the baseline manifold structure, the neighborhood range of the real-time mapped coordinates is defined to obtain the local neighborhood set of the real-time mapped coordinates; The local distance between the reference coordinates in the local neighborhood set and the real-time mapped coordinates is evaluated to obtain the local distance magnitude of the real-time mapped coordinates. Based on the residual evolution characteristics of power plant heat exchangers under healthy operating conditions, threshold boundaries are set for local detachment amplitudes to obtain the detachment judgment criteria for power plant heat exchangers. By matching and verifying the local alienation amplitude with the alienation degree judgment criteria, an outlier status indication of the real-time mapped coordinates is obtained. According to the outlier status indication, when the local outlier amplitude exceeds the outlier determination standard, the power plant heat exchanger degradation early warning is activated to generate a performance degradation early warning signal for the power plant heat exchanger.

[0021] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, which are executed by at least one processor to enable the at least one processor to perform the method of this disclosure.

[0022] Compared with existing technologies, the power plant heat exchanger early warning system based on time series modeling disclosed herein has the following advantages: By collecting multi-dimensional operating parameters under the health condition of the heat exchanger, the system sequentially performs stable operating condition judgment to eliminate drastically fluctuating non-steady-state data, time domain alignment to unify the time base of each parameter, and fusion and compilation to form a structured time series set. Finally, a pure and synchronized historical parameter time series set is constructed, thereby accurately screening effective raw data, eliminating non-steady-state interference, providing a reliable data source for subsequent analysis, and ensuring the accuracy of the benchmark manifold structure construction.

[0023] By analyzing the time-series data of historical parameters to obtain autocorrelation evolution characteristics, theoretical predictions for each parameter dimension are derived and compared with measured values ​​to obtain residuals. The residual sequence is then transformed into an instantaneous trend pointing vector through sliding window segmentation, trend feature extraction, and situation mapping, and is temporally linked to form a historical residual trend vector. Subsequently, through local neighborhood detection, reconstructed weight evaluation, and feature space mapping, the high-dimensional trend vector is embedded into a low-dimensional manifold and the core evolution path is traced. Finally, a benchmark manifold structure representing the residual evolution law under healthy conditions is constructed, thereby effectively extracting low-dimensional degradation features from high-dimensional data and achieving accurate quantification of the difference between the equipment operating status and the health benchmark. Attached Figure Description

[0024] The above and other objects, features, and advantages of this disclosure will become readily apparent from the following detailed description of exemplary embodiments, taken in conjunction with the accompanying drawings. Several embodiments of this disclosure are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0025] Figure 1 A flowchart illustrating the workflow of the power plant heat exchanger early warning method based on time series modeling according to an embodiment of this disclosure is shown. Figure 2 A functional block diagram of a power plant heat exchanger early warning system based on time series modeling, according to an embodiment of this disclosure, is shown. Figure 3 A schematic diagram of the composition structure of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0026] To make the objectives, features, and advantages of this disclosure more apparent and understandable, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0027] Reference Figure 1 The diagram shown is a flowchart illustrating a power plant heat exchanger early warning method based on time series modeling according to an embodiment of the present invention. In this embodiment, the power plant heat exchanger early warning method based on time series modeling includes: S1: Collect historical multi-dimensional operating parameters of the power plant heat exchanger under healthy operating conditions to construct a time series set of historical parameters of the power plant heat exchanger.

[0028] In one possible implementation, S1 specifically includes sub-steps S101 to S104: S101: Extract multi-dimensional measurement parameters of the power plant heat exchanger under healthy operating conditions from the historical operation database of the power plant heat exchanger.

[0029] S102: Determine the steady-state operating condition of the multi-dimensional measuring point parameters to obtain the steady-state operating condition segment parameters of the power plant heat exchanger.

[0030] S103: Time-domain alignment of steady-state operating condition segment parameters to obtain the synchronization time sequence of the power plant heat exchanger.

[0031] S104: Merge and assemble the synchronous timing sequences to obtain the historical parameter timing set of the power plant heat exchanger.

[0032] Multidimensional measurement parameters of the heat exchanger under healthy operating conditions are extracted from the historical operating database of the heat exchanger. The historical operating database is a complete data set formed by continuously collecting and storing various operating data during the daily operation of the heat exchanger. The criteria for judging the healthy operating state are that all operating indicators of the heat exchanger are within the rated normal range specified by the equipment manufacturer and there are no fault alarm information or performance degradation related records. The multidimensional measurement parameters are physical quantity data directly related to the operating state of the heat exchanger, covering data such as temperature, pressure, flow rate, and heat exchange efficiency at different monitoring points of the heat exchanger. During extraction, the full operating data of all monitoring points of the corresponding heat exchanger are first matched according to the unique equipment identifier of the heat exchanger. Then, the data of all monitoring points during the entire healthy operating period of the heat exchanger are filtered according to the timestamp. The final product is the multidimensional measurement parameters of the heat exchanger under healthy operating conditions. These parameters are the original data sequence of different monitoring points changing continuously over time during the healthy operating period.

[0033] Stable operating conditions are determined by analyzing the multi-dimensional measuring point parameters to obtain the steady-state operating condition segment parameters of the heat exchanger. The criterion for stable operating conditions is that the numerical variation amplitude of each physical quantity in the multi-dimensional measuring point parameters is controlled within the rated fluctuation threshold within a continuous fixed time window. This threshold is set according to the equipment performance indicators of the heat exchanger. Specifically, the difference between the maximum and minimum values ​​of each physical quantity within the corresponding time window does not exceed a fixed proportion of the rated value of that physical quantity. During the determination, the multi-dimensional measuring point parameters are continuously slid-trimmed according to the above-mentioned fixed time window. The numerical variation amplitude of each physical quantity within each trimmed time window is calculated, and the variation amplitude of each physical quantity is checked one by one to verify whether it meets the criteria for stable operating conditions. All multi-dimensional measuring point parameter segments corresponding to the time windows that meet the criteria are extracted and retained, while parameter segments that do not meet the criteria are directly discarded. The final product is the steady-state operating condition segment parameters of the heat exchanger. These parameters are the operating data segments of each monitoring point when the heat exchanger is in a stable operating condition under healthy operating conditions. Each segment is continuous and numerically stable multi-dimensional data within the corresponding time window.

[0034] Time-domain alignment of steady-state operating condition segment parameters yields the synchronous timing sequence of the heat exchanger. The core of time-domain alignment is to uniformly calibrate the time axes of all steady-state operating condition segment parameters, ensuring that the start and time intervals of each parameter segment are completely consistent. During operation, the shortest time length among all steady-state operating condition segment parameters is first selected as a unified time length benchmark. Steady-state operating condition segment parameters exceeding this benchmark length are then truncated at equal time intervals, ensuring that the time length of each truncated parameter segment is completely consistent with the benchmark. Finally, the original timestamps of all steady-state operating condition segment parameters are recalibrated. A unified time axis is constructed using fixed time intervals. Each data point of each parameter segment is precisely matched to the corresponding time node on the unified time axis. If data is missing at a certain time node, it is supplemented according to the changing trend of adjacent data points within that parameter segment. The supplemented values ​​are highly consistent with the changing patterns of adjacent data points. The final product is the synchronous time sequence of the heat exchanger. This sequence is a continuous multi-dimensional operating data sequence of each monitoring point on the unified time axis after all steady-state operating condition segment parameters have been uniformly calibrated on the time axis. All data points achieve precise synchronous correspondence in the time domain.

[0035] Synchronous time series sequences are fused and compiled to obtain the historical parameter time series set of the heat exchanger. The fusion and compilation process involves classifying and integrating all time-domain aligned synchronous time series sequences according to the category of monitoring points. First, the data of the same monitoring points in each synchronous time series are continuously spliced ​​according to the order of timestamps to ensure the temporal continuity of the spliced ​​data. Then, the spliced ​​data sequences of different monitoring points are integrated as a whole to construct a multi-dimensional data set with time as the horizontal axis and the physical quantities of each monitoring point as the vertical axis. During the integration process, the integrity of the data is fully verified, and redundant data corresponding to duplicate timestamps are directly removed to ensure that there is unique and complete data for each monitoring point under each timestamp. The final product is the historical parameter time series set of the heat exchanger. This time series set is the multi-dimensional operating data of all monitoring points under the healthy operating conditions and stable operating conditions of the heat exchanger. It is presented in the form of a time series and includes all processed valid operating data. The data is uniformly integrated in both the time domain and the monitoring dimension.

[0036] The beneficial effects of this invention are as follows: it accurately selects effective raw data, strengthens the data foundation, improves the accuracy of subsequent processing, ensures the reliability of the data source for constructing the benchmark manifold structure, eliminates non-steady-state invalid data, reduces interference with residual analysis, improves the accuracy of generating historical residual time series, achieves unified calibration of parameters in the time domain, eliminates time domain misalignment, ensures data time synchronization, improves the accuracy of residual trend vector extraction, integrates into a unified structured time series set, provides standardized data for time series extrapolation and residual analysis, improves modeling efficiency and accuracy, provides accurate data support for early warning, and improves overall early warning efficiency.

[0037] S2: Perform time series extrapolation and residual analysis on the historical parameter time series set to obtain the historical residual time series of the power plant heat exchanger.

[0038] In one possible implementation, S2 specifically includes sub-steps S201 to S204: S201: Perform time-series dependency analysis on the parameter dimensions in the historical parameter time series set to obtain the autocorrelation evolution characteristics of the historical parameter time series set.

[0039] S202: Based on autocorrelation evolution characteristics, the time-series evolution trajectory of the parameter dimension is fitted with a trend to obtain the theoretical predicted value of the parameter dimension.

[0040] S203: Based on theoretical predictions, the measured values ​​in the historical parameter time series set are compared hour by hour to obtain the degree of deviation between the measured values ​​and the theoretical predictions.

[0041] S204: The numerical deviations are serialized and arranged in chronological order to obtain the historical residual time series of the power plant heat exchanger.

[0042] The process of analyzing the time-series dependency characteristics of parameter dimensions in the historical parameter time series set is as follows: First, extract all time-series values ​​of each parameter dimension in the historical parameter time series set on the complete time axis. The historical parameter time series set is a set of synchronous time series sequences formed after steady-state condition discrimination, time-domain alignment, and fusion assembly under the healthy operating state of the heat exchanger. The parameter dimensions are various independent monitoring dimensions that characterize the operating state of the heat exchanger in this set. Then, starting from a time step of 1, the time step is increased sequentially, and the correlation degree between the current time node value of the parameter dimension and the corresponding historical time node value at each time step is calculated. Next, the correlation degree result corresponding to each time step is correlated with the time step. Then, continuous feature extraction is performed on the correlation degree results at each time step to extract the changing pattern and core features of the correlation degree of each parameter dimension at different time steps. The set of such changing patterns and core features of all parameter dimensions is the autocorrelation evolution feature of the historical parameter time series set.

[0043] The process of trend fitting of the temporal evolution trajectory of parameter dimensions based on autocorrelation evolution characteristics is as follows: First, the core time dependency relationship of the numerical change of each parameter dimension is determined according to the autocorrelation evolution characteristics of each parameter dimension. The autocorrelation evolution characteristics are the set of change rules and core features of the temporal dependency relationship of each parameter dimension at different time steps. The temporal evolution trajectory of the parameter dimension is the continuous trajectory formed by the numerical change of each parameter dimension on the historical time axis. Then, based on the existing time series values ​​of the parameter dimension, the values ​​of each subsequent time node are continuously extrapolated and calculated along the extension direction of the time axis according to the determined core time dependency relationship. During the extrapolation and calculation process, the change rules of the historical time series values ​​are kept consistent. The value obtained by extrapolation and calculation of each parameter dimension at each time node is the theoretical prediction value of the parameter dimension at the corresponding time node. The set of theoretical prediction values ​​of all time nodes arranged in chronological order is the theoretical prediction value of the parameter dimension.

[0044] The process of comparing measured values ​​in the historical parameter time series set with theoretical predictions time-by-time is as follows: First, for each parameter dimension, the theoretical prediction value and the measured value at the same time point are compared one-to-one. The theoretical prediction value is the set of values ​​for each parameter dimension obtained through extrapolation and calculation at each time point. The measured values ​​in the historical parameter time series set are the values ​​for each parameter dimension actually monitored at the corresponding time point under the healthy operation state of the heat exchanger. Then, the difference between the theoretical prediction value and the measured value at the same time point is calculated. Next, the absolute value of the calculated difference is processed. The result after processing is the degree of deviation of the parameter dimension at that time point. The set of the degree of deviation of each parameter dimension at all time points is the degree of deviation between the measured value and the theoretical prediction value.

[0045] The process of serializing the numerical deviation in chronological order is as follows: First, the numerical deviation of all parameter dimensions at the same time node is summarized to obtain the comprehensive numerical deviation result corresponding to each time node. The numerical deviation is the set of deviation results calculated for each parameter dimension at each time node. Then, the comprehensive numerical deviation results of all time nodes are arranged in chronological order to form a continuous numerical sequence distributed by time node. This sequence of comprehensive numerical deviation results arranged in chronological order is the historical residual time series of the heat exchanger.

[0046] The beneficial effects of this invention are as follows: it accurately mines the temporal correlation patterns of each parameter dimension, improves the accuracy of autocorrelation evolution characteristics, lays a reliable feature foundation for subsequent trend fitting, makes the theoretical prediction values ​​highly consistent with the actual temporal change trends of heat exchanger parameters, improves the accuracy of theoretical prediction values, provides an accurate reference benchmark for calculating the degree of numerical deviation, accurately reflects the actual difference between measured values ​​and theoretical prediction values, improves the objectivity and accuracy of the calculation of the degree of numerical deviation, provides accurate basic data for the construction of historical residual time series, and allows the historical residual time series to continuously and completely characterize the deviation changes of each parameter of the heat exchanger in the time dimension, improves the continuity and comprehensiveness of its characterization of residual evolution, and provides standardized and orderly basic time series data for subsequent residual trend vectorization.

[0047] S3: Perform time-series trend vectorization on the historical residual time series to obtain the historical residual trend vector of the power plant heat exchanger.

[0048] In one possible implementation, S3 specifically includes sub-steps S301 to S304: S301: Perform sliding window segmentation on the historical residual time series to obtain local time segments of the historical residual time series.

[0049] S302: Extract trend features from local time series segments to obtain the slope and amplitude of trend changes in the local time series segments.

[0050] S303: Based on the slope of trend change and the amplitude of fluctuation, the evolution of the historical residual time series is mapped to obtain the instantaneous trend pointing vector of the historical residual time series.

[0051] S304: Perform a time-series connection on the instantaneous trend pointing vector to obtain the historical residual trend vector of the power plant heat exchanger.

[0052] Historical residual time series is a continuous data sequence formed by arranging the deviations between measured values ​​and theoretical predicted values ​​under healthy operating conditions of a heat exchanger in chronological order. Sliding window segmentation involves setting a fixed-length window and a fixed sliding step size for this continuous sequence. Starting from the beginning time position of the historical residual time series, all residual data within the window coverage area are extracted. Then, the window is moved forward along the time axis with the set sliding step size. After each movement, all residual data within the window coverage area are extracted until the window moves to the end time position of the historical residual time series. Each independent residual data segment extracted by the window is a local time series segment. All the extracted independent residual data segments together constitute the local time series segments of the historical residual time series, and each local time series segment retains the temporal correlation of the residual data within the corresponding time range.

[0053] A local time series segment is an independent fragment of residual data within a time range obtained by segmenting a historical residual time series using a sliding window. Trend feature extraction, for each independent local time series segment, first traces the numerical change trajectory of the residual data over time, then calculates the slope of this trajectory. The residual values ​​corresponding to the first and last endpoints of the time dimension within the segment are selected. The numerical difference is obtained by subtracting the residual value at the beginning from the value at the end, and the time difference is obtained by subtracting the time value at the beginning from the value at the end. The slope of the trend change for this local time series segment is obtained by dividing the numerical difference by the time difference. A positive slope indicates an upward trend in the residual data within the local time series segment over time, a negative slope indicates a downward trend, and a slope of zero indicates... The table shows that the residual data within the local time series segment does not exhibit a significant trend change over time. Next, the fluctuation amplitude of this local time series segment is calculated. First, the average value of all residual data within the local time series segment is obtained. Then, the absolute value of the difference between each residual data point and the average value is calculated sequentially. Finally, the average of all absolute values ​​is calculated to obtain the fluctuation amplitude of the local time series segment. The magnitude of the fluctuation amplitude directly reflects the degree of dispersion of the residual data around the average value within the local time series segment; a larger value indicates a higher degree of dispersion, and a smaller value indicates a lower degree of dispersion. After completing the above slope and fluctuation amplitude calculations for all local time series segments, each local time series segment corresponds to a unique trend change slope and fluctuation amplitude. The trend change slope and fluctuation amplitude corresponding to all local time series segments are the extracted local time series segment trend characteristics.

[0054] The trend change slope and fluctuation amplitude are the core quantitative trend features obtained after trend feature extraction of each local time series segment. The evolution of the historical residual time series is a comprehensive reflection of the changing trend and discrete state of the residual data at each time stage. The situation mapping uses the trend change slope corresponding to each local time series segment as the vertical component of the vector and the fluctuation amplitude corresponding to the local time series segment as the horizontal component of the vector. Combining the time attributes of the local time series segments, a two-dimensional vector is constructed for each local time series segment. The direction of the two-dimensional vector is determined by the numerical magnitude of the slope and the fluctuation amplitude. The magnitude of the two-dimensional vector is obtained by taking the square root of the sum of the square of the slope and the square of the fluctuation amplitude. Each two-dimensional vector corresponds to the residual evolution trend within a time window in the historical residual time series. This type of two-dimensional vector, which corresponds one-to-one with the local time series segment, is the instantaneous trend pointing vector of the historical residual time series. All local time series segments correspond to a unique instantaneous trend pointing vector, and the time order of the instantaneous trend pointing vector is completely consistent with the time order of the corresponding local time series segment in the historical residual time series.

[0055] The instantaneous trend pointing vector is a two-dimensional vector obtained after situation mapping of each local time segment. Each instantaneous trend pointing vector has a temporal correlation consistent with the original residual sequence. The temporal concatenation is to sequentially connect all instantaneous trend pointing vectors according to the temporal order corresponding to the instantaneous trend pointing vectors, so that the end of the previous instantaneous trend pointing vector is connected to the beginning of the next instantaneous trend pointing vector, forming a continuous vector sequence with temporal correlation. This continuous vector sequence can completely reflect the continuous evolution of the trend change and fluctuation degree of the historical residual time series in the entire time dimension. This continuous vector sequence that perfectly matches the time dimension of the historical residual time series is the historical residual trend vector of the heat exchanger. The vector features of each node in this vector sequence maintain a precise correlation with the residual features of the corresponding time window of the historical residual time series.

[0056] The beneficial effects of this invention are as follows: sliding window segmentation can refine the analysis of residual time series data, preserve time correlation, avoid information loss, improve the targeting of feature extraction, and trend feature extraction can quantify the local residual situation in two dimensions, improve feature objectivity, provide quantitative basis for situation mapping, and the situation mapping will concretize the abstract residual trend features into vectors to realize quantitative visualization, lay a structured foundation for time series connection, and the time series connection integrates a continuous vector sequence to fully restore the overall evolution trend of the residuals, providing accurate structured vector support for manifold space embedding.

[0057] S4: Embed the historical residual trend vector into the manifold space to obtain the baseline manifold structure of the residual evolution trend of the power plant heat exchanger under healthy conditions.

[0058] In one possible implementation, S4 specifically includes sub-steps S401 to S405: S401: Perform local neighborhood detection on the historical residual trend vector to obtain the nearest neighbor index set of the historical residual trend vector.

[0059] S402: Based on the nearest neighbor index set, perform local weight evaluation on the historical residual trend vector to obtain the local reconstruction weight of the historical residual trend vector.

[0060] In one possible implementation, S402 specifically includes sub-steps S4021 to S4023: S4021: Calculate the spatial distance between the neighbor vectors indicated by the nearest neighbor index set and the historical residual trend vector to obtain the local neighborhood distance metric of the historical residual trend vector.

[0061] S4022: Based on the local neighborhood distance metric, perform linear characterization analysis on the neighboring vectors to obtain the local reconstruction coefficients of the historical residual trend vector.

[0062] S4023: Normalize and arrange the local reconstruction coefficients to obtain the local reconstruction weights of the historical residual trend vector.

[0063] S403: Based on the local reconstruction weights, perform feature space mapping on the historical residual trend vector to obtain the low-dimensional manifold coordinates of the historical residual trend vector.

[0064] S404: Tracing the evolution path of low-dimensional manifold coordinates to obtain the core evolution path of low-dimensional manifold coordinates.

[0065] S405: Based on the core evolution path, the topological structure of the geometry of the low-dimensional manifold coordinates is characterized to obtain the reference manifold structure of the power plant heat exchanger.

[0066] For each historical residual trend vector in the set of historical residual trend vectors formed by the temporal connection of the instantaneous trend pointing vectors of the historical residual time series of the heat exchanger, a fixed neighborhood range is defined with the vector as the center based on the actual characteristics of residual evolution under the healthy operating state of the heat exchanger. All other historical residual trend vectors within the neighborhood range with a spatial correlation degree higher than a preset threshold are screened out. The threshold of spatial correlation degree is set according to the distribution density of residual trend vectors under the healthy state. The higher the distribution density, the higher the threshold is set, and the lower the distribution density, the lower the threshold is set. Then, the position identifiers of these screened vectors in the set of historical residual trend vectors are summarized and organized to form an ordered identifier set. This ordered identifier set is the nearest neighbor index set of the historical residual trend vectors.

[0067] The nearest neighbor index set indicates all historical residual trend vectors pointed to by the index set. For each historical residual trend vector, the spatial distance between the vector and each of its corresponding nearest neighbor vectors in the feature space is calculated in turn. During the calculation, the differences of each dimension component of the vector are calculated in turn. All differences are squared and then summed. The square root of the sum is then taken to obtain the single spatial distance between the two vectors. The single spatial distances between the historical residual trend vector and all its nearest neighbor vectors are summarized and arranged in descending order of value. Each single spatial distance is associated with a one-to-one relationship with its corresponding nearest neighbor vector. The resulting ordered set of distances is the local neighborhood distance metric of the historical residual trend vector.

[0068] Based on the ordered set of spatial distances between the historical residual trend vector and each neighboring vector, i.e., the local neighborhood distance metric, a representation weight is assigned to the corresponding neighboring vector according to the magnitude of each spatial distance in the set. The smaller the spatial distance value, the larger the representation weight assigned, and vice versa. Then, all neighboring vectors are linearly combined according to the assigned representation weights. By adjusting the representation weight values ​​of each neighboring vector, the feature deviation between the linearly combined vector and the target historical residual trend vector is minimized. The feature deviation is determined by the sum of the deviations of each dimension component of the vector being within the preset minimum deviation range determined based on the fluctuation characteristics of the residual trend under the health state of the heat exchanger. At this point, the representation weight values ​​corresponding to each neighboring vector are the local reconstruction coefficients of the historical residual trend vector.

[0069] The sum of all values ​​in the set of representation weights corresponding to each neighboring vector is calculated to obtain the total sum of coefficients. Each local reconstruction coefficient in the set is divided by this total sum of coefficients to obtain the normalized value corresponding to each coefficient. Then, all the normalized values ​​are sorted in order according to the correspondence between the original local reconstruction coefficients and neighboring vectors to form a normalized set of values. The sum of all values ​​in this set is 1. This set of values ​​is the local reconstruction weight of the historical residual trend vector.

[0070] Based on the normalized set of representation weights, i.e., the local reconstruction weights, the historical residual trend vector is weighted and fused with all its corresponding neighboring vectors. During fusion, each dimension component of each neighboring vector is multiplied by the corresponding local reconstruction weight, and then the same dimension components of all weighted neighboring vectors are summed to obtain the fused vector features. The fused vector features are then mapped from the original high-dimensional feature space to a low-dimensional feature space determined by the number of core features of the residual evolution trend under the health state. During the mapping process, the core evolution features of the vector are retained. The core evolution features are the residual trend features that play a key role in determining the health state of the heat exchanger. Finally, a unique position identifier is determined for the mapped vector in the low-dimensional feature space. This position identifier is the low-dimensional manifold coordinate of the historical residual trend vector.

[0071] Low-dimensional manifold coordinates are a set of positional identifiers for historical residual trend vectors in a low-dimensional feature space. Each coordinate in this set carries a corresponding timestamp. All low-dimensional manifold coordinates are sequentially connected in order of their timestamps from earliest to latest to form an initial evolution path. The initial evolution path is then smoothed to remove abnormal coordinate points caused by minor fluctuations in the residuals. Abnormal coordinate points are determined when the spatial distance between the coordinate and its surrounding continuous coordinates exceeds a preset distance threshold determined based on the distribution characteristics of low-dimensional manifold coordinates under healthy conditions. The smoothed evolution path is then simplified to retain key inflection points where the trend changes and the overall trend. The resulting continuous path is the core evolution path of the low-dimensional manifold coordinates.

[0072] Centered on the core evolution path, which is the continuous evolution path of the processed low-dimensional manifold coordinates, a spatial range encompassing all low-dimensional manifold coordinates under healthy conditions is delineated. The boundary of this spatial range is determined based on the maximum spatial distance between all low-dimensional manifold coordinates and the core evolution path. Then, the distribution characteristics of all low-dimensional manifold coordinates within this spatial range are extracted. The extracted features include the clustering regions, dispersion, and trend extension direction of the coordinates. Finally, a geometric topology matching the extracted distribution characteristics is constructed. This topology can fully present the overall characteristics and evolution law of the residual evolution trend in the low-dimensional manifold space under healthy conditions of the heat exchanger. The constructed geometric topology is the benchmark manifold structure of the residual evolution trend under healthy conditions of the heat exchanger.

[0073] The beneficial effects of this invention are as follows: it accurately selects neighboring vectors, improves the pertinence and accuracy of subsequent weight evaluation, accurately reflects the correlation of vector space, provides an objective basis for weight allocation, improves the rationality of allocation, accurately represents the linear contribution of vectors, improves the accuracy of linear representation, strengthens the foundation of normalized data processing, achieves standardization of weight proportions, avoids mapping bias, improves the accuracy of feature space mapping results, realizes the dimensionality reduction of high-dimensional vectors, simplifies processing complexity while retaining core information, improves data processing efficiency and effectiveness, accurately presents the residual evolution law, eliminates invalid interference, improves the accuracy and simplicity of trend extraction, fully presents the core features of residual evolution, provides a standardized and accurate benchmark, and improves the accuracy and reliability of heat exchanger degradation early warning.

[0074] S5: During the real-time monitoring phase, the real-time operating parameters of the power plant heat exchanger are spatially mapped to the reference manifold structure to obtain the real-time mapped coordinates of the power plant heat exchanger.

[0075] In one possible implementation, S5 specifically includes sub-steps S501 to S503: S501: Based on the residual evolution law corresponding to the benchmark manifold structure, residual analysis is performed on the real-time multidimensional operating parameters of the power plant heat exchanger to obtain the real-time residual evolution time series of the power plant heat exchanger.

[0076] S502: Based on the trend evolution law corresponding to the baseline manifold structure, the real-time residual evolution time series is mapped with trend features to obtain the real-time trend vector of the power plant heat exchanger.

[0077] S503: Based on the embedding mapping rules of the reference manifold structure, the real-time trend vector is located in manifold space to obtain the real-time mapping coordinates of the power plant heat exchanger.

[0078] When performing residual analysis on the real-time multidimensional operating parameters of a heat exchanger based on the residual evolution law corresponding to the baseline manifold structure, the time-series data of the temperature, pressure, flow rate, and other multidimensional operating parameters continuously collected during the real-time monitoring of the heat exchanger are first extracted. This data is the raw data of the real-time multidimensional operating parameters. Then, following the inherent logic of residual analysis under the healthy state in the baseline manifold structure, the time-series features of each parameter dimension of the real-time multidimensional operating parameters are extracted sequentially. Next, the time-series trend fitting is performed on the real-time operating parameters of each dimension to obtain the real-time theoretical prediction value of each dimension. Then, the time-by-time numerical difference between the real-time measured values ​​of each dimension and the real-time theoretical prediction value of the corresponding dimension is calculated. Finally, the time-by-time numerical differences of all parameter dimensions are serialized and arranged in chronological order. The resulting continuous time-series data is the real-time residual evolution time series of the heat exchanger.

[0079] When mapping the trend features of the real-time residual sequence according to the trend evolution law corresponding to the benchmark manifold structure, the real-time residual evolution time series is first divided into sliding window segments with a predetermined window length in the benchmark manifold structure. After segmentation, several continuous and ordered local real-time residual evolution time series segments are obtained. Then, the trend features of each local real-time residual evolution time series segment are calculated. First, the trend change slope of the segment is obtained by linear fitting of the time series points of the residual values ​​within the segment. Then, the fluctuation amplitude of the segment is obtained by calculating the difference between the maximum and minimum values ​​of the residual values ​​within the segment. Next, according to the predetermined situation mapping rules in the healthy state of the benchmark manifold structure, the trend change slope and fluctuation amplitude of each local real-time residual evolution time series segment are mapped to a unique spatial pointing feature to form the instantaneous trend pointing vector of each local segment. Finally, all instantaneous trend pointing vectors are continuously linked in time sequence. The resulting vector sequence is the real-time trend vector of the heat exchanger.

[0080] When locating the real-time trend vector in manifold space according to the embedding mapping rules of the benchmark manifold structure, the eigenvalues ​​of all feature dimensions of the real-time trend vector are first extracted. Then, the spatial mapping relationship and topological correspondence rules for converting the historical residual trend vector in the benchmark manifold structure into low-dimensional manifold coordinates are retrieved. These rules include a one-to-one correspondence between the feature dimensions of the historical residual trend vector and the coordinate axes of the low-dimensional manifold space, and a fixed conversion ratio between the vector eigenvalues ​​and the coordinate values ​​of the manifold space. Next, according to the correspondence between the feature dimensions and the coordinate axes of the manifold space in the rules, the eigenvalues ​​of each feature dimension of the real-time trend vector are numerically converted according to the fixed conversion ratio to obtain the coordinate values ​​corresponding to each coordinate axis of the low-dimensional manifold space. Finally, these coordinate values ​​are mapped to the low-dimensional manifold space of the benchmark manifold structure. The coordinates corresponding to the unique spatial location point determined in this space are the real-time mapped coordinates of the heat exchanger.

[0081] The beneficial effects of this invention are as follows: it improves the matching between real-time residuals and the baseline manifold structure, ensures the accuracy of the evolution timeline of real-time residuals, provides a suitable time-series data foundation for subsequent trend feature mapping, unifies the extraction standards of trend vectors in real-time and healthy states, ensures the feature effectiveness and dimensional matching of real-time trend vectors, provides accurate vector feature support for subsequent manifold space positioning, realizes the accurate conversion of real-time trend vectors to the low-dimensional space of the baseline manifold, ensures the positional accuracy of real-time mapping coordinates, provides a reliable coordinate basis for subsequent alienation degree determination, and improves the accuracy of the heat exchanger early warning data foundation.

[0082] S6: Determine the outlier degree of the local alienation of the real-time mapped coordinates on the reference manifold structure to obtain the performance degradation early warning signal of the power plant heat exchanger.

[0083] In one possible implementation, S6 specifically includes sub-steps S601 to S605: S601: Based on the baseline manifold structure, the neighborhood range of the real-time mapped coordinates is defined to obtain the local neighborhood set of the real-time mapped coordinates.

[0084] S602: Perform a separation evaluation between the reference coordinates within the local neighborhood set and the real-time mapped coordinates to obtain the local separation amplitude of the real-time mapped coordinates.

[0085] S603: Based on the residual evolution characteristics of the power plant heat exchanger under healthy operating conditions, a threshold boundary is set for the local alienation amplitude to obtain the alienation judgment standard of the power plant heat exchanger.

[0086] S604: Match and verify the local alienation amplitude with the alienation determination criteria to obtain the outlier status indication of the real-time mapped coordinates.

[0087] S605: Based on the outlier status indication, when the local outlier amplitude exceeds the outlier determination standard, the power plant heat exchanger degradation warning is activated to generate a power plant heat exchanger performance degradation warning signal.

[0088] Using the baseline manifold structure representing the residual evolution trend under the healthy state of the heat exchanger as a reference, this baseline manifold structure contains the low-dimensional manifold coordinates corresponding to all residual trend vectors during the healthy operation of the heat exchanger. With the real-time mapped coordinates as the spatial center point, the number of coordinates of all healthy coordinates in the baseline manifold structure within a unit volume in space is first counted to obtain the overall spatial distribution density. Then, the overall average value of this spatial distribution density is calculated and the average density interval is determined. The spatial region where the number of coordinates within a unit volume is within this average density interval is defined as the neighborhood boundary of the real-time mapped coordinates. Finally, all healthy coordinates in the baseline manifold structure contained within this neighborhood boundary are aggregated. The resulting coordinate set is the local neighborhood set of the real-time mapped coordinates.

[0089] For each reference coordinate in the local neighborhood set, the spatial Euclidean distance between it and the real-time mapped coordinate is calculated sequentially. After the distance calculation of all reference coordinates is completed, the arithmetic mean of all the spatial Euclidean distances is calculated to obtain the corresponding average spatial distance. Then, the spatial Euclidean distances calculated between all healthy coordinates in the reference manifold structure are statistically analyzed and the arithmetic mean is calculated to obtain the overall average spatial distance of the healthy coordinates. The difference between the average spatial distance corresponding to the real-time mapped coordinate and the overall average spatial distance of the healthy coordinate is calculated. The non-negative difference result is the local alienation amplitude of the real-time mapped coordinate. The magnitude of this amplitude directly reflects the degree of spatial deviation of the real-time mapped coordinate relative to the reference coordinates in the local neighborhood.

[0090] The residual evolution characteristics, such as residual fluctuation range and residual evolution rate, are extracted from the historical residual time series under the healthy operating state of the heat exchanger. Based on these characteristics, the low-dimensional manifold coordinates corresponding to all historical residual trend vectors under the healthy operating state are retrieved. Then, the alienation amplitude of these healthy coordinates in their respective local neighborhoods is calculated. All the obtained historical alienation amplitudes are sorted in ascending order. The historical alienation amplitude with the value at the 99th percentile after sorting is selected as the upper limit threshold, while the lower limit threshold is set to 0. The numerical range formed by the upper limit threshold and the lower limit threshold is the alienation judgment standard of the heat exchanger. This standard is the exclusive numerical basis for judging whether the real-time mapped coordinates are in an outlier state.

[0091] The local alienation amplitude of the real-time mapped coordinates is precisely compared with the numerical range corresponding to the alienation determination criterion. If the local alienation amplitude is within the specified range, the real-time mapped coordinates are directly determined to be in a normal state, and an outlier state indicator representing this state is generated. If the local alienation amplitude exceeds the upper threshold of the specified range, the real-time mapped coordinates are directly determined to be in an outlier state, and an outlier state indicator representing this state is generated. The outlier state indicator is used to clearly identify the spatial position of the real-time mapped coordinates within the reference manifold structure.

[0092] The system identifies the outlier status indicator, which includes two status markers: normal and outlier. When the identification result is an outlier status marker, it means that the local alienation amplitude of the real-time mapped coordinates exceeds the alienation judgment standard. At this time, the activation program of the heat exchanger degradation early warning is immediately triggered. This program automatically retrieves the heat exchanger's real-time operating parameter information, real-time mapped coordinate information, and local alienation amplitude information, integrates this information with the preset early warning marker to form an information set, and then encodes the information set according to the preset signal format of the heat exchanger monitoring system. The information formed after encoding is the heat exchanger performance degradation early warning signal, which is directly transmitted to the heat exchanger monitoring terminal.

[0093] The beneficial effects of this invention are as follows: defining the neighborhood according to the reference manifold structure density improves the rationality of the definition, ensures the accuracy of subsequent evaluation data, quantitatively calculates the local alienation amplitude, accurately reflects the degree of coordinate deviation, improves the evaluation accuracy and objectivity, sets judgment criteria in combination with the health characteristics of the heat exchanger, improves adaptability and scientificity, provides reliable judgment basis, generates outlier state indication by amplitude comparison, improves the clarity of judgment, accurately outputs the coordinate space state, and promptly activates early warning and integrates generated signals when outliers occur, improving the completeness of early warning information and response efficiency, and achieving timely early warning. Example

[0094] like Figure 2 As shown in the figure, this embodiment also provides a functional module diagram of a power plant heat exchanger early warning system based on time series modeling.

[0095] The power plant heat exchanger early warning system 100 based on time series modeling described in this embodiment can be installed in an electronic device. Depending on the functions implemented, the power plant heat exchanger early warning system 100 based on time series modeling may include a health time series parameter construction module 101, a historical residual sequence analysis module 102, a residual trend vectorization module 103, a baseline manifold structure construction module 104, a real-time manifold coordinate mapping module 105, and a real-time coordinate manifold analysis module 105. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0096] In this embodiment, the functions of each module / unit are as follows: The health time series parameter construction module 101 is used to collect historical multi-dimensional operating parameters of the power plant heat exchanger under healthy operating conditions in order to construct a historical parameter time series set of the power plant heat exchanger.

[0097] The historical residual sequence parsing module 102 is used to perform time series extrapolation and residual parsing on the historical parameter time series set to obtain the historical residual time series of the power plant heat exchanger.

[0098] The residual trend vectorization module 103 is used to perform time-series trend vectorization on the historical residual time series to obtain the historical residual trend vector of the power plant heat exchanger.

[0099] The reference manifold structure construction module 104 is used to embed the historical residual trend vector into the manifold space to obtain the reference manifold structure of the residual evolution trend of the power plant heat exchanger under healthy conditions.

[0100] The real-time manifold coordinate mapping module 105 is used to map the real-time operating parameter space of the power plant heat exchanger to the reference manifold structure during the real-time monitoring phase, so as to obtain the real-time mapped coordinates of the power plant heat exchanger.

[0101] The real-time coordinate manifold analysis module 106 is used to determine the degree of local alienation of the real-time mapped coordinates on the reference manifold structure, so as to obtain the performance degradation early warning signal of the power plant heat exchanger.

[0102] In one possible implementation, the health timing parameter construction module is specifically used for: Multidimensional measurement parameters of power plant heat exchangers under healthy operating conditions are extracted from the historical operation database of power plant heat exchangers.

[0103] By performing steady-state condition determination on the parameters of multi-dimensional measuring points, the steady-state condition segment parameters of the power plant heat exchanger are obtained.

[0104] By aligning the parameters of the steady-state operating condition segment in the time domain, the synchronous timing sequence of the power plant heat exchanger is obtained.

[0105] The synchronous time sequence is fused and assembled to obtain the historical parameter time sequence set of the power plant heat exchanger.

[0106] In one possible implementation, the historical residual sequence parsing module is specifically used for: We analyze the time-dependent characteristics of the parameter dimensions in the historical parameter time series set to obtain the autocorrelation evolution characteristics of the historical parameter time series set.

[0107] Based on autocorrelation evolution characteristics, the time-series evolution trajectory of the parameter dimension is trend-fitted to obtain the theoretical predicted value of the parameter dimension.

[0108] Based on theoretical predictions, the measured values ​​in the historical parameter time series are compared hour by hour to obtain the degree of deviation between the measured values ​​and the theoretical predictions.

[0109] The numerical deviations are serialized and arranged in chronological order to obtain the historical residual time series of the power plant heat exchanger.

[0110] In one possible implementation, the residual trend vectorization module is specifically used for: By performing sliding window segmentation on the historical residual time series, local time segments of the historical residual time series can be obtained.

[0111] Trend features are extracted from local time series segments to obtain the slope and amplitude of trend changes in the local time series segments.

[0112] Based on the slope of trend change and the amplitude of fluctuation, the evolution of the historical residual time series is mapped to obtain the instantaneous trend pointing vector of the historical residual time series.

[0113] By performing a time-series concatenation on the instantaneous trend pointing vector, the historical residual trend vector of the power plant heat exchanger is obtained.

[0114] In one possible implementation, the baseline manifold structure building module is specifically used for: Local neighborhood detection is performed on the historical residual trend vector to obtain the nearest neighbor index set of the historical residual trend vector.

[0115] Based on the nearest neighbor index set, local weight evaluation is performed on the historical residual trend vector to obtain the local reconstruction weight of the historical residual trend vector.

[0116] Based on the local reconstruction weights, the historical residual trend vector is mapped to the feature space to obtain the low-dimensional manifold coordinates of the historical residual trend vector.

[0117] By tracing the evolution path of low-dimensional manifold coordinates, the core evolution path of low-dimensional manifold coordinates can be obtained.

[0118] Based on the core evolution path, the topological structure of the low-dimensional manifold coordinates is characterized to obtain the reference manifold structure of the power plant heat exchanger.

[0119] In one possible implementation, the baseline manifold structure building module is further used for: Spatial distance is calculated between the neighbor vectors indicated by the nearest neighbor index set and the historical residual trend vector to obtain the local neighborhood distance metric of the historical residual trend vector.

[0120] Based on the local neighborhood distance metric, the neighboring vectors are linearly represented and analyzed to obtain the local reconstruction coefficients of the historical residual trend vector.

[0121] The local reconstruction coefficients are normalized and arranged to obtain the local reconstruction weights of the historical residual trend vector.

[0122] In one possible implementation, the real-time manifold coordinate mapping module is specifically used for: Based on the residual evolution law corresponding to the baseline manifold structure, residual analysis is performed on the real-time multidimensional operating parameters of the power plant heat exchanger to obtain the real-time residual evolution time series of the power plant heat exchanger.

[0123] Based on the trend evolution law corresponding to the baseline manifold structure, the real-time residual evolution time series is mapped with trend features to obtain the real-time trend vector of the power plant heat exchanger.

[0124] Based on the embedding mapping rules of the reference manifold structure, the real-time trend vector is located in manifold space to obtain the real-time mapped coordinates of the power plant heat exchanger.

[0125] In one possible implementation, the real-time coordinate manifold analysis module is specifically used for: Based on the baseline manifold structure, the neighborhood range of the real-time mapped coordinates is defined to obtain the local neighborhood set of the real-time mapped coordinates.

[0126] The local distance between the reference coordinates in the local neighborhood set and the real-time mapped coordinates is evaluated to obtain the local distance magnitude of the real-time mapped coordinates.

[0127] Based on the residual evolution characteristics of power plant heat exchangers under healthy operating conditions, a threshold boundary is set for the local alienation amplitude to obtain the alienation judgment standard for power plant heat exchangers.

[0128] By matching and verifying the local alienation amplitude with the alienation determination criteria, an outlier status indication of the real-time mapped coordinates is obtained.

[0129] According to the outlier status indication, when the local outlier amplitude exceeds the outlier determination standard, the power plant heat exchanger degradation early warning is activated to generate a performance degradation early warning signal for the power plant heat exchanger.

[0130] In detail, each module in the power plant heat exchanger early warning system 100 based on time series modeling described in this embodiment of the invention uses the same technical means as the power plant heat exchanger early warning method based on time series modeling described in Embodiment 1, and can produce the same technical effect, which will not be repeated here. Example

[0131] According to embodiments of this disclosure, this disclosure also provides an electronic device.

[0132] Figure 3 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0133] like Figure 3 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0134] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0135] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as a power plant heat exchanger early warning method based on time series modeling. For example, in some embodiments, the power plant heat exchanger early warning method based on time series modeling can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the power plant heat exchanger early warning method based on time series modeling described above can be performed. Alternatively, in other embodiments, the computing unit 801 may be configured by any other suitable means (e.g., by means of firmware) to perform a power plant heat exchanger early warning method based on time series modeling.

[0136] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0137] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0138] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0139] The systems and technologies described herein can be implemented in computing systems that include back-end components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0140] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0141] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0142] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means two or more, unless otherwise explicitly specified.

[0143] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A power plant heat exchanger early warning method based on time series modeling, characterized in that, The method includes: S1: Collect historical multi-dimensional operating parameters of the power plant heat exchanger under healthy operating conditions to construct a time series set of historical parameters of the power plant heat exchanger; S2: Perform time series extrapolation and residual analysis on the historical parameter time series set to obtain the historical residual time series of the power plant heat exchanger; S3: Perform time-series trend vectorization on the historical residual time series to obtain the historical residual trend vector of the power plant heat exchanger; S4: Embed the historical residual trend vector into the manifold space to obtain the baseline manifold structure of the residual evolution trend of the power plant heat exchanger under healthy conditions; S5: During the real-time monitoring phase, the real-time operating parameters of the power plant heat exchanger are spatially mapped to the reference manifold structure to obtain the real-time mapped coordinates of the power plant heat exchanger. S6: Determine the degree of local alienation of the real-time mapped coordinates on the reference manifold structure to obtain the performance degradation early warning signal of the power plant heat exchanger.

2. The power plant heat exchanger early warning method based on time series modeling according to claim 1, characterized in that, S1 specifically includes: S101: Extract the multi-dimensional measurement parameters of the power plant heat exchanger under healthy operating conditions from the historical operation database of the power plant heat exchanger; S102: Perform steady-state condition discrimination on the multi-dimensional measuring point parameters to obtain the steady-state condition segment parameters of the power plant heat exchanger; S103: Time-domain alignment of the steady-state operating condition segment parameters to obtain the synchronization time sequence of the power plant heat exchanger; S104: The synchronous timing sequence is fused and compiled to obtain the historical parameter timing set of the power plant heat exchanger.

3. The power plant heat exchanger early warning method based on time series modeling according to claim 1, characterized in that, S2 specifically includes: S201: Perform time-series dependency analysis on the parameter dimensions in the historical parameter time series set to obtain the autocorrelation evolution characteristics of the historical parameter time series set; S202: Based on the autocorrelation evolution characteristics, perform trend fitting on the time-series evolution trajectory of the parameter dimension to obtain the theoretical predicted value of the parameter dimension; S203: Based on the theoretical prediction value, the measured values ​​in the historical parameter time series set are compared time-by-time to obtain the degree of deviation between the measured values ​​and the theoretical prediction value. S204: The deviation of the values ​​is serialized and arranged in chronological order to obtain the historical residual time series of the power plant heat exchanger.

4. The power plant heat exchanger early warning method based on time series modeling according to claim 1, characterized in that, S3 specifically includes: S301: Perform sliding window segmentation on the historical residual time series to obtain local time segments of the historical residual time series; S302: Perform trend feature extraction on the local time series segment to obtain the trend change slope and fluctuation amplitude of the local time series segment; S303: Based on the trend change slope and the fluctuation amplitude, perform a trend mapping on the evolution of the historical residual time series to obtain the instantaneous trend pointing vector of the historical residual time series; S304: Perform a time-series concatenation on the instantaneous trend pointing vector to obtain the historical residual trend vector of the power plant heat exchanger.

5. The power plant heat exchanger early warning method based on time series modeling according to claim 1, characterized in that, S4 specifically includes: S401: Perform local neighborhood detection on the historical residual trend vector to obtain the nearest neighbor index set of the historical residual trend vector; S402: Based on the nearest neighbor index set, perform local weight evaluation on the historical residual trend vector to obtain the local reconstruction weight of the historical residual trend vector; S403: Based on the local reconstruction weights, perform feature space mapping on the historical residual trend vector to obtain the low-dimensional manifold coordinates of the historical residual trend vector; S404: Tracing the evolution path of the low-dimensional manifold coordinates to obtain the core evolution path of the low-dimensional manifold coordinates; S405: Based on the core evolution path, the geometric shape of the low-dimensional manifold coordinates is topologically characterized to obtain the reference manifold structure of the power plant heat exchanger.

6. The power plant heat exchanger early warning method based on time series modeling according to claim 5, characterized in that, Specifically, S402 includes: S4021: Calculate the spatial distance between the neighbor vectors indicated by the nearest neighbor index set and the historical residual trend vector to obtain the local neighborhood distance metric of the historical residual trend vector. S4022: Based on the local neighborhood distance metric, perform linear characterization analysis on the neighbor vector to obtain the local reconstruction coefficients of the historical residual trend vector; S4023: Normalize and arrange the local reconstruction coefficients to obtain the local reconstruction weights of the historical residual trend vector.

7. The power plant heat exchanger early warning method based on time series modeling according to claim 1, characterized in that, S5 specifically includes: S501: Based on the residual evolution law corresponding to the reference manifold structure, residual analysis is performed on the real-time multidimensional operating parameters of the power plant heat exchanger to obtain the real-time residual evolution time series of the power plant heat exchanger. S502: Based on the trend evolution law corresponding to the reference manifold structure, perform trend feature mapping on the real-time residual evolution time series to obtain the real-time trend vector of the power plant heat exchanger; S503: Based on the embedding mapping rules of the reference manifold structure, perform manifold space positioning on the real-time trend vector to obtain the real-time mapping coordinates of the power plant heat exchanger.

8. The power plant heat exchanger early warning method based on time series modeling according to claim 1, characterized in that, S6 specifically includes: S601: Based on the reference manifold structure, the neighborhood range of the real-time mapped coordinates is defined to obtain the local neighborhood set of the real-time mapped coordinates; S602: Perform a separation evaluation between the reference coordinates in the local neighborhood set and the real-time mapped coordinates to obtain the local separation amplitude of the real-time mapped coordinates; S603: Based on the residual evolution characteristics of the power plant heat exchanger under healthy operating conditions, a threshold boundary is set for the local alienation amplitude to obtain the alienation determination standard of the power plant heat exchanger. S604: Match and verify the local alienation amplitude with the alienation determination criterion to obtain the outlier status indication of the real-time mapped coordinates; S605: According to the outlier status indication, when the local alienation amplitude exceeds the alienation degree judgment standard, the power plant heat exchanger is activated for degradation early warning to generate a performance degradation early warning signal for the power plant heat exchanger.

9. A power plant heat exchanger early warning system based on time series modeling, characterized in that, The system is used to implement the power plant heat exchanger early warning method based on time series modeling as described in any one of claims 1 to 8, the system comprising: The health time series parameter construction module is used to collect historical multi-dimensional operating parameters of the power plant heat exchanger under healthy operating conditions in order to construct the historical parameter time series set of the power plant heat exchanger; The historical residual sequence parsing module is used to perform time series extrapolation and residual parsing on the historical parameter time series set to obtain the historical residual time series of the power plant heat exchanger. The residual trend vectorization module is used to perform time-series trend vectorization on the historical residual time series to obtain the historical residual trend vector of the power plant heat exchanger. A baseline manifold structure construction module is used to embed the historical residual trend vector into the manifold space to obtain the baseline manifold structure of the residual evolution trend of the power plant heat exchanger under healthy conditions. The real-time manifold coordinate mapping module is used to map the real-time operating parameter space of the power plant heat exchanger to the reference manifold structure during the real-time monitoring phase, so as to obtain the real-time mapped coordinates of the power plant heat exchanger. The real-time coordinate manifold analysis module is used to determine the degree of local alienation of the real-time mapped coordinates on the reference manifold structure, so as to obtain the performance degradation early warning signal of the power plant heat exchanger.

10. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the power plant heat exchanger early warning method based on time series modeling as described in any one of claims 1 to 8.

Images