An intelligent monitoring method and system for gas turbine unit state evaluation

Abstract

The present application relates to the technical field of intelligent monitoring, in particular to an intelligent monitoring method and system for state evaluation of a gas unit, comprising the following steps: performing differential calculation on oxygen concentration information to generate a horizontal distribution graph, comparing adjacent measurement point trends to filter positions to generate a starting area graph, deriving a connection line to generate a combustion diffusion trajectory graph, dividing levels according to the trajectory to generate a start-stop distribution graph, and combining information to judge to generate a unit monitoring scheme. In the present application, a horizontal change gradient is constructed based on a multi-sensor sequence differential evolution process, an initial area and a spread path are defined through multi-point time sequence comparison, a monitoring deduction mechanism based on space-time correlation is constructed to eliminate the risk of misjudgment, parameter abnormality signs are captured in advance to establish unit operation evolution characteristics, and the limitations of lagging review and static rule benchmarking are abandoned, realizing the tracking of parameter abnormality to overall state deviation, improving the forward-looking nature of equipment hazard exploration and the effectiveness of monitoring analysis.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, and in particular to an intelligent monitoring method and system for assessing the condition of gas turbine units. Background Technology

[0002] Intelligent monitoring technology is primarily used for comprehensive analysis and operational decision-making regarding the operating status of industrial equipment. Its foundation relies on the acquisition and processing of multi-source sensor data. During the operation of gas turbine units, sensors typically acquire operational data such as temperature, pressure, current, vibration, and combustion parameters. This data is then combined with data comparison, threshold judgment, and historical operating condition analysis to identify the equipment's operating status. Simultaneously, a centralized control system aggregates and displays operational information from multiple devices, providing operators with a basis for status assessment and control, thus forming an online monitoring and status evaluation technology system for industries such as power. Traditional gas turbine unit monitoring methods typically rely on operational data collected by a monitoring system. This involves displaying and comparing steam parameters, combustion parameters, and unit load data for equipment such as boilers, turbines, and generators. Abnormal data is marked based on preset operating curves and alarm thresholds. Operators then review historical records and trend curves to make status judgments and operational decisions.

[0003] However, existing gas turbine monitoring methods mainly rely on static comparison of data output by the monitoring system and preset operating curves, which has obvious limitations: fixed thresholds are difficult to adapt to complex operating conditions, single-point anomaly alarms lack multi-parameter correlation analysis capabilities, and it is difficult to identify the spatial evolution characteristics of the combustion process in a timely manner; at the same time, the monitoring process is highly dependent on human experience, and when faced with multiple measurement points and multiple time series data, manual judgment is inefficient and inconsistent, making it difficult to achieve continuous tracking and forward-looking early warning of the combustion state evolution process, resulting in a lag in the identification of unit operation risks. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent monitoring method for gas turbine unit condition assessment, comprising the following steps: S1: Multiple oxygen sensors are arranged at the same height cross section of the boiler furnace to collect continuous time series oxygen concentration change information. The oxygen concentration change information at adjacent sampling times is differentially calculated, and the change trend of the measuring points is calculated according to the time sequence to generate a horizontal change distribution map. S2: Compare and analyze the horizontal variation distribution map, collect the change trend of adjacent measuring points and compare them point by point, determine the position of the measuring point according to the size law of the change trend of the continuous sampling period, screen the measuring point positions that meet the conditions, and generate a combustion initiation area identification map. S3: Based on the combustion initiation zone identification map, derive the horizontal measurement point path, collect the measurement point change trend for continuous comparison, connect the measurement point positions according to the progressive relationship of the change trend, analyze the connection results, and generate a combustion diffusion trajectory map; S4: Divide the combustion diffusion trajectory map into segments, collect the positions of the measuring points in the combustion diffusion trajectory map and sort them according to the trajectory order, filter the sorting results and divide the measuring point levels, and generate a measuring point start-stop distribution map. S5: Construct monitoring content for the start-stop distribution map of the measuring points, collect oxygen concentration change information of the measuring points in the start-stop distribution map of the measuring points and combine them in time order, judge the combination result, and generate a smart monitoring scheme for the gas turbine unit status.

[0005] As a further aspect of the present invention, the lateral variation distribution map includes a cross-sectional oxygen gradient field, a measuring point differential amplitude grid, and a trend variation thermogram; the combustion initiation region identification map includes an initial ignition point coordinate set, a boundary of intense oxidation reaction, and geometric features of the combustion pilot zone; the combustion diffusion trajectory map includes flame spread spatial lines, oxygen mutation propagation routes, and a combustion zone expansion vector network; the measuring point start-stop distribution map includes a sensor activation priority sequence, a measuring point monitoring depth echelon, and a data acquisition sleep interval; and the gas turbine unit status intelligent monitoring scheme includes a furnace combustion condition diagnostic panel, oxygen abnormality alarm threshold settings, and a dynamic unit load adaptation strategy.

[0006] As a further aspect of the present invention, the step of determining the location of the measuring point based on the magnitude of the change trend of the continuous sampling period refers to identifying and verifying the spatial location of the measuring point that conforms to a specific change pattern based on the characteristics of the change amplitude, direction and stability of oxygen concentration in the continuous time series.

[0007] As a further aspect of the present invention, the progressive relationship of the changing trend refers to the orderly transmission and gradual enhancement and weakening of the oxygen concentration change trend at the measuring point in space.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire sampling sequences from multiple oxygen sensors at the same height cross-section of the boiler furnace, perform time alignment processing on the oxygen concentration values ​​at the measuring points and arrange them according to a uniform sampling period to generate an oxygen concentration time series matrix. S102: Based on the oxygen concentration time series matrix, perform differential operation on the oxygen concentration values ​​at adjacent sampling times of the measuring points, calculate the weighted dynamic difference value of oxygen concentration, calculate the concentration change corresponding to the time interval and arrange them in time order to obtain the oxygen concentration difference sequence set. S103: Based on the oxygen concentration difference sequence set, and combined with the spatial coordinates of the measuring points, perform lateral mapping processing, sort the difference values ​​according to the cross-sectional position, construct a two-dimensional grid relationship, and generate a lateral variation distribution map.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Obtain the lateral change distribution map, extract the change value sequence corresponding to the measurement point location, construct measurement point pairs according to spatial adjacency, and perform difference comparison and change direction determination on the change values ​​of adjacent measurement points to obtain a set of measurement point change trend pairs; S202: Based on the set of measurement point change trends and combined with the preset change trend judgment benchmark, perform point-by-point comparison on the relationship between the consistency of the change direction and the change amplitude of the measurement points within the continuous sampling period, filter the measurement point positions that meet the judgment benchmark, and obtain the set of trend-matching measurement points. S203: Based on the trend matching measurement point set, combined with the spatial distribution relationship of the measurement points, perform regional aggregation processing, determine the connectivity of adjacent measurement points that meet the conditions, establish a regional mapping structure, and generate a combustion initiation area identification map.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Obtain the combustion initiation area identification map, extract the position coordinates and change trend sequence of the measuring points within the identification area, construct a set of candidate connecting measuring points based on spatial adjacency, and generate a set of candidate paths for measuring points; S302: Based on the candidate set of measurement point paths, perform continuous comparison on the changing trends of adjacent measurement points in the candidate paths, determine the progressive relationship of the trend direction and the decreasing relationship of the change magnitude, filter the path combinations that satisfy the progressive relationship, and obtain the set of trend progressive paths. S303: Based on the aforementioned trend progression path set, and combined with the spatial distribution order of the measuring points, perform path connection integration processing, sequentially associate the path nodes, construct the path topology structure, and generate a combustion diffusion trajectory map.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Obtain the combustion diffusion trajectory map, extract the coordinates of the measuring points and the path order index in the trajectory, and perform serialization and arrangement of the measuring points according to the trajectory connection order to generate a measuring point trajectory sequence set; S402: Based on the measured point trajectory sequence set, perform a filtering process on the measured point positions in the sequence, remove discontinuous nodes and retain continuous node combinations according to the preset path continuity benchmark, and obtain a continuous measured point sequence set; S403: Based on the continuous measurement point sequence set, perform hierarchical division processing according to the sequential relationship of the trajectory, establish a forward and backward hierarchical mapping relationship for the measurement point positions and form a hierarchical structure, and generate a measurement point start and stop distribution map.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Obtain the start-stop distribution map of the measuring points, extract the start-stop status and oxygen concentration change sequence corresponding to the measuring point location, perform permutation and combination processing on the oxygen concentration values ​​of the measuring points according to the time sequence, and generate a set of measuring point time sequence combinations. S502: Based on the combined set of time series of measuring points, perform time-by-time comparisons on the magnitude and direction of oxygen concentration changes in the measuring point sequence, and perform classification judgments on the sequence according to the preset state judgment criteria to obtain a set of measuring point state judgments. S503: Based on the set of measurement point status determinations, and combined with the start-stop relationship of the measurement points, perform correlation integration processing, establish a time-series mapping structure for the measurement point status results and form an overall distribution relationship, and generate a smart monitoring scheme for the gas turbine unit status.

[0013] An intelligent monitoring system for assessing the condition of gas turbine units includes: The horizontal oxygen concentration trend analysis module is used to achieve S1: by arranging multiple oxygen sensors at the same height cross section of the boiler furnace to collect continuous time series oxygen concentration change information, perform differential calculation on the oxygen concentration change information at adjacent sampling times, calculate the change trend of the measuring points according to the time sequence, and generate a horizontal change distribution map. The combustion initiation zone identification module is used to implement S2: compare and analyze the lateral change distribution map, collect the change trend of adjacent measuring points for point-by-point comparison, determine the position of measuring points based on the magnitude of the change trend of continuous sampling period, filter the measuring point positions that meet the conditions, and generate a combustion initiation zone identification map. The combustion diffusion trajectory derivation module is used to implement S3: based on the combustion initiation area identification map, derive the lateral measuring point path, collect the changing trend of the measuring points for continuous comparison, connect the measuring point positions according to the progressive relationship of the changing trend, analyze the connection results, and generate a combustion diffusion trajectory map; The measuring point layering and start / stop determination module is used to implement S4: segmenting the combustion diffusion trajectory map, collecting the measuring point positions in the combustion diffusion trajectory map and sorting them according to the trajectory order, filtering the sorting results and dividing the measuring point layers, and generating a measuring point start / stop distribution map. The intelligent monitoring module for unit status is used to implement S5: construct monitoring content for the start-stop distribution map of the measuring points, collect oxygen concentration change information of the measuring points in the start-stop distribution map of the measuring points and combine them in time order, judge the combination result, and generate an intelligent monitoring scheme for the gas turbine unit status.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a lateral change gradient is constructed based on the differential evolution process of multi-sensor sequences. The initial combustion area and propagation path nodes are defined through multi-point time-series comparison. Then, the acquisition level and monitoring wake-up sequence are dynamically planned to break the judgment barrier of the preset threshold of inherent static parameters. A dynamic state deduction mechanism that adapts to the spatiotemporal evolution law is constructed to eliminate the risk of misjudgment of operating conditions caused by a single isolated feature. Based on the combustion evolution characteristics, early signs of parameter anomalies are captured in advance. A panoramic portrait of the unit's operating condition evolution with strong mapping to the physical process is established. The outdated manual experience review and static rule benchmarking are abandoned to achieve a comprehensive tracking from parameter anomalies to overall state deviation, thereby improving the foresight of hidden danger discovery and the effectiveness of monitoring. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides an intelligent monitoring method for assessing the condition of gas turbine units, comprising the following steps: S1: Oxygen concentration change information in a continuous time series is collected by multiple oxygen sensors arranged at the same height cross section of the boiler furnace, and differential calculation is performed on the oxygen concentration change information at adjacent sampling times. The change trend of the measuring points is calculated and processed according to the time sequence to generate a horizontal change distribution map. S2: Compare and analyze the horizontal variation distribution map, collect the variation trend corresponding to adjacent measuring points and compare them point by point, judge the position of measuring points according to the preset variation trend size law within the continuous sampling period, filter the measuring point positions that meet the variation trend conditions, and generate a combustion start area identification map. S3: Based on the combustion initiation zone identification map, the path of the transverse measuring points is deduced, the changing trend of the measuring points is collected and continuously compared, the measuring point positions are connected according to the progressive relationship of the changing trend, the connection results are analyzed and processed, and a combustion diffusion trajectory map is generated. S4: Divide the combustion diffusion trajectory map into segments, collect the measurement point positions in the combustion diffusion trajectory map and sort them according to the trajectory order, filter the sorting results, divide the measurement point levels according to the distribution relationship before and after the trajectory, and generate a measurement point start and stop distribution map. S5: Construct monitoring content for the start-stop distribution map of measuring points, collect oxygen concentration change information of corresponding measuring points in the start-stop distribution map and combine them in chronological order, judge and process the combination results, and generate a smart monitoring scheme for the gas turbine unit status.

[0020] The lateral variation distribution map includes the cross-sectional oxygen gradient field, the differential amplitude grid of measuring points, and the trend variation thermogram. The combustion initiation area identification map includes the initial ignition point coordinate set, the boundary of the violent oxidation reaction, and the geometric features of the combustion pilot zone. The combustion diffusion trajectory map includes the flame spread spatial connection line, the oxygen mutation propagation route, and the combustion zone expansion vector network. The measuring point start-stop distribution map includes the sensor activation priority sequence, the measuring point monitoring depth echelon, and the data acquisition dormancy interval. The intelligent monitoring scheme for gas turbine unit status includes the furnace combustion condition diagnosis panel, the oxygen abnormality alarm threshold setting, and the dynamic unit load adaptation strategy.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire sampling sequences from multiple oxygen sensors at the same height cross-section of the boiler furnace, perform time alignment processing on the oxygen concentration values ​​at the measuring points and arrange them according to a uniform sampling period to generate an oxygen concentration time series matrix. Polling data was collected from 16 zirconia oxygen sensors deployed at the same cross-section at an elevation of 45m in the boiler furnace. The probes of each oxygen sensor were positioned in the high-temperature region inside the furnace, and their signals were led out via high-temperature resistant leads to a signal processing unit in the low-temperature region outside the furnace. In the signal processing unit, analog-to-digital conversion was performed to generate a digital signal, which was then transmitted and acquired via an RS485 communication bus at a baud rate of 9600bps. To ensure signal transmission stability, the signal processing unit was located in the low-temperature environment outside the furnace to avoid the influence of high temperatures on electronic components and communication lines. The initial electrical signals acquired were converted from analog to digital and mapped to the oxygen concentration dimensional space ranging from 0% to 21%. For each sensor's sampling data stream, a baseline sampling period of 1000ms is set to acquire macroscopic trend information during combustion. Due to the thermal inertia of the furnace combustion process, a second-level sampling period is sufficient for unit operation monitoring and anomaly early warning. For rapid transient changes in combustion, the sampling frequency can be increased or a high-frequency detection device can be introduced for auxiliary analysis, but this is not a necessary limitation of this implementation method. 3600 consecutive sampling periods are extracted to form an initial one-dimensional sampling sequence. Data cleaning is performed on the extracted 16 initial sampling sequences, using a median filtering algorithm to eliminate impulse noise interference. Specifically, the filtering window length is set to 5. The third value is extracted after arranging the five consecutive sampled values ​​within the window in ascending order as the effective oxygen concentration value at the current moment. To address the timestamp offset problem caused by transmission delay, the local clock pulse of measuring point 1 is used as the global synchronization reference, and the microsecond-level deviation of the sampling timestamps of the remaining 15 measuring points relative to the baseline timestamp is calculated. When the deviation value exceeds 50ms, a linear interpolation alignment algorithm is initiated. This algorithm calls the coordinates of adjacent forward and backward valid sampling points, assigning weights based on the time deviation ratio to reconstruct the oxygen concentration value at the current alignment moment. A sequence of 16 measurement points with fully synchronized timestamps and a uniform sampling interval of 1000ms is row-sorted from 1 to 16 according to the physical clockwise installation number of the sensors on the furnace cross-section. The 3600 discrete sampling time points are then column-sorted. A contiguous two-dimensional floating-point array is allocated in memory, and finally, a 16-row, 3600-column oxygen concentration time series matrix is ​​written and locked. Each element in the matrix is ​​precisely located to a specific measurement point and the absolute sampling second.

[0022] S102: Based on the oxygen concentration time series matrix, perform a difference operation on the oxygen concentration values ​​at adjacent sampling times of the measuring points, using the following formula: ; Calculate the weighted dynamic difference value of oxygen concentration, calculate the concentration change corresponding to the time interval and arrange them in chronological order to obtain the oxygen concentration difference sequence set; in, Representing the Each measuring point The weighted oxygen concentration dynamic difference at time points. The first element in the oxygen concentration time series matrix Each measuring point The oxygen concentration value at that time. The first element in the oxygen concentration time series matrix Each measuring point The oxygen concentration value at that time. Represents the differential gain coefficient between adjacent time points. Represents the total number of historical backtracking steps. The sequence number represents the summation association of historical backtracking steps. The first element in the time series matrix representing oxygen concentration Each measuring point The oxygen concentration value at that time. Representing the A decaying weight factor for each backtracking step size, Representing the The baseline static differential bias concentration threshold for each measuring point; For the constructed oxygen concentration time series matrix with dimensions of 16 rows and 3600 columns, a difference extraction operation is initiated for data from adjacent sampling times. The oxygen concentration value at time 100 (the first measurement point in the first row of the matrix) is extracted to be 4.15%, and the oxygen concentration value of the previous sample at time 99 (the previous measurement point) is extracted to be 4.10%. Before performing formula calculations, the specific numerical settings and physical quantification procedures for each parameter need to be obtained. Parameters The data was obtained by collecting 50 sets of transient fluctuation data under standard steady-state combustion conditions beforehand, and then conducting comparative tests within the range of 0.5 to 2.0 using a grid search method. When set to 1.2, it can maximize the suppression of background noise and highlight effective concentration fluctuations of over 0.05%; Parameter Setting it to 5 represents tracing back 5 consecutive 1000ms sampling periods; parameter Quantified according to the exponential decay model, the calculation method is as follows: Therefore, the decay weighting factors for the 1st to 5th backtracking steps are 0.8, 0.64, 0.512, 0.4096, and 0.3276, respectively; parameters By collecting 24 hours of zero-drift data at this measuring point under unit shutdown and cooling conditions, the arithmetic mean was taken and an additional term was added. Substituting this with 4.15% and 4.10%, the basic absolute difference value was calculated to be 0.06%. Subsequently, historical oxygen concentration values ​​from the 95th to the 99th second (4.12%, 4.13%, 4.11%, 4.09%, and 4.10%, respectively) were retrieved, and the squared difference and square root of the difference were calculated with the 4.15% value at the 100th second. The results were then multiplied sequentially by the corresponding index. The numerical values, summed and divided by 5, yield a historical fluctuation weighted term of 0.028%. Adding the gain term of 0.06% to the historical fluctuation term of 0.028%, and subtracting the bias term of 0.02%, the final result is 0.0596472%. This result indicates that the first measuring point experienced a slight but definite oxygen concentration transient at time 100, effectively eliminating sensor static drift and high-frequency random noise. The advantage of this formula lies in its ability to effectively smooth out pseudo-difference spikes caused by single-point mutations by introducing a decay weight factor for the historical backtracking step. By iteratively performing the above substitution operation on the 16 measuring points and 3600 time points in the matrix, a 16-row, 3600-column numerical set with the same dimensions as the original matrix is ​​generated. The 16 weighted oxygen concentration dynamic difference values ​​corresponding to each time point are extracted and encapsulated and combined in ascending order from second 1 to second 3600 to obtain an oxygen concentration difference sequence set containing the complete differential evolution process.

[0023] S103: Based on the oxygen concentration difference sequence set, combined with the spatial coordinates of the measuring points, perform lateral mapping processing, sort the difference values ​​according to the cross-sectional position and construct a two-dimensional grid relationship to generate a lateral variation distribution map; Sixteen valid difference values ​​corresponding to the 100th second of the aforementioned 16 rows and 3600 columns of oxygen concentration difference sequence are extracted. Simultaneously, the three-dimensional Cartesian coordinate data of 16 measuring points pre-stored in a relational database are retrieved, removing the vertical Z-axis coordinates and retaining the X and Y-axis planar coordinate pairs. A 10m x 10m reference two-dimensional plane system is established with the geometric center of the furnace cross-section as the origin. This plane system is divided into a 128 x 128 pixel matrix using a mesh generation algorithm, and the row and column indices of the 16 measuring points in the pixel matrix are calculated. The 16 discrete difference values ​​are filled into the corresponding pixel indices. For the 128 x 128 grid blank nodes without sensors, a bicubic interpolation algorithm is initiated to perform pixel-level smooth filling. Specifically, the algorithm calls the difference values ​​of the 16 nearest known measuring points around the blank node, constructs a cubic polynomial surface function, and calculates the partial derivatives and cross derivatives in the X and Y directions respectively to obtain the mapping interpolation for each grid node. The 16,384 grid nodes with interpolated values ​​were subjected to bubble sort in row-first order, and a node position mapping table was established based on the descending order of the difference values. For each node in the pixel matrix, a corresponding pseudo-color RGB channel value was assigned according to its calculated difference value. The difference value of 0.00% was set to correspond to dark blue (RGB: 0, 0, 255), and the maximum difference value of 0.15% was set to correspond to dark red (RGB: 255, 0, 0). The intermediate values ​​were mapped using a linear color gradient. Finally, a horizontal distribution map with a resolution of 128 by 128 pixels was generated, which intuitively reflects the degree of drastic change in oxygen content in different areas of the same cross section.

[0024] Please see Figure 3 The specific steps of S2 are as follows: S201: Obtain the horizontal change distribution map, extract the change value sequence corresponding to the measurement point location, construct measurement point pairs according to spatial adjacency, and perform difference comparison and change direction determination on the change values ​​of adjacent measurement points to obtain the measurement point change trend pair set; Using a 128×128 pixel horizontal variation distribution map in memory as input, the gray-quantitative change sequences of the grid nodes containing 16 physical measurement points are extracted according to a preset coordinate matching dictionary. Based on the plane X-axis and Y-axis coordinates of the measurement points, the Euclidean distance between points is calculated. A spatial proximity threshold of 3.5 meters is set, and the spatial distance between any measurement point and the other 15 measurement points is calculated. Combinations with distances less than or equal to 3.5 meters are retained, thus constructing an undirected connected graph and generating 42 pairs of measurement points that satisfy spatial adjacency. For adjacent measurement point pairs labeled "Measurement Point 1 - Measurement Point 2", the change value of measurement point 1 (0.068%) and the change value of measurement point 2 (0.045%) at the current time are extracted, and a subtraction operation is performed to obtain a difference of 0.023%. The sign bit of measurement point 1 at the current time compared to the previous time is positive (representing an increase), and the sign bit of measurement point 2 is also positive, determining that the change directions of both are increasing in the same direction. Based on the difference calculation results and direction determination results, a quadruple data structure containing "node A, node B, absolute value of difference, and direction identifier (1 indicates same direction, -1 indicates opposite direction, 0 indicates one moving and one stationary)" is established for each pair of measurement points. All 42 pairs of measurement points are traversed to complete the quadruple encapsulation. Finally, a set of 42 spatially adjacent measurement point change trend pairs containing quantified difference characteristics is generated in the memory list.

[0025] S202: Based on the set of measurement point change trends and combined with the preset change trend judgment benchmark, perform point-by-point comparison on the relationship between the consistency of the change direction and the change amplitude of measurement points within a continuous sampling period, screen the measurement point positions that meet the judgment benchmark, and obtain the set of trend-matching measurement points. Based on this, it is further determined that the absolute value of oxygen concentration at the measuring point is at a relatively high level and shows a continuous downward trend. Combined with the spatial location of the measuring point, it is determined whether it is located in the frontier region of change and diffusion. Measuring point locations that meet the above conditions are selected to obtain a set of trend-matching measuring points. For a set of measurement point change trends containing 42 sets of quadruples, a deep comparison was performed on datasets from five consecutive 1000ms sampling periods. The directional identifiers of the first quadruple, "Measurement Point 1 - Measurement Point 2," were extracted within these five periods, resulting in the sequence [1, 1, 1, 1, 1], indicating that the directional consistency met the preset benchmark of five consecutive unidirectional measurements. The absolute value sequence of the difference between these measurement points within these five periods was further retrieved: [0.023%, 0.021%, 0.024%, 0.020%, 0.022%]. A deviation threshold of 0.015% was set for the allowable change magnitude. The difference between the maximum and minimum values ​​in this sequence was calculated: 0.024% minus 0.020% equals 0.004%. Since 0.004% is strictly less than the preset benchmark of 0.015%, the measurement point pair was deemed to have passed the change magnitude relationship review. To verify the effectiveness of the preset benchmark, comparative tests were conducted at different allowable deviation thresholds.

[0026] Table 1 Comparison of Test Results for Deviation Allowable Threshold Setting 0.005 12 24 78 0.010 45 15 89 0.015 88 4 97 As shown in Table 1, the consistency and amplitude deviation comparison calculations were performed on the remaining 41 pairs of measuring points in the set one by one. Pairs of measuring points containing opposite direction identifiers or with a range greater than 0.015% were removed, ultimately selecting 28 pairs of measuring point combinations that fully met the judgment criteria. All non-repeating independent measuring point numbers involved in these 28 pairs were extracted, and a trend-matching measuring point set containing 11 valid measuring point identifiers was generated in the array space.

[0027] S203: Based on the trend matching measurement point set, combined with the spatial distribution relationship of the measurement points, perform regional aggregation processing, determine the connectivity of adjacent measurement points that meet the conditions, establish a regional mapping structure, and generate a combustion initiation area identification map; After completing the division of the connected regions, based on the absolute level and trend of oxygen concentration at the measuring points in each connected region, the connected regions with relatively high oxygen concentration and a continuous decreasing trend, and located at the spatial diffusion front, are selected first and marked as combustion initiation regions, generating a combustion initiation region identification map. For a trend-matching measurement point set containing 11 measurement point markers, extract its planar position parameters in a 10×10 meter reference coordinate system. Initialize a two-dimensional binary matrix that maps to the physical space, setting the coordinate positions of the 11 measurement points to 1 and the rest to 0. Start a seed-filling algorithm based on eight-neighborhood search, randomly selecting any coordinate point in the set as the starting seed node and pushing it onto the stack. Pop the top node from the stack and check its eight adjacent coordinate points. If the value of an adjacent point in the two-dimensional binary matrix is ​​1 and it has not been visited, calculate the range of the change values ​​between the adjacent point and the currently popped node, and determine whether the range is less than the connectivity threshold of 0.05%. If the range is less than 0.05%, mark the two points as connected, establish an undirected edge relationship, and push the adjacent point onto the stack. Repeat the above push and pop operations until the stack is completely empty. At this point, all the traversed nodes form a connected subgraph. If there are still unvisited nodes with a value of 1 in the 2D matrix, a new seed node is selected for a new round of connected subgraph extraction. After all searches are completed, the connected subgraph with the most nodes (e.g., containing 8 measurement points) is marked as the core active region, and the circumscribed polygon contour of these 8 outer coordinate points is extracted using the GrahamScan algorithm. Since the oxygen concentration in the region corresponding to the initial stage of combustion usually decreases from a higher level, while the oxygen concentration in the vigorous combustion region has decreased significantly, it is easy to misjudge the region of intense combustion reaction as the starting region based solely on the difference in oxygen concentration. Therefore, the absolute level of oxygen concentration and the spatial front position constraint are introduced in the judgment process to distinguish between the combustion starting region and the region of intense combustion reaction, thereby improving the physical rationality and accuracy of the identification results. Specific combustion attribute data structures are assigned to the internal regions of the extracted polygons, and red marker channel values ​​representing intense oxidation reactions are filled in. These are then output in the graphics rendering interface, and finally, a combustion starting region identification map with clear boundaries and a well-defined geographical coordinate range is generated on the screen terminal.

[0028] Please see Figure 4 The specific steps of S3 are as follows: S301: Obtain the combustion initiation area identification map, extract the position coordinates and change trend sequence of measuring points within the identification area, construct a candidate set of measuring point connections based on spatial adjacency, and generate a candidate set of measuring point paths; The system receives the combustion initiation area identification map data matrix from the terminal's video memory and extracts the numbers of eight core measuring points and their X and Y coordinates within the polygonal region. It synchronously retrieves the dynamic differential change values ​​of these eight measuring points over the past 60 seconds from the global historical data buffer and constructs a 60-bit floating-point trend sequence with a time resolution of once per second. The Delaunay triangulation algorithm is used to spatially mesh the coordinates of these eight measuring points, generating a mesh topology composed of non-intersecting triangles. Using any measuring point as a vertex, directed line segments are extracted along the edges of the triangles to connect with other adjacent measuring points, and the Euclidean distance of each edge is calculated. Longer edges than 4.0 meters are removed, and the remaining edge relationships are retained as traversable edges. The measuring point closest to the furnace center (e.g., measuring point 5 with coordinates X:5.0, Y:5.0) is used as the unique root node for the search, and a depth-first search (DFS) traversal algorithm is initiated. The maximum search depth is set to 4 layers. Adjacent nodes are traversed sequentially downwards along the previously preserved traversable edges. During the search, the sequence of node numbers visited is recorded. When no unvisited adjacent nodes are reached or the maximum depth is reached, the recorded node sequence (e.g., "measurement point 5 to measurement point 7 to measurement point 9 to measurement point 2") is packaged into a complete path chain. This search strategy is repeated until all possible branch structures are explored. The resulting 24 unique path chains are then summarized and stored in a dynamic array list, generating a candidate set of measurement point paths for subsequent evaluation and selection.

[0029] S302: Based on the candidate set of measurement point paths, perform continuous comparison on the changing trends of adjacent measurement points in the candidate paths, determine the progressive relationship of the trend direction and the decreasing relationship of the change magnitude, filter the path combinations that satisfy the progressive relationship, and obtain the set of trend progressive paths. The 24 path chains stored in the candidate path set for measuring points are traversed, and the first candidate path, "Measuring Point 5 to Measuring Point 7 to Measuring Point 9," is extracted. For this path chain, the difference changes of its three nodes at the current time are extracted: 0.082% for measuring point 5, 0.065% for measuring point 7, and 0.041% for measuring point 9. In the directional progression determination stage, the distances between measuring point 5 and the central region are calculated to be 1.2 meters, 3.5 meters for measuring point 7, and 6.1 meters for measuring point 9. The distance values ​​show a strictly increasing trend, confirming that this path conforms to the physical trend of spatial diffusion from the inside out. In the decrease in the magnitude of change determination stage, the difference is subtracted: 0.082% minus 0.065% equals 0.017%, and 0.065% minus 0.041% equals 0.024%. The critical value that both of these difference results are greater than 0 indicates that as the diffusion distance increases, the fluctuation amplitude of the oxygen concentration difference shows a strictly monotonically decreasing relationship. To eliminate false diffusion caused by turbulent flow, a lower limit of 10% for amplitude attenuation rate was set. Calculations showed that (0.082%-0.065%) / 0.082%=20.7% and (0.065%-0.041%) / 0.065%=36.9%, both meeting the lower limit requirement. Candidate paths meeting both spatial progression and amplitude reduction conditions were marked as valid. Distance calculation, difference calculation, and attenuation rate percentage judgment were iteratively performed on the remaining 23 paths, eliminating 18 paths exhibiting amplitude reversal (e.g., subsequent node changes exceeding those of preceding nodes) or spatial reversals. The remaining 6 paths that passed all calculations were extracted, memory space was reallocated, and a set of trend-progressing paths with strict physical diffusion logic was generated.

[0030] S303: Based on a trend-progressive path set, combined with the spatial distribution of measuring points, path connection integration processing is performed to sequentially associate path nodes and construct a path topology to generate a combustion diffusion trajectory map. The algorithm calls a set of trend progression paths containing 6 path chains and initializes a Directed Acyclic Graph (DAG) data structure in the graphics memory. It iterates through each measurement point element in these 6 paths, instantiating it as an independent node in the DAG. The sequential relationship between two adjacent measurement points in each path is instantiated as a connecting edge with a direction attribute (e.g., a directed edge pointing from measurement point 5 to measurement point 7). For measurement points and connecting edges that appear repeatedly in different paths, a deduplication and merging operation is performed, and the weight value of each edge is calculated. The weight value is calculated based on the arithmetic mean of the changes in the values ​​of the two endpoints; for example, the average value of measurement points 5 and 7 is (0.082% + 0.065%) / 2 = 0.0735%. A topological sorting algorithm is used to calculate the hierarchy of nodes in the entire DAG. Nodes with an in-degree of 0 are assigned to the first level, and the in-degree is decremented by 1 with each forward pass, sequentially dividing the graph into different diffusion levels. In the topology, a variant of Dijkstra's shortest path algorithm is used, modifying the search for "shortest distance" to find "maximum cumulative weight," thereby calculating the dominant diffusion artery from the initial level node to the final level node. Along the direction of the dominant artery, nodes at different levels are given a dynamic gradient color from yellow to dark red in the graphics rendering pipeline, and Bézier curve drawing instructions with arrow markers and width gradients are added to the directed connection edges. All drawn primitives are output to the display frame buffer, and finally, a combustion diffusion trajectory diagram clearly showing the oxygen fluctuation and spread process is presented on the human-computer interaction interface.

[0031] Please see Figure 5 The specific steps of S4 are as follows: S401: Obtain the combustion diffusion trajectory map, extract the coordinates of the measuring points and the path order index in the trajectory, perform serialization and arrangement of the measuring points according to the trajectory connection order, and generate a set of measuring point trajectory sequences; The underlying topology data of the combustion diffusion trajectory diagram is read back from the rendering pipeline of the human-computer interaction interface, and the core path object marked as the "dominant diffusion artery" is extracted. This path object is parsed, and the sequence of measurement point identifiers (e.g., measurement point 5, measurement point 7, measurement point 9, measurement point 12) is extracted according to the direction of the directed edges. The system's static configuration table is queried using the measurement point identifiers to extract the three-dimensional coordinates (X, Y, Z) of these four measurement points in the furnace physical coordinate system. An array of structures is created, containing the fields: "index number", "measurement point ID", "X coordinate", and "Y coordinate". The index number of the starting measurement point 5 is assigned as 1. For each node advanced along the path connection edge, the index number is incremented by 1, assigning index number 2 to measurement point 7, index number 3 to measurement point 9, and index number 4 to measurement point 12. For cases involving multiple branch trajectories, a depth-first traversal sequence is used to record the descending order. A decimal-level hierarchical method is employed to assign sub-indices to branch nodes (e.g., if measurement point 9 branches downwards to measurement point 10, then the index number of measurement point 10 is 3.1). In memory, the QuickSort algorithm is used to sort all structures in ascending order according to their index numbers, completing the transformation from spatial graphical topology to a one-dimensional linear data structure. Finally, the sorted array of structures is serialized and written to a local disk file, generating a measurement point trajectory sequence set with strict temporal and spatial correlation.

[0032] S402: Based on the measurement point trajectory sequence set, perform filtering processing on the measurement point positions in the sequence, remove discontinuous nodes and retain continuous node combinations according to the preset path continuity benchmark, and obtain a continuous measurement point sequence set; The system loads the measurement point trajectory sequence set from the local disk and sequentially reads the measurement point identifiers and coordinate information from the structure array. It extracts the physical coordinate data corresponding to two adjacent index numbers (e.g., measurement point 5 at index 1 and measurement point 7 at index 2), calls the spatial distance calculation function between the two points, and finds the actual physical distance to be 2.8 meters. It compares this to a preset path spatial continuity benchmark value, set at 3.0 meters. Since 2.8 meters is less than 3.0 meters, it is determined that there is no spatial discontinuity between the two points. Simultaneously, it extracts the timestamps of the oxygen concentration difference peaks at measurement points 5 and 7. The peak time for measurement point 5 is at 102 seconds, and for measurement point 7 it is at 104 seconds, a time difference of 2 seconds. It compares this to a preset path time continuity benchmark value, setting the maximum tolerable delay to 3 seconds. Since 2 seconds is less than 3 seconds, the time transmission is considered continuous. If the spatial distance between adjacent nodes (such as index 3 and index 4) is 4.5 meters, which is greater than the spatial reference of 3.0 meters, or the time difference is 5 seconds, which is greater than the time reference of 3 seconds, then all nodes associated with index 4 and thereafter are marked as discontinuous nodes. A complete forward traversal is performed on the entire sequence set. The array removal instruction (such as Array.splice) is used to permanently delete the structure data marked as discontinuous nodes from the memory list. All node combinations that satisfy the condition of spatial span less than 3.0 meters and time delay less than 3 seconds are retained. The clean data array after removal is repackaged and backed up to obtain a continuous measurement point sequence set that has eliminated interference and miscellaneous items.

[0033] S403: Based on a continuous set of measurement points, hierarchical division is performed according to the sequential relationship of the trajectory. A forward and backward hierarchical mapping relationship is established for the measurement point positions to form a hierarchical structure and generate a measurement point start and stop distribution map. Using a continuous sequence of measurement points as input, the measurement point corresponding to the first index in the sequence is extracted as the level 0 reference point. For the remaining nodes in the sequence, a forward and backward hierarchical mapping table is established based on the order of timestamps. Measurement points with timestamps 0 to 2 seconds later than the reference point are mapped to the first forward level, and those 2 to 4 seconds later are mapped to the second forward level. For the node group whose time division is complete, the corresponding device status register is checked. If the oxygen concentration fluctuation associated with the current measurement point is higher than 0.08%, the device virtual status flag is set to 1, representing an "on" state; if the fluctuation drops and remains below 0.02% for more than 10 seconds, the flag is set to 0, representing a "off" state. A blank hierarchical view is created, and the level 0 reference point icon is drawn centered at the top of the view. The icon is filled with color according to the status register value: 1 is rendered as green, and 0 as gray. Below this, draw all the measurement point icons of the first forward level at equal horizontal intervals and fill them with color according to their values. Connect the icons of the upper and lower levels with solid straight lines. Draw the nodes of lower levels in a loop, constructing a hierarchical graphic structure similar to a tree directory. Output the set of primitives with all coordinate positions calculated and color attributes attached as a vector graphic file, generating a measurement point start and stop distribution map that intuitively reflects the state transition relationship of each spatial node over time.

[0034] Please see Figure 6 The specific steps of S5 are as follows: S501: Obtain the start-stop distribution map of the measuring points, extract the start-stop status and oxygen concentration change sequence corresponding to the measuring point location, perform permutation and combination processing on the oxygen concentration values ​​of the measuring points according to the time sequence, and generate a set of measuring point time series combinations. Parse the start / stop distribution map data of the measurement points in the vector graphics file. Based on the primitive attribute table, extract the binary start / stop state sequence (a one-dimensional array of 0s and 1s) of 16 physical measurement points over the past 120 consecutive sampling periods (i.e., 120 seconds) and the corresponding floating-point absolute oxygen concentration change sequence. Establish a multi-dimensional correlation matrix in the cache, where the row index corresponds to measurement point numbers 1 to 16. For measurement point number 1, extract its state sequence (e.g., 1 for the first 60 seconds and 0 for the next 60 seconds) and oxygen concentration change sequence (containing 120 floating-point data points). Start a time-axis aligned cursor, setting the cursor step size to 1 second, and traverse from second 1 to second 120. Under the time scale indicated by the cursor, extract a comprehensive data packet containing "measurement point identifier, current absolute second count, binary start / stop state value, and floating-point oxygen concentration change value". The 120 integrated data packets from measurement point 1 are concatenated into a long vector in chronological order. This same truncation and concatenation operation is then repeated for the remaining 15 measurement points, resulting in 16 long vectors. These 16 long vectors are then stacked horizontally and combined in memory to generate a two-dimensional array of structures. This structure perfectly encapsulates the data features of the spatial dimension (different measurement points), the temporal dimension (120-second time series), and the physical dimension (state and concentration values), ultimately solidifying into a time-series combination set of measurement points for use by the backend diagnostic logic.

[0035] S502: Based on the time series combination set of measurement points, perform time-by-time comparison of the magnitude and direction of oxygen concentration change in the measurement point sequence, and perform classification judgment on the sequence according to the preset state judgment criteria to obtain the measurement point state judgment set. Load the combined time series data of the measuring points and start the time-by-time classification engine with multiple conditions and branches. Extract the comprehensive data packet of measuring point 1 at 60 seconds, and obtain its oxygen concentration change amplitude value of 0.12%, with the calculated change direction being negative (i.e., the absolute value of oxygen concentration decreases). Call the preset combustion state judgment classification rule tree: the first layer judges whether the change amplitude is greater than the lower limit threshold of 0.10% for violent fluctuations. Since 0.12% is greater than 0.10%, it enters the active judgment branch; the second layer judges the change direction. A negative direction indicates that oxygen is being rapidly consumed, entering the oxygen-deficient combustion branch; the third layer combines its start-stop status value. If the current status value is 1 (start), the final judgment output is the status label "local combustion is extremely active and air supply is insufficient" (the label code is set to 3). If another measuring point has a change amplitude of 0.01% at a certain moment (less than the stable threshold of 0.03%) and a status value of 0, the judgment output is the label "stable / inactive combustion" (the label code is set to 0). To verify the reliability of the classification rule threshold, historical labeled samples were collected for state classification tests.

[0036] Table 2 Test Data Table for State Judgment Classification Rules 0.05 82 68 35 0.08 91 85 18 0.10 98 96 5 As shown in Table 2, for 120 time points of 16 measurement points, a total of 1920 data packets are processed in a loop. The three-level classification rule tree judgment is executed, and the calculated status label code (0, 1, 2, 3, etc.) is added to each data packet. All labeled data dictionaries are repackaged, and a measurement point status judgment set with discrete classification characteristics is generated in the database.

[0037] S503: Based on the set of measurement point status judgments, combined with the start-stop relationship of measurement points, perform correlation and integration processing, establish a time-series mapping structure for the measurement point status results and form an overall distribution relationship, and generate a smart monitoring scheme for the gas turbine unit status. The complete set of measurement point status judgments is extracted from the database. A data cleaning script removes all redundant data items with a status marker code of 0 (i.e., stable / inactive combustion), retaining anomalous or highly active data points with marker codes 1 to 3. A time-series hash table is built according to the chronological order of timestamps, using the absolute number of seconds as the key, and storing the measurement point number and marker code of the anomalous state occurring at the same time as the value. Measurement point clusters with consecutive occurrences of the same type of anomalous state on the timeline are extracted. For example, measurement points 4 and 5 are found to have a status marker code of 3 (extremely active local combustion and insufficient air supply) between seconds 80 and 100. Furthermore, a query using the start-stop relationship table shows that both points are in the "on" state and at the same physical diffusion level. Based on the clustering characteristics and temporal continuity of this area, an intelligent adjustment strategy script is triggered. The burner secondary damper control object interface corresponding to this measurement point cluster is extracted, and the compensation air volume value is calculated: the basic compensation opening is increased by 5%, and a penalty coefficient of 1.5 corresponding to status code 3 is added, resulting in a final damper opening adjustment of 7.5%. The regional location information (furnace cross-section areas 4 to 5), status diagnosis results (extremely active and insufficient air supply), and quantitative operation suggestions (corresponding to a 7.5% increase in secondary air damper opening) are packaged into a standardized instruction set in JSON format. This instruction set is then pushed to the operator station display panel of the DCS (Distributed Control System) via an API gateway and written to the historical operation suggestion log table, ultimately generating a smart monitoring solution for the gas turbine unit status that integrates spatiotemporal analysis, quantitative diagnosis, and precise control intervention.

[0038] An intelligent monitoring system for assessing the condition of gas turbine units includes: The horizontal oxygen concentration trend analysis module is used to achieve S1: by arranging multiple oxygen sensors at the same height cross section of the boiler furnace to collect continuous time series oxygen concentration change information, perform differential calculation on the oxygen concentration change information at adjacent sampling times, calculate the change trend of the measuring points according to the time sequence, and generate a horizontal change distribution map. The combustion initiation zone identification module is used to implement S2: compare and analyze the lateral change distribution map, collect the change trend of adjacent measuring points for point-by-point comparison, determine the position of measuring points based on the magnitude of the change trend of continuous sampling period, filter the measuring point positions that meet the conditions, and generate a combustion initiation zone identification map. The combustion diffusion trajectory derivation module is used to implement S3: derive the lateral measuring point path based on the combustion initiation area identification map, collect the changing trend of the measuring points for continuous comparison, connect the measuring point positions according to the progressive relationship of the changing trend, analyze the connection results, and generate a combustion diffusion trajectory map; The measurement point layering and start / stop determination module is used to implement S4: segmenting the combustion diffusion trajectory map, collecting the measurement point positions in the combustion diffusion trajectory map and sorting them according to the trajectory order, filtering the sorting results and dividing the measurement point layers, and generating a measurement point start / stop distribution map. The intelligent monitoring panel generation module for unit status is used to implement S5: construct monitoring content from the start-stop distribution map of measuring points, collect oxygen concentration change information of measuring points in the start-stop distribution map of measuring points and combine them in time order, judge the combination result, and generate an intelligent monitoring panel scheme for gas turbine unit status. The horizontal oxygen concentration trend analysis module is used to perform differential calculations and spatial mapping processing based on time series data of oxygen concentration from multiple measurement points, and to construct a horizontal variation distribution map. The combustion initiation zone identification module is used to perform consistency analysis on the changing trends of measuring points in the lateral variation distribution map, and, in combination with the absolute level of oxygen concentration and the spatial front position determination conditions, to screen measuring points and aggregate connected regions in order to identify the combustion initiation zone. The combustion diffusion trajectory derivation module is used to construct a set of measurement point paths based on the progressive relationship of the measurement point change trend and the spatial adjacency relationship, and generate a combustion diffusion trajectory map through path filtering and topological connection; The measuring point stratification and start / stop determination module is used to serialize the combustion diffusion trajectory and, in combination with temporal sequence and spatial continuity, classify the measuring points into levels and determine their start / stop status. The intelligent monitoring module for generating unit status is used to combine and analyze the start-up and shutdown distribution map of measuring points and the oxygen concentration change sequence, and output the unit operation status monitoring results and control suggestions according to the preset status judgment rules.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the technical solution.

Claims

1. An intelligent monitoring method for condition assessment of gas turbine units, characterized in that, Includes the following steps: S1: Multiple oxygen sensors are arranged at the same height cross section of the boiler furnace to collect continuous time series oxygen concentration change information. The oxygen concentration change information at adjacent sampling times is differentially calculated, and the change trend of the measuring points is calculated according to the time sequence to generate a horizontal change distribution map. S2: Compare and analyze the horizontal variation distribution map, collect the change trend of adjacent measuring points and compare them point by point, determine the position of the measuring point according to the size law of the change trend of the continuous sampling period, screen the measuring point positions that meet the conditions, and generate a combustion initiation area identification map. The specific steps of S2 are as follows: S201: Obtain the lateral change distribution map, extract the change value sequence corresponding to the measurement point location, construct measurement point pairs according to spatial adjacency, and perform difference comparison and change direction determination on the change values ​​of adjacent measurement points to obtain a set of measurement point change trend pairs; S202: Based on the set of measurement point change trends and combined with the preset change trend judgment benchmark, perform point-by-point comparison on the relationship between the consistency of the change direction and the change amplitude of the measurement points within the continuous sampling period, filter the measurement point positions that meet the judgment benchmark, and obtain the set of trend-matching measurement points. S203: Based on the trend matching measurement point set, combined with the spatial distribution relationship of the measurement points, perform regional aggregation processing, determine the connectivity of adjacent measurement points that meet the conditions, establish a regional mapping structure, and generate a combustion initiation area identification map. S3: Based on the combustion initiation zone identification map, derive the horizontal measurement point path, collect the measurement point change trend for continuous comparison, connect the measurement point positions according to the progressive relationship of the change trend, analyze the connection results, and generate a combustion diffusion trajectory map; The specific steps for S3 are as follows: S301: Obtain the combustion initiation area identification map, extract the position coordinates and change trend sequence of the measuring points within the identification area, construct a set of candidate connecting measuring points based on spatial adjacency, and generate a set of candidate paths for measuring points; S302: Based on the candidate set of measurement point paths, perform continuous comparison on the changing trends of adjacent measurement points in the candidate paths, determine the progressive relationship of the trend direction and the decreasing relationship of the change magnitude, filter the path combinations that satisfy the progressive relationship, and obtain the set of trend progressive paths. S303: Based on the aforementioned trend progression path set, and combined with the spatial distribution order of the measuring points, perform path connection integration processing, sequentially associate the path nodes and construct the path topology structure to generate a combustion diffusion trajectory map; S4: Divide the combustion diffusion trajectory map into segments, collect the positions of the measuring points in the combustion diffusion trajectory map and sort them according to the trajectory order, filter the sorting results and divide the measuring point levels, and generate a measuring point start-stop distribution map. The specific steps of S4 are as follows: S401: Obtain the combustion diffusion trajectory map, extract the coordinates of the measuring points and the path order index in the trajectory, and perform serialization and arrangement of the measuring points according to the trajectory connection order to generate a measuring point trajectory sequence set; S402: Based on the measured point trajectory sequence set, perform a filtering process on the measured point positions in the sequence, remove discontinuous nodes and retain continuous node combinations according to the preset path continuity benchmark, and obtain a continuous measured point sequence set; S403: Based on the continuous measurement point sequence set, perform hierarchical division processing according to the sequential relationship of the trajectory, establish a forward and backward hierarchical mapping relationship for the measurement point positions and form a hierarchical structure, and generate a measurement point start and stop distribution map; S5: Construct monitoring content for the start-stop distribution map of the measuring points, collect oxygen concentration change information of the measuring points in the start-stop distribution map of the measuring points and combine them in time order, judge the combination result, and generate a smart monitoring scheme for the gas turbine unit status. The lateral variation distribution map includes the cross-sectional oxygen gradient field, the differential amplitude grid of measuring points, and the trend variation thermogram. The combustion initiation area identification map includes the initial ignition point coordinate set, the boundary of the violent oxidation reaction, and the geometric features of the combustion pilot zone. The combustion diffusion trajectory map includes the flame spread spatial connection line, the oxygen mutation propagation route, and the combustion zone expansion vector network. The measuring point start-stop distribution map includes the sensor activation priority sequence, the measuring point monitoring depth echelon, and the data acquisition sleep interval. The intelligent monitoring scheme for the gas turbine unit status includes the furnace combustion condition diagnosis panel, the oxygen abnormality alarm threshold setting, and the dynamic unit load adaptation strategy.

2. The intelligent monitoring method for gas turbine unit condition assessment according to claim 1, characterized in that, The method of determining the location of the measuring point based on the magnitude of the change trend in the continuous sampling period refers to identifying and verifying the spatial location of the measuring point that conforms to the change pattern based on the magnitude, direction and stability characteristics of the change in oxygen concentration in the continuous time series. The aforementioned conformity change pattern refers to the change characteristics in which the oxygen concentration difference, change direction, and continuity at the measuring point meet preset threshold conditions.

3. The intelligent monitoring method for gas turbine unit condition assessment according to claim 1, characterized in that, The aforementioned progressive relationship of change trends refers to the spatially ordered transmission and gradual strengthening and weakening of the oxygen concentration change trend at the measuring point.

4. The intelligent monitoring method for gas turbine unit condition assessment according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire sampling sequences from multiple oxygen sensors at the same height cross-section of the boiler furnace, perform time alignment processing on the oxygen concentration values ​​at the measuring points and arrange them according to a uniform sampling period to generate an oxygen concentration time series matrix. S102: Based on the oxygen concentration time series matrix, perform differential operation on the oxygen concentration values ​​at adjacent sampling times of the measuring points, calculate the weighted dynamic difference value of oxygen concentration, calculate the concentration change corresponding to the time interval and arrange them in time order to obtain the oxygen concentration difference sequence set. S103: Based on the oxygen concentration difference sequence set, and combined with the spatial coordinates of the measuring points, perform lateral mapping processing, sort the difference values ​​according to the cross-sectional position, construct a two-dimensional grid relationship, and generate a lateral variation distribution map.

5. The intelligent monitoring method for gas turbine unit condition assessment according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Obtain the start-stop distribution map of the measuring points, extract the start-stop status and oxygen concentration change sequence corresponding to the measuring point location, perform permutation and combination processing on the oxygen concentration values ​​of the measuring points according to the time sequence, and generate a set of measuring point time sequence combinations. S502: Based on the combined set of time series of measuring points, perform time-by-time comparisons on the magnitude and direction of oxygen concentration changes in the measuring point sequence, and perform classification judgments on the sequence according to the preset state judgment criteria to obtain a set of measuring point state judgments. S503: Based on the set of measurement point status determinations, and combined with the start-stop relationship of the measurement points, perform correlation integration processing, establish a time-series mapping structure for the measurement point status results and form an overall distribution relationship, and generate a smart monitoring scheme for the gas turbine unit status.

6. An intelligent monitoring system for assessing the condition of gas turbine units, characterized in that, The system is used to implement the intelligent monitoring method for gas turbine unit condition assessment as described in any one of claims 1-5, comprising: The horizontal oxygen concentration trend analysis module is used to achieve S1: by arranging multiple oxygen sensors at the same height cross section of the boiler furnace to collect continuous time series oxygen concentration change information, perform differential calculation on the oxygen concentration change information at adjacent sampling times, calculate the change trend of the measuring points according to the time sequence, and generate a horizontal change distribution map. The combustion initiation zone identification module is used to implement S2: compare and analyze the lateral change distribution map, collect the change trend of adjacent measuring points for point-by-point comparison, determine the position of measuring points based on the magnitude of the change trend of continuous sampling period, filter the measuring point positions that meet the conditions, and generate a combustion initiation zone identification map. The combustion diffusion trajectory derivation module is used to implement S3: based on the combustion initiation area identification map, derive the lateral measuring point path, collect the changing trend of the measuring points for continuous comparison, connect the measuring point positions according to the progressive relationship of the changing trend, analyze the connection results, and generate a combustion diffusion trajectory map; The measuring point layering and start / stop determination module is used to implement S4: segmenting the combustion diffusion trajectory map, collecting the measuring point positions in the combustion diffusion trajectory map and sorting them according to the trajectory order, filtering the sorting results and dividing the measuring point layers, and generating a measuring point start / stop distribution map. The intelligent monitoring module for unit status is used to implement S5: construct monitoring content for the start-stop distribution map of the measuring points, collect oxygen concentration change information of the measuring points in the start-stop distribution map of the measuring points and combine them in time order, judge the combination result, and generate an intelligent monitoring scheme for the gas turbine unit status.

Images