Real-time control simulation system for air cooling island anti-freezing

Abstract

The application belongs to the technical field of data processing, and particularly relates to a real-time control simulation system for air cooling island anti-freezing, which acquires the temperature field, back pressure field and fan operation state in a target area in real time through a monitoring module; a analysis module constructs the space-time distribution field of temperature and back pressure according to the temperature field and the back pressure field, and further fuses into a temperature-back pressure causal distribution mapping field; an inversion module discriminates the fault type based on the mapping field and a back pressure anomaly recognition model, and generates an inversion mapping space containing early warning level, start time, air volume, air speed and air direction in combination with a fan operation inversion function; a response simulation module uses a reinforcement adjustment strategy and a simulation algorithm to perform real-time anti-freezing control, so that the temperature and the back pressure are maintained in a standard interval; the application realizes intelligent identification, accurate prediction and self-adaptive regulation and control of the air cooling island freezing hazard, and effectively improves the accuracy of anti-freezing control and the system operation reliability.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and in particular relates to a real-time control simulation system for antifreeze in air-cooled islands. Background Technology

[0002] Air-cooled islands, as critical cooling equipment in thermal power plants, are prone to back pressure increases due to the freezing of heat sink tube bundles during operation in low-temperature environments, affecting unit operating efficiency and safety. Currently, anti-freezing control of air-cooled islands mainly relies on manual experience or simple threshold adjustment based on local temperature monitoring, making it difficult to accurately perceive the spatiotemporal evolution of temperature and back pressure across the entire field and to uncover causal relationships between anomalies. Existing systems lack in-depth fusion analysis of multi-dimensional temperature fields, back pressure fields, and fan operating status, failing to effectively construct a distributed mapping field reflecting the coupling relationship between temperature and back pressure, resulting in delayed anomaly identification and inaccurate type judgment. Furthermore, traditional control strategies do not fully consider the dynamic impact of fan airflow, speed, and direction on the spatiotemporal changes of temperature gradients, making it difficult to achieve adaptive control based on anomaly levels. This leads to insufficient real-time anti-freezing response and control precision, resulting in blind spots and low efficiency. In addition, the lack of a real-time evaluation and feedback mechanism for the efficiency and uniformity of the control process prevents closed-loop optimization control, impacting the intelligence level of air-cooled island anti-freezing early warning and control. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a real-time control simulation system for anti-freezing in air-cooled islands. This system uses a monitoring module to acquire real-time temperature, back pressure, and fan operating status within the target area. An analysis module constructs the spatiotemporal distribution fields of temperature and back pressure, further fusing them into a temperature-back pressure causal distribution mapping field. An inversion module uses this mapping field and a back pressure anomaly identification model to determine fault types and, combined with a fan operation inversion function, generates an inversion mapping space including warning level, start-up time, air volume, wind speed, and wind direction. A response simulation module employs enhanced adjustment strategies and simulation algorithms for real-time anti-freezing control, maintaining temperature and back pressure within the standard range. This invention achieves intelligent identification, accurate prediction, and adaptive control of freezing damage risks in air-cooled islands, effectively improving the accuracy of anti-freezing control and the reliability of system operation.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A real-time control simulation system for antifreeze in air-cooled islands includes: a monitoring module, an analysis module, an inversion module, and a response simulation module;

[0006] The monitoring module is used to monitor the first temperature field information, back pressure distribution field and air-cooled island fan operation status information in the target area in real time.

[0007] The analysis module is used to calculate the temperature spatiotemporal distribution field and the back pressure spatiotemporal distribution field representing the target area based on the first temperature field information and the back pressure distribution field, and to calculate the temperature-back pressure causal distribution mapping field representing the real-time state of the target area based on the temperature spatiotemporal distribution field and the back pressure spatiotemporal distribution field combined with the correlation analysis algorithm.

[0008] The inversion module is used to determine the type of back pressure anomaly based on the temperature-back pressure causal distribution mapping field of the target area combined with the back pressure anomaly identification model. When the anomaly is determined to be antifreeze anomaly, the inversion mapping space is calculated based on the temperature-back pressure causal distribution mapping field of the target area combined with the operation inversion function of the air-cooled island fan. The inversion mapping space includes at least the warning level index, the fan start time, air volume, wind direction and wind speed. The operation inversion function of the air-cooled island fan is used to obtain the corresponding fan start time, running time, wind direction and wind speed based on the maximum working area of ​​each fan and the current temperature and back pressure change trend, so that the temperature and back pressure of the target area are kept in the standard temperature range and the standard back pressure range in real time.

[0009] The response simulation module is used to perform real-time antifreeze simulation control by combining the real-time inversion mapping space with the preset enhancement adjustment strategy and simulation algorithm, so that the real-time temperature and back pressure of the monitored target area meet the preset temperature standard range and back pressure standard range.

[0010] Specifically, the process of constructing the temperature-backpressure causal distribution mapping field includes:

[0011] Based on the topological relationships of the target monitoring area, determine the patrol coordinate system of the target monitoring area;

[0012] Obtain the temperature information of each location point corresponding to each anomaly type in the historical inspection coordinate system, and determine the temperature spatial gradient field of adjacent location points.

[0013] Specifically, the construction process of the temperature-backpressure causal distribution mapping field also includes:

[0014] Obtain temperature trend information at each location point at different time points, and obtain the temperature time gradient field and temperature change curve at each location point;

[0015] The temperature time gradient field and temperature change curve at each location point are mapped to the corresponding location point in the temperature spatial gradient field to obtain the temperature spatiotemporal distribution field. For the temperature change curves at locations with spatial gradient anomalies, chromaticity and luminance are marked according to the magnitude of the spatial temperature gradient anomaly deviation and the HSV color space.

[0016] Specifically, the construction process of the temperature-backpressure causal distribution mapping field also includes:

[0017] Obtain the back pressure information of each location point under the corresponding anomaly type in the patrol coordinate system, and obtain the back pressure spatiotemporal distribution field in the same way as the temperature spatiotemporal distribution field.

[0018] The spatial coordinates of the back pressure spatiotemporal distribution field and the temperature spatiotemporal distribution field are aligned to obtain the aligned temperature-back pressure spatial gradient pair and temperature-back pressure temporal gradient pair.

[0019] Based on the aligned temperature-backpressure spatial gradient pairs, the correlation degree of temperature-backpressure spatial gradient changes is obtained through correlation analysis algorithms.

[0020] Specifically, the construction process of the temperature-backpressure causal distribution mapping field also includes:

[0021] Based on the aligned temperature-backpressure time gradient pairs, the correlation degree of temperature-backpressure time gradient changes is obtained.

[0022] Using the correlation between the temporal and spatial gradient changes of temperature and back pressure as weights and combined with a radial kernel function, a gradient change mapping of temperature and back pressure corresponding to time and space is constructed. The established gradient change mapping of temperature and back pressure corresponding to time and space is then embedded into the temperature spatiotemporal distribution field to obtain a temperature-back pressure causal distribution mapping field.

[0023] Based on the temperature-backpressure causal distribution mapping field combined with real-time collected temperature gradient data and backpressure gradient data, a bidirectional prediction verification is performed through a causal inference algorithm to obtain the temperature-backpressure causal distribution mapping field under each verified anomaly type.

[0024] Specifically, the construction process of the back pressure anomaly identification model includes:

[0025] Obtain text information of actual operation fault scenarios of air-cooled island, extract the corresponding fault type and the spatial gradient change state and temporal gradient change state of temperature and back pressure under the corresponding fault type, and construct an anomaly type information matrix.

[0026] Based on the anomaly type information matrix and statistical analysis algorithm, the range of the consistency anomaly level index of temperature and back pressure under each anomaly type is obtained; the consistency anomaly level index of temperature and back pressure is obtained by multiplying the ratio of the spatial gradient change state of temperature and back pressure at each time point by the ratio of the area of ​​abnormal temperature to the area of ​​abnormal back pressure at the corresponding time point.

[0027] Based on the anomaly type information matrix and the consistency anomaly level index of temperature and back pressure, a back pressure anomaly identification model is obtained by combining the random forest tree algorithm with the validated temperature-back pressure causal distribution mapping field and simulation algorithm for simulation training.

[0028] Specifically, the process of performing real-time antifreeze simulation control includes:

[0029] The historical air-cooled island fan operation status information is obtained and statistical algorithms are used to extract the fan's air volume, wind speed, wind direction status information at different times, as well as the spatial temperature gradient information and temporal temperature gradient change information under the corresponding operation status.

[0030] Based on the information on air volume, wind speed, and wind direction, as well as the spatial temperature gradient information under the corresponding operating conditions, the contribution of air volume, wind speed, and wind direction to the adjustment of spatial temperature gradient changes is obtained through factor analysis algorithm.

[0031] Based on the current wind turbine's air volume, operating time, wind speed, wind direction status information, as well as the corresponding spatial adjustment contribution and spatial temperature gradient information under the operating state, a spatial temperature gradient inversion function is constructed using a support vector machine.

[0032] Similarly, by utilizing information on air volume, operating time, wind speed, wind direction, and the time-temperature gradient change information under the corresponding operating conditions, a time gradient inversion function is constructed.

[0033] Specifically, the process of conducting real-time antifreeze simulation control also includes:

[0034] The temperature-backpressure causal distribution mapping field corresponding to the real-time target monitoring area is obtained. Combined with the preset temperature anomaly threshold value and the preset backpressure anomaly threshold value, the temperature anomaly time point and the backpressure anomaly time point are determined. The average of the temperature anomaly time point and the backpressure anomaly time point is used as the fan start time point.

[0035] Simultaneously, based on the temperature-back pressure causal distribution mapping field corresponding to the real-time target monitoring area and the anomaly type information matrix, the causal type of back pressure anomaly corresponding to the wind turbine start-up time point and the identification accuracy are predicted.

[0036] Simultaneously, based on the number of points at each time point that reach the critical value of temperature anomaly and the critical value of back pressure anomaly in the temperature-back pressure causal distribution mapping field, the area of ​​the temperature anomaly monitoring interval or the area of ​​the back pressure anomaly monitoring interval corresponding to each time point is determined.

[0037] Based on the ratio of the area of ​​the abnormal monitoring interval corresponding to each time point to the area of ​​the working area covered by the maximum rated power of each wind turbine, combined with the ratio of the back pressure gradient to the temperature gradient at the corresponding time point and a comprehensive fuzzy algorithm, the consistency anomaly level index at each time point is obtained.

[0038] Specifically, the process of conducting real-time antifreeze simulation control also includes:

[0039] Based on the spatial and temporal gradient changes of temperature under the corresponding index level, and combined with the corresponding spatial temperature gradient inversion function and temporal gradient inversion function, the real-time air volume and wind speed of each fan in the corresponding working area are obtained.

[0040] At the fan start-up time, the air volume, air direction and wind speed are obtained in real time, and combined with the deep strategy algorithm, antifreeze control commands are generated for real-time antifreeze control.

[0041] Real-time monitoring of the adjusted temperature-backpressure causal distribution mapping field yields the ratio of the area of ​​the abnormal temperature monitoring interval at the corresponding time point after adjustment to the area of ​​the abnormal temperature monitoring interval at the corresponding time point predicted before adjustment. A temperature antifreeze regulation efficiency index is constructed, and the temperature spatial gradient field at the corresponding time point after adjustment is monitored in conjunction with a uniformity algorithm to calculate the temperature gradient change uniformity at the corresponding time point.

[0042] The temperature antifreeze regulation efficiency index and temperature gradient change uniformity at the corresponding time points are mapped in real time to the HSV color space for visualization processing, to obtain a temperature anomaly regulation efficiency uniformity comparison layer at each time point, and then mapped to the temperature-backpressure causal distribution mapping field for three-dimensional spatial and temporal display.

[0043] Specifically, the temperature antifreeze regulation efficiency index and the temperature gradient change uniformity at the corresponding time point are compared with the preset temperature antifreeze regulation efficiency index threshold and temperature gradient change uniformity threshold, respectively. If either the temperature antifreeze regulation efficiency index or the temperature gradient change uniformity at the corresponding time point does not meet the corresponding threshold, the corresponding temperature abnormal regulation efficiency comparison layer, the antifreeze regulation efficiency index, and the temperature gradient change uniformity at the corresponding time point are fed back to the deep strategy algorithm to adjust the antifreeze control command at the corresponding time point in real time.

[0044] Simultaneously, based on the currently predicted temperature antifreeze regulation efficiency index and the temperature gradient change uniformity at the corresponding time point, the temperature-backpressure gradient change mapping in the temperature-backpressure causal distribution mapping field is used to obtain the currently predicted backpressure antifreeze regulation efficiency index and the backpressure gradient change uniformity at the corresponding time point.

[0045] Based on the currently predicted back pressure antifreeze regulation efficiency index and the uniformity of back pressure gradient change at the corresponding time point, and the temperature antifreeze regulation efficiency index and the uniformity of temperature gradient change at the corresponding time point, the consistency anomaly level index of temperature and back pressure is adjusted to obtain the consistency anomaly level index of temperature and back pressure at the current moment. The anomaly type of the air-cooled island is judged in real time by the back pressure anomaly identification model that has been trained. When it is determined to be an antifreeze anomaly again, the above antifreeze control command adjustment process is repeated until the temperature antifreeze regulation efficiency index and the uniformity of temperature gradient change at the corresponding time point meet the corresponding threshold in real time.

[0046] When it is determined again that it is not an antifreeze anomaly, the temperature and back pressure consistency anomaly level index is combined with the temperature-back pressure causal distribution mapping field. The corresponding type is searched and located according to the direction of increasing deviation between temperature gradient fluctuation and back pressure gradient fluctuation. The corresponding non-antifreeze anomaly location point is obtained and mapped into the temperature-back pressure causal distribution mapping field for real-time display and early warning.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] This invention addresses the shortcomings of existing technologies by employing four modules—monitoring, analysis, inversion, and response simulation—to collaboratively construct a bidirectionally validated temperature-backpressure causal distribution mapping field. It then combines this with a random forest algorithm to train an accurate backpressure anomaly identification model, and finally uses a support vector machine to construct spatial and temporal temperature gradient inversion functions, achieving intelligent full-process anti-freezing for air-cooled islands. This allows for precise identification of anomaly types, determination of fan start-up timing, and quantification of control parameters, avoiding the blindness of traditional manual operation. By real-time monitoring of adjustment efficiency and uniformity, combined with closed-loop feedback to optimize control commands, it ensures that temperature and backpressure remain stable within the standard range, reducing energy consumption. Furthermore, this invention can locate non-anti-freezing anomalies and provide real-time warnings, balancing anti-freezing effectiveness with fault prevention, comprehensively improving the safety, economy, and intelligence level of air-cooled island operation. Attached Figure Description

[0049] Figure 1 This is a block diagram of a real-time control simulation system for antifreeze in an air-cooled island, according to Embodiment 1 of the present invention.

[0050] Figure 2 This is a diagram of the temperature-backpressure causal distribution mapping field architecture of Embodiment 1 of the present invention. Detailed Implementation

[0051] Example 1

[0052] Please see Figure 1 The present invention provides an embodiment of a real-time control simulation system for antifreezing of air-cooled islands, comprising: a monitoring module, an analysis module, an inversion module and a response simulation module;

[0053] The monitoring module is used to monitor the first temperature field information, back pressure distribution field and air-cooled island fan operation status information in the target area in real time.

[0054] The analysis module is used to calculate the temperature spatiotemporal distribution field and the back pressure spatiotemporal distribution field representing the target area based on the first temperature field information and the back pressure distribution field, and to calculate the temperature-back pressure causal distribution mapping field representing the real-time state of the target area based on the temperature spatiotemporal distribution field and the back pressure spatiotemporal distribution field combined with the correlation analysis algorithm.

[0055] The inversion module is used to determine the type of back pressure anomaly based on the temperature-back pressure causal distribution mapping field of the target area combined with the back pressure anomaly identification model. When the anomaly is determined to be antifreeze anomaly, the inversion mapping space is calculated based on the temperature-back pressure causal distribution mapping field of the target area combined with the operation inversion function of the air-cooled island fan. The inversion mapping space includes at least the warning level index, the fan start time, air volume, wind direction and wind speed. The operation inversion function of the air-cooled island fan is used to obtain the corresponding fan start time, running time, wind direction and wind speed based on the maximum working area of ​​each fan and the current temperature and back pressure change trend, so that the temperature and back pressure of the target area are kept in the standard temperature range and the standard back pressure range in real time.

[0056] The response simulation module is used to perform real-time antifreeze simulation control by combining the real-time inversion mapping space with the preset enhancement adjustment strategy and simulation algorithm, so that the real-time temperature and back pressure of the monitored target area meet the preset temperature standard range and back pressure standard range.

[0057] Further explanation is needed; please refer to [link / reference]. Figure 2 The construction process of the temperature-backpressure causal distribution mapping field in this embodiment includes:

[0058] Based on the topological relationships of the target monitoring area, determine the patrol coordinate system of the target monitoring area;

[0059] It should be further explained that the specific implementation process of determining the patrol coordinate system of the target monitoring area in this embodiment includes the following steps:

[0060] A three-dimensional structural model of the air-cooled island heat sink tube bundle array is obtained, the row and column topology relationship of the tube bundle arrangement and the spacing data of adjacent tube bundles are extracted, and a topology network diagram with tube bundle units as basic nodes is established.

[0061] Based on the connection relationships and relative positions of each node in the topology network diagram, determine the origin and axis of the inspection coordinate system. Set the center point of the central fan group at the bottom of the air-cooled island as the origin of the coordinate system, with the direction perpendicular to the tube bundle array plane as the Z-axis, the direction parallel to the tube bundle row as the X-axis, and the direction parallel to the tube bundle column as the Y-axis.

[0062] Based on the tube spacing data, a grid is generated along the X and Y axes to create a uniformly distributed coordinate point array. Each coordinate point corresponds to an actual monitoring location, and the spacing between the coordinate points is determined according to the tube row spacing and column spacing.

[0063] Based on the operating characteristics of the wind turbines and the airflow distribution patterns, the core coverage area of ​​each wind turbine is identified in a coordinate matrix. Using the turbine impeller diameter and tilt angle data, the projected coverage area of ​​each wind turbine on the XY plane is calculated, and a mapping is established between the coordinate points within this area and the corresponding wind turbine.

[0064] The initially established patrol coordinate system was verified and optimized. The uniformity of coordinate point coverage over the target area was calculated by simulating the distribution of temperature sensors within the coordinate system. Based on the distribution characteristics of historical anomaly data, the coordinate points in high-incidence areas were densified to increase monitoring density, resulting in the final optimized patrol coordinate system.

[0065] Obtain the temperature information of each location point corresponding to each anomaly type in the historical inspection coordinate system, and determine the temperature spatial gradient field of adjacent location points.

[0066] It should be further explained that the specific implementation steps for determining the temperature spatial gradient field of adjacent locations in this embodiment include:

[0067] Extract temperature data sequences recorded at each location point in the inspection coordinate system under various abnormal operating conditions from the historical database, and store them as a temperature information matrix according to the type of abnormality.

[0068] Data preprocessing was performed on the temperature information matrix for each anomaly type. A time series alignment algorithm was used to unify the data sampling frequency, and Lagrange interpolation was applied to compensate for missing temperature data points, ensuring that the temperature data at each location point was complete and continuous.

[0069] Based on the preprocessed temperature data, the temperature gradient value between each location point and its neighboring locations is calculated. The system iterates through each location point in the coordinate system, finding its nearest neighbors in the east, south, west, and north directions. For each location point, the temperature difference between it and its four neighboring points is calculated, and divided by the actual distance in the corresponding coordinate direction to obtain the temperature gradient components in the four directions.

[0070] Based on the temperature gradient components in four directions, the comprehensive temperature gradient amplitude at each location point is calculated. The comprehensive gradient value is calculated using the Euclidean norm, and the direction angle of the maximum gradient component is recorded. The calculated gradient amplitude is compared with the preset normal operating condition gradient threshold, and abnormal gradient points that exceed the threshold are identified.

[0071] Cluster analysis is performed on the abnormal gradient regions. Density clustering algorithm is used to identify a set of continuous abnormal gradient points in space to form abnormal gradient regions. The geometric center coordinates, average gradient magnitude and main gradient direction of each abnormal region are calculated to generate an abnormal gradient field map with spatial distribution characteristics. The map is associated with the inspection coordinate system and stored to establish the mapping relationship between the location coordinates and gradient characteristics.

[0072] Obtain temperature trend information at each location point at different time points, and obtain the temperature time gradient field and temperature change curve at each location point;

[0073] It should be further explained that the process of obtaining the temperature time gradient field and temperature change curve at each location point in this embodiment includes:

[0074] Based on the coordinate identifiers of each location point in the inspection coordinate system, temperature sampling data for the corresponding location point over a continuous time series are extracted from the historical database. A timestamp alignment algorithm is used to unify the data acquisition time points of each location point to ensure the synchronization of the time series.

[0075] The temperature time series data at each location point are preprocessed, and a sliding window averaging filter algorithm is applied to eliminate random noise interference. Cubic spline interpolation is used to complete the temperature data at missing time points, forming a continuous and smooth temperature time series.

[0076] The first-order differential of the temperature time series at each location point is calculated, and the instantaneous temperature change rate at each time point is obtained using the central difference method. The temperature time gradient value at each time point is calculated based on the temperature difference between adjacent time points and the sampling time interval.

[0077] The processed temperature time series data was converted into a time-temperature coordinate point set, and a Bézier curve fitting algorithm was used to generate a smooth temperature change curve. Feature parameters extracted from the curve included the number of extreme points, inflection points in the trend, average slope, and fluctuation frequency.

[0078] The temporal gradient value of each location point is stored as a gradient vector in time series, and associated with the corresponding temperature change curve feature parameters. Spatiotemporal coding technology is used to map and associate the temporal gradient field with the spatial position of the inspection coordinate system, forming a gradient distribution matrix with time dimension.

[0079] The generated temporal gradient field is validated for consistency by comparing the gradient change trends of adjacent points over the same time period to ensure the accuracy of the data calculation. The validated temperature temporal gradient field and temperature change curve features are stored in a database to provide a data foundation for the subsequent construction of the spatiotemporal distribution field.

[0080] The temperature temporal gradient field and temperature change curve at each location point are connected and mapped to the corresponding location point in the temperature spatial gradient field to obtain the temperature spatiotemporal distribution field. For the temperature change curves at locations with spatial gradient anomalies, chromaticity and luminance are marked according to the magnitude of the spatial temperature gradient anomaly deviation and the HSV color space. Specifically:

[0081] Based on the coordinate mapping relationship of location points, the temperature temporal gradient field and temperature change curve of each location point are spatiotemporally fused with the corresponding location point in the temperature spatial gradient field to obtain the temperature spatiotemporal distribution field. A spatial gradient anomaly deviation calculation algorithm is used to process location points with spatial gradient anomalies, calculating the spatial gradient anomaly deviation value for each anomaly location point. Based on the HSV color space conversion model, the spatial gradient anomaly deviation value is linearly mapped to the hue channel to obtain the corresponding chroma value, and simultaneously linearly mapped to the lightness channel to obtain the corresponding brightness value. A color marking algorithm is used to assign the obtained chroma and brightness values ​​to the temperature change curve of the corresponding anomaly location point, achieving visual marking based on the degree of spatial gradient anomaly. Finally, spatiotemporal data integration technology is used to fuse the marked temperature change curve with the temperature spatiotemporal distribution field to form a temperature spatiotemporal distribution field with spatiotemporal anomaly markings.

[0082] Obtain the back pressure information of each location point under the corresponding anomaly type in the patrol coordinate system, and obtain the back pressure spatiotemporal distribution field in the same way as the temperature spatiotemporal distribution field.

[0083] The spatial coordinates of the back pressure spatiotemporal distribution field and the temperature spatiotemporal distribution field are aligned to obtain the aligned temperature-back pressure spatial gradient pair and temperature-back pressure temporal gradient pair.

[0084] Based on the aligned temperature-backpressure spatial gradient pairs, the correlation degree of temperature-backpressure spatial gradient changes is obtained through correlation analysis algorithms.

[0085] Based on the aligned temperature-backpressure time gradient pairs, the correlation degree of temperature-backpressure time gradient changes is obtained.

[0086] Using the correlation between the temporal and spatial gradient changes of temperature and back pressure as weights and combined with a radial kernel function, a gradient change mapping of temperature and back pressure corresponding to time and space is constructed. The established gradient change mapping of temperature and back pressure corresponding to time and space is then embedded into the temperature spatiotemporal distribution field to obtain a temperature-back pressure causal distribution mapping field.

[0087] It should be further explained that the specific process for obtaining the temperature-backpressure gradient change mapping over time and space in this embodiment includes:

[0088] Step 1: Based on the constructed temperature-backpressure causal distribution mapping field and the preset air-cooled island backpressure anomaly type classification system, a data filtering algorithm is used to divide the real-time gradient dataset corresponding to each backpressure anomaly type from the real-time collected global gradient data of the air-cooled island. The preset air-cooled island backpressure anomaly type classification system includes at least three anomaly types, such as backpressure anomalies caused by low temperature freezing, backpressure anomalies caused by dust blockage, and backpressure anomalies caused by fin damage. The data filtering algorithm is based on the feature thresholds corresponding to each backpressure anomaly type in the temperature-backpressure causal distribution mapping field (such as the upper limit threshold of temperature gradient and the lower limit threshold of backpressure gradient for low temperature freezing anomalies, and the lower limit threshold of temperature gradient and the upper limit threshold of backpressure gradient for dust blockage anomalies). The algorithm filters the gradient data that meets the feature thresholds of the anomaly type from the real-time collected global gradient data of the air-cooled island, forming a real-time gradient dataset exclusive to each backpressure anomaly type. This dataset contains the real-time temperature gradient data and real-time backpressure gradient data of each inspection coordinate system position point under the corresponding anomaly type, and the real-time temperature gradient data and real-time backpressure data of each position point are accompanied by a unique spatial coordinate identifier and time node identifier.

[0089] Step 2: Based on the temperature gradient-backpressure gradient positive correlation rule specific to each backpressure anomaly type in the temperature-backpressure causal distribution mapping field, the real-time temperature gradient data of each location point in the real-time gradient dataset of the corresponding anomaly type is matched one-to-one with the spatial coordinates and time nodes of the inspection coordinate system and input into the positive mapping module of the temperature-backpressure causal distribution mapping field. The predicted backpressure gradient data of each location point under the corresponding anomaly type is obtained through the gradient correlation calculation model built into the positive mapping module. Among them, the temperature gradient-backpressure gradient positive correlation rule specific to each backpressure anomaly type is obtained by using the historical spatiotemporal gradient correlation data corresponding to that anomaly type, and employing linear regression and nonlinear fitting. The algorithm training generates rules that include the quantitative correlation between temperature gradient and backpressure gradient under this anomaly type (e.g., for every unit decrease in temperature gradient corresponding to a low-temperature freezing anomaly, the corresponding increase in backpressure gradient). The gradient correlation calculation model built into the forward mapping module is constructed based on the spatiotemporal gradient correlation parameters (including spatial correlation weight, temporal correlation weight, and radial kernel function parameters) of this anomaly type. The model receives real-time temperature gradient data with spatial coordinate identifiers and time node identifiers, calls the forward correlation rules corresponding to the anomaly type, and outputs predicted backpressure gradient data that accurately matches the real-time temperature gradient data in terms of spatial coordinates and time nodes.

[0090] Step 3: Based on the error analysis algorithm, the predicted backpressure gradient data and real-time backpressure gradient data at each location point under the corresponding anomaly type are compared point by point to calculate the deviation value between the two. By comparing with the preset backpressure gradient prediction deviation threshold, the accuracy of the temperature-backpressure positive mapping under this anomaly type is determined. The error analysis algorithm uses a combination of root mean square error (RMSE) calculation method and mean absolute error (MAE) calculation method to calculate the RMSE and MAE values ​​between the predicted backpressure gradient data and real-time backpressure gradient data at each location point. The preset backpressure gradient prediction deviation threshold is set based on the error statistics of historical verification data for this anomaly type, including a first-level deviation threshold and a second-level deviation threshold. If both the RMSE and MAE values ​​are less than or equal to the first-level deviation threshold, the positive mapping verification is considered to have passed completely. If both the RMSE and MAE values ​​are less than or equal to the second-level deviation threshold and at least one is greater than the first-level deviation threshold, the positive mapping verification is considered to have passed initially. If any error value is greater than the second-level deviation threshold, it is marked as a positive mapping anomaly, and the spatial coordinates and time node of the anomaly location point are recorded.

[0091] Step 4: Based on the backpressure gradient-temperature gradient inverse correlation rule specific to each backpressure anomaly type in the temperature-backpressure causal distribution mapping field, the real-time backpressure gradient data of each location point in the real-time gradient dataset of the corresponding anomaly type is matched one by one with the spatial coordinates and time nodes of the inspection coordinate system and input into the inverse mapping module of the temperature-backpressure causal distribution mapping field. The predicted temperature gradient data of each location point under the corresponding anomaly type is obtained through the inverse gradient correlation calculation model built into the inverse mapping module. Among them, the backpressure gradient-temperature gradient inverse correlation rule specific to each backpressure anomaly type is obtained through the historical spatiotemporal gradient inverse correlation data corresponding to that anomaly type (i.e., historical data from backpressure gradient changes to deduce temperature gradient changes). The back pressure gradient is generated using an inverse regression algorithm. The rules include the quantified inverse correlation between the back pressure gradient and the temperature gradient under this anomaly type (e.g., for every unit decrease in the back pressure gradient corresponding to a low-temperature freezing anomaly, the corresponding increase in the temperature gradient). The inverse mapping module has a built-in inverse gradient correlation calculation model, which is constructed based on the spatiotemporal inverse gradient correlation parameters (including inverse spatial correlation weights, inverse temporal correlation weights, and inverse radial kernel function parameters) of this anomaly type. The model receives real-time back pressure gradient data with spatial coordinates and time nodes, calls the inverse correlation rules of the corresponding anomaly type, and outputs predicted temperature gradient data that accurately matches the real-time back pressure gradient data in terms of spatial coordinates and time nodes.

[0092] Step 5: Based on the same error analysis algorithm as in Step 3, the predicted temperature gradient data and real-time temperature gradient data at each location point under the corresponding anomaly type are compared point by point. The deviation between the two is calculated, and the accuracy of the temperature-backpressure inverse mapping under this anomaly type is determined by comparing it with the preset temperature gradient prediction deviation threshold. The error analysis algorithm also uses the root mean square error (RMSE) and mean absolute error (MAE) calculation methods to calculate the RMSE and MAE values ​​between the predicted temperature gradient data and real-time temperature gradient data at each location point. The preset temperature gradient prediction deviation threshold is set based on the error statistics of historical verification data for this anomaly type, and also includes a first-level deviation threshold and a second-level deviation threshold. If both the RMSE and MAE values ​​are less than or equal to the first-level deviation threshold, the inverse mapping verification is considered fully passed. If both the RMSE and MAE values ​​are less than or equal to the second-level deviation threshold, and at least one is greater than the first-level deviation threshold, the inverse mapping verification is considered preliminarily passed. If any error value is greater than the second-level deviation threshold, it is marked as an inverse mapping anomaly, and the spatial coordinates and time node of the anomaly location point are recorded.

[0093] Step 6: Based on the bidirectional mapping deviation comprehensive judgment logic, integrate the verification results of forward and reverse mapping under each back pressure anomaly type to determine the bidirectional mapping verification result of the temperature-back pressure causal distribution mapping field under that anomaly type. The bidirectional mapping deviation comprehensive judgment logic first judges the combination of the verification results of forward and reverse mapping: if both forward and reverse mapping verifications are fully passed, then the bidirectional mapping verification is directly determined to be passed; if both forward and reverse mapping verifications are initially passed, or one verification is fully passed while the other is initially passed, then the bidirectional mapping verification is initially determined to be passed. This requires considering the error distribution of the bidirectional mapping (such as the spatial region of the error concentration and the trend of error change over time) to determine the gradient relationship corresponding to the anomaly type. Fine-tune the linkage parameters (spatial correlation weight, temporal correlation weight, and radial kernel function parameter). If any mapping is marked as an anomaly, adjust the gradient correlation parameter corresponding to the anomaly type based on the deviation analysis results of the anomaly mapping (including the spatial coordinate distribution of the anomaly location points, the time node pattern of the anomaly occurrence, and the magnitude level of the deviation). For example, for the low temperature freezing anomaly, if the positive mapping deviation is concentrated at the edge location points of the inspection coordinate system, increase the spatial correlation weight of the edge region; if the deviation fluctuates significantly over time, adjust the temporal correlation weight or the bandwidth parameter of the radial kernel function. After adjustment, repeat steps 2 to 5 until the root mean square error and mean absolute error of the bidirectional mapping both meet the first-level deviation threshold, and the bidirectional mapping verification is deemed successful.

[0094] Step 7: Based on the spatiotemporal continuity requirement of the inspection coordinate system, for each backpressure anomaly type, the temperature-backpressure causal distribution mapping field that has passed the bidirectional mapping verification is compared with the predicted gradient change trend and the real-time gradient change trend of each adjacent location point and continuous time node using a spatiotemporal trend consistency verification algorithm to confirm the spatiotemporal correlation validity of the mapping field. Specifically, the spatiotemporal trend consistency verification algorithm first determines the preset relative trend relationship between the temperature gradient and the backpressure gradient for each backpressure anomaly type. For example, the preset relative trend relationship for the low-temperature freezing anomaly is: when the temperature gradient shows an upward trend, the backpressure gradient shows a downward trend; the temperature gradient... When the temperature gradient decreases, the back pressure gradient increases. This relationship is derived from the physical principle that increased temperature reduces tube freezing and improves heat dissipation efficiency, leading to a decrease in back pressure. The preset relative trend for dust blockage anomalies is: when the temperature gradient increases, the back pressure gradient also increases. This is derived from the physical principle that dust blockage reduces heat dissipation efficiency and causes a synchronous increase in temperature and back pressure. Subsequently, the algorithm extracts the predicted temperature gradient change direction (increasing, decreasing) of each adjacent point in the mapping field (adjacent points in the same row or column of the inspection coordinate system) at continuous time nodes (e.g., continuous time points at unit time intervals). The algorithm first extracts the predicted backpressure gradient trend direction (including the predicted temperature gradient and the real-time backpressure gradient trend direction at the corresponding location points within the same continuous time node.) Then, it performs trend comparison in three dimensions: first, whether the predicted temperature gradient trend direction and the real-time temperature gradient trend direction are consistent at adjacent location points; second, whether the predicted backpressure gradient trend direction and the real-time backpressure gradient trend direction are consistent at adjacent location points; and third, whether the relative trend relationship between the predicted temperature gradient and the backpressure gradient is consistent with the preset relative trend relationship for this anomaly type. If all adjacent location points within the same continuous time node... If all nodes meet the trend consistency requirements of the above three dimensions, then the temperature-backpressure causal distribution mapping field under this anomaly type is confirmed to satisfy the spatiotemporal correlation validity. If there are adjacent locations that do not meet the trend consistency requirements (such as in the low temperature freezing anomaly, when the predicted temperature gradient of an adjacent location increases, the backpressure gradient also increases, which is inconsistent with the preset relative trend relationship), then the spatiotemporal gradient trend correlation rules of this anomaly type in the mapping field are corrected (such as adjusting the relative trend coefficients of temperature gradient and backpressure gradient). After correction, the bidirectional mapping verification of steps 2 to 6 is executed again until all adjacent locations meet the trend consistency requirements at consecutive time nodes.

[0095] Based on the results of all the above verification steps, the temperature-backpressure causal distribution mapping field under each backpressure anomaly type that meets the first-level deviation threshold and has a completely consistent spatiotemporal trend is marked as the verified temperature-backpressure causal distribution mapping field for that anomaly type. The data storage module records all verification parameters corresponding to this verified mapping field, including the root mean square error value of the bidirectional mapping, the mean absolute error value, the finally determined gradient correlation parameters, and the trend consistency rate of the spatiotemporal trend verification (the proportion of adjacent location points that meet the trend consistency requirements to the total number of adjacent location points). Simultaneously, the applicable operating conditions range of this mapping field (such as the ambient temperature range and the unit load range) is recorded, forming a dedicated verified mapping field file for each anomaly type. This provides accurate and condition-adaptive causal correlation basis for subsequent air-cooled island anomaly identification and control. Based on the temperature-backpressure causal distribution mapping field combined with real-time acquired temperature gradient data and backpressure gradient data, a causal inference algorithm is used for bidirectional prediction verification to obtain the verified temperature-backpressure causal distribution mapping field for each anomaly type.

[0096] It should be further explained that the process of verifying the temperature-backpressure causal distribution mapping field for each anomaly type in this embodiment includes:

[0097] S1. Based on the established temperature-backpressure causal distribution mapping field and the preset air-cooled island backpressure anomaly type classification system, the real-time gradient dataset corresponding to each backpressure anomaly type (such as anomalies caused by low temperature freezing, dust blockage, fin damage, etc.) is divided from the real-time collected global gradient data of the air-cooled island through a data filtering algorithm. This dataset contains the real-time temperature gradient data and real-time backpressure gradient data of each inspection coordinate system position point under the corresponding anomaly type.

[0098] S2. Based on the "temperature gradient-backpressure gradient" positive correlation rule specific to each backpressure anomaly type in the temperature-backpressure causal distribution mapping field (this rule is generated by training the historical spatiotemporal gradient correlation data of this anomaly type), the real-time temperature gradient data of each location point in the real-time gradient dataset of the corresponding anomaly type is matched one by one with the spatial coordinates and time nodes of the inspection coordinate system and input into the positive mapping module of the temperature-backpressure causal distribution mapping field. Through the gradient correlation calculation model built into the positive mapping module (built based on the spatiotemporal gradient correlation parameters of this anomaly type), the predicted backpressure gradient data of each location point under the corresponding anomaly type is obtained.

[0099] S3. Based on the error analysis algorithm, the predicted back pressure gradient data of each location point under the corresponding anomaly type is compared with the real-time back pressure gradient data point by point, and the deviation value between the two (such as root mean square error, mean absolute error) is calculated. By comparing with the preset back pressure gradient prediction deviation threshold, the accuracy of the temperature-back pressure positive mapping (temperature gradient predicts back pressure gradient) under this anomaly type is determined. If the deviation value is less than or equal to the preset threshold, the positive mapping verification is initially passed; if the deviation value is greater than the preset threshold, it is marked as a positive mapping anomaly.

[0100] S4. Based on the inverse correlation rule of "backpressure gradient - temperature gradient" specific to each backpressure anomaly type in the temperature-backpressure causal distribution mapping field (this rule is trained and generated from the historical spatiotemporal gradient inverse correlation data of this anomaly type, and complements the forward correlation rule), the real-time backpressure gradient data of each location point in the real-time gradient dataset of the corresponding anomaly type is matched one by one with the spatial coordinates and time nodes of the inspection coordinate system and input into the inverse mapping module of the temperature-backpressure causal distribution mapping field. Through the inverse gradient correlation calculation model built into the inverse mapping module (built based on the spatiotemporal inverse gradient correlation parameters of this anomaly type), the predicted temperature gradient data of each location point under the corresponding anomaly type is obtained.

[0101] S5. Based on the same error analysis algorithm as S3, the predicted temperature gradient data of each location point under the corresponding anomaly type is compared with the real-time temperature gradient data point by point, and the deviation value between the two is calculated. By comparing it with the preset temperature gradient prediction deviation threshold, the accuracy of the temperature-backpressure inverse mapping (backpressure gradient predicts temperature gradient) under this anomaly type is determined. If the deviation value is less than or equal to the preset threshold, the inverse mapping verification is initially passed; if the deviation value is greater than the preset threshold, it is marked as an inverse mapping anomaly.

[0102] S6. Based on the comprehensive judgment logic of bidirectional mapping deviation, the verification results of forward mapping and reverse mapping under each back pressure anomaly type are integrated. If both forward mapping and reverse mapping pass the initial test (the deviation values ​​all meet the corresponding thresholds), the bidirectional mapping verification of the temperature-back pressure causal distribution mapping field under this anomaly type is determined to be passed. If there is any mapping anomaly (forward or reverse deviation exceeds the threshold), the gradient correlation parameters corresponding to this anomaly type in the temperature-back pressure causal distribution mapping field are adjusted based on the deviation analysis results of the anomaly mapping (such as the location point and time node of the deviation concentration). S2 to S5 are repeated until the bidirectional mapping deviations all meet the preset thresholds.

[0103] S7. Based on the spatiotemporal continuity requirement of the inspection coordinate system, for the temperature-backpressure causal distribution mapping field that has passed the bidirectional mapping verification under each backpressure anomaly type, the spatiotemporal trend consistency verification algorithm is used to compare the predicted gradient change trend of each adjacent location point and continuous time node with the real-time gradient change trend (such as the consistency of the change direction of the backpressure gradient when the temperature gradient rises). If the trends are consistent, it is confirmed that the temperature-backpressure causal distribution mapping field under this anomaly type satisfies the spatiotemporal correlation validity; if the trends are inconsistent, the spatiotemporal gradient trend correlation rule of this anomaly type in the mapping field is corrected, and the bidirectional mapping verification is performed again until the spatiotemporal trends are consistent.

[0104] Based on the results of all the above verification steps, the temperature-backpressure causal distribution mapping field under each backpressure anomaly type that meets the threshold for bidirectional mapping deviation and has consistent spatiotemporal trends is marked as the verified temperature-backpressure causal distribution mapping field under that anomaly type. The verification parameters (such as error values, gradient correlation parameters, and spatiotemporal trend verification results) corresponding to the mapping field are recorded through the data storage module, forming a unique verified mapping field file for each anomaly type, providing accurate causal correlation basis for subsequent air-cooled island anomaly identification and control.

[0105] It should be further explained that the specific construction process of the back pressure anomaly identification model in this embodiment includes:

[0106] The process involves acquiring textual information about actual operational fault scenarios in the air-cooled island, extracting the corresponding fault types and the spatial and temporal gradient changes of temperature and back pressure under each fault type, and constructing an anomaly type information matrix. The specific steps include:

[0107] Based on the actual operational fault scenario text information recorded by the air-cooled island operation and maintenance management system, including on-site operation and maintenance logs, fault diagnosis and analysis reports and equipment maintenance records at the time of the fault, irrelevant redundant characters, repetitive descriptive statements and auxiliary information unrelated to the fault characteristics are removed from the text through text cleaning algorithms. Then, the fault time descriptions of different recorders are unified into the year-month-day-hour-minute-second format through text format standardization processing, and the fault location descriptions are unified into the inspection coordinate system coordinate description, to obtain a normalized fault scenario text dataset.

[0108] Based on the normalized fault scene text dataset, the text classification algorithm in natural language processing is combined with the back pressure anomaly type system of the air-cooled island to initially label the fault type of each fault scene text. Then, the clustering algorithm is used to classify and merge the initially labeled fault types, eliminate the fault type descriptions with the same meaning but different names, and obtain the classified fault type set and the fault scene text subsets corresponding to each fault type.

[0109] Based on the fault scenario text subsets corresponding to each fault type after classification, the entity recognition algorithm extracts the descriptive information related to temperature spatial distribution and back pressure spatial distribution for each fault scenario from the text. Then, combined with the spatial gradient feature transformation logic, the above descriptive information is transformed into the corresponding temperature spatial gradient change state and back pressure spatial gradient change state. The temperature spatial gradient change state includes the temperature spatial gradient direction and the temperature spatial gradient amplitude range, and the back pressure spatial gradient change state includes the back pressure spatial gradient direction and the back pressure spatial gradient amplitude range, thus obtaining the set of temperature and back pressure spatial gradient change states for each fault type.

[0110] Based on the text subset of fault scenarios corresponding to each fault type, the time series information extraction algorithm is used to extract the descriptive information related to temperature and time changes and the descriptive information related to back pressure and time changes in each fault scenario. Then, combined with the time gradient feature transformation logic, the above descriptive information is transformed into the corresponding temperature and time gradient change states and back pressure and time gradient change states. The temperature and time gradient change states include the temperature and time gradient change rate and the duration of the temperature and time gradient, and the back pressure and time gradient change states include the back pressure and time gradient change rate and the duration of the back pressure and time gradient, thus obtaining the set of temperature and back pressure and time gradient change states under each fault type.

[0111] Based on the obtained set of fault types, the set of temperature and back pressure spatial gradient change states under each fault type, and the set of temperature and back pressure temporal gradient change states, the row dimension of the anomaly type information matrix is ​​defined as each fault type after classification, and the column dimension is the preset gradient change state feature items, including temperature spatial gradient direction, temperature spatial gradient amplitude range, back pressure spatial gradient direction, back pressure spatial gradient amplitude range, temperature temporal gradient change rate, temperature temporal gradient duration, back pressure temporal gradient change rate, and back pressure temporal gradient duration. Through the feature mapping algorithm, each gradient change state corresponding to each fault type is filled into the cell of the corresponding fault type row and the corresponding feature item column in the matrix to obtain the initial anomaly type information matrix.

[0112] Based on the verified temperature and back pressure gradient correlation data in the historical fault database of the air-cooled island, the gradient change state of each cell in the initial anomaly type information matrix is ​​verified by a data verification algorithm. The algorithm determines whether the gradient change state in the matrix is ​​consistent with the historical verification data. If there is an inconsistency, the corresponding fault scenario text subset is returned and the gradient change state extraction operation is re-executed. At the same time, the integrity verification algorithm checks whether there is any missing information in each cell of the matrix. Cells with missing information are reasonably supplemented based on the gradient change state characteristics of other fault scenarios under the same fault type to obtain the verified and optimized anomaly type information matrix.

[0113] Based on the preset gradient change state feature term representation specification, the gradient change state representation of each cell in the verified and optimized anomaly type information matrix is ​​unified into a standard format through matrix standardization processing. Then, the standardized anomaly type information matrix is ​​stored in the feature database of the air-cooled island intelligent early warning system according to the fault type through the data storage module, providing structured anomaly type feature data support for the subsequent construction of temperature and back pressure causal distribution mapping field and training of back pressure anomaly identification model.

[0114] Based on the anomaly type information matrix and statistical analysis algorithm, the interval of the consistency anomaly level index of temperature and back pressure under each anomaly type is obtained; the consistency anomaly level index of temperature and back pressure is constructed by multiplying the ratio of the spatial gradient change state of temperature and back pressure at each time point by the ratio of the area of ​​abnormal temperature to the area of ​​abnormal back pressure at the corresponding time point.

[0115] Based on the anomaly type information matrix and the consistency anomaly level index of temperature and back pressure, a back pressure anomaly identification model is obtained by combining the random forest tree algorithm with the validated temperature-back pressure causal distribution mapping field and simulation algorithm for simulation training.

[0116] It should be further explained that the training process of the back pressure anomaly identification model in this embodiment includes:

[0117] Based on the spatial and temporal gradient changes of temperature and back pressure corresponding to each fault type in the anomaly type information matrix, and the spatiotemporal gradient correlation data specific to each anomaly type in the verified temperature and back pressure causal distribution mapping field, the two types of data are matched and integrated according to fault type through a feature fusion algorithm. Joint features under each fault type are extracted, including the coupling features of temperature spatial gradient direction and spatial correlation, temperature temporal gradient rate and temporal correlation, back pressure spatial gradient magnitude and spatial correlation, and back pressure temporal gradient duration and temporal correlation. At the same time, the consistency anomaly level index of back pressure is incorporated. This index is calculated based on the degree of coordinated change of back pressure anomaly level at different locations under the same fault type, to obtain the feature vector set corresponding to each fault type.

[0118] A multi-condition anomaly simulation scenario library for air-cooled islands is constructed based on simulation algorithms. This scenario library covers various back pressure anomaly scenarios under different ambient temperatures, different unit loads, and different equipment aging degrees. It corresponds one-to-one with the fault types in the anomaly type information matrix. The simulation algorithm simulates the spatiotemporal gradient change process of temperature and back pressure under each scenario, and outputs the temperature gradient data, back pressure gradient data, and back pressure consistency anomaly level data corresponding to the simulation scenario. The simulation data is converted into simulation feature vectors with the same format as the above feature vectors. The simulation data is then merged with the above real feature vector set to construct a comprehensive sample dataset for the back pressure anomaly identification model.

[0119] Based on the comprehensive sample dataset, each feature vector is associated with the corresponding fault type label and back pressure consistency anomaly level label through a data annotation algorithm to form labeled training samples. A stratified sampling algorithm is used to divide the labeled training samples into training set, validation set and test set according to a preset ratio. The training set is used for model parameter learning, the validation set is used for model hyperparameter optimization, and the test set is used for model generalization ability evaluation to ensure that the sample distribution ratio of each type of fault in each set is consistent.

[0120] The basic framework for a backpressure anomaly identification model is constructed based on the random forest algorithm. The initial hyperparameters of the model are determined by the grid search algorithm, including the number of decision trees, the maximum depth of the decision trees, the node splitting criterion, and the minimum number of samples in the leaf nodes. The training set is input into the basic framework for initial training. During the training process, a sub-sample set is extracted from the training set using the bootstrap sampling method to allocate training data to each decision tree. At the same time, based on the spatiotemporal gradient correlation weights of each fault type in the validated temperature and backpressure causal distribution mapping field, dynamic feature weights are set for the feature vectors in the model. Features with high causal correlation are given higher weights, thereby improving the model's focus on key features.

[0121] Based on the validation set, the hyperparameters of the initially trained random forest model are optimized. The cross-validation algorithm is used to calculate the model's performance metrics on the validation set under different hyperparameter combinations, including fault type identification accuracy, precision, recall, and back pressure consistency anomaly level prediction error. The hyperparameter combination with the best performance metrics is selected to update the model. At the same time, the spatiotemporal gradient data of a certain fault type in the validated temperature and back pressure causal distribution mapping field is input into the model. By analyzing the deviation between the model's output fault type prediction results and the actual fault types, the feature weights corresponding to the fault type are adjusted. The above optimization process is repeated until the model's performance metrics on the validation set are stable above the preset threshold.

[0122] New abnormal scenario data not included in the comprehensive sample dataset is generated based on simulation algorithms. The new scenario data is transformed into feature vectors and input into the optimized model. The model's ability to identify fault types in new scenarios is evaluated through simulation verification algorithms. If the model's recognition accuracy for a certain type of new scenario is lower than a preset threshold, the new scenario data is added to the training set, the model is incrementally trained, the splitting rules of the decision tree and feature weights are adjusted, and the model's ability to identify rare abnormal scenarios is enhanced.

[0123] The generalization ability of the incrementally trained model is evaluated based on the test set. The model’s recognition effect on various fault types is analyzed by confusion matrix, with a focus on the recognition accuracy of easily confused fault types. At the same time, the back pressure consistency anomaly level data in the test set is input into the model, and the average absolute error between the anomaly level predicted by the model and the actual anomaly level is calculated. If the error exceeds the preset range, the relevant decision tree node rules for anomaly level prediction in the model are corrected based on the source of error, and the test evaluation is carried out again until the model’s recognition accuracy and anomaly level prediction accuracy on the test set meet the preset requirements.

[0124] Based on the above training and verification process, the random forest model that finally meets the performance requirements is determined as the back pressure anomaly identification model after training. The model's structural parameters, feature weight parameters, hyperparameters, and corresponding fault type label mapping relationships are solidified and stored through the model solidification algorithm. At the same time, a model training report is generated, recording the sample source, hyperparameter optimization process, performance evaluation results, and the correlation logic with the causal distribution mapping field of temperature and back pressure during the model training process, providing a basis for subsequent model maintenance and updates.

[0125] It should be further explained that the specific process of performing real-time antifreeze simulation control in this embodiment includes:

[0126] The historical air-cooled island fan operation status information is obtained and statistical algorithms are used to extract the fan's air volume, wind speed, wind direction status information at different times, as well as the spatial temperature gradient information and temporal temperature gradient change information under the corresponding operation status.

[0127] Based on the information on air volume, wind speed, and wind direction, as well as the spatial temperature gradient information under the corresponding operating conditions, the contribution of air volume, wind speed, and wind direction to the adjustment of spatial temperature gradient changes is obtained through factor analysis algorithm.

[0128] Based on the current wind turbine's air volume, operating time, wind speed, wind direction status information, as well as the corresponding spatial adjustment contribution and spatial temperature gradient information under the operating state, a spatial temperature gradient inversion function is constructed using a support vector machine. The spatial temperature gradient information includes the normal temperature gradient change area and the angle between the abnormal temperature gradient change and the X-axis in the inspection coordinate system.

[0129] Similarly, by using information on air volume, operating time, wind speed, wind direction, and the time-temperature gradient change information under the corresponding operating conditions, a time gradient inversion function is constructed.

[0130] It should be further explained that the construction and training process of the spatial temperature gradient inversion function and the temporal gradient inversion function in this embodiment includes:

[0131] Based on the historical air-cooled island fan operation status information, including the fan operation parameter records at different times and the temperature monitoring data of the target area of ​​the air-cooled island during the corresponding period, the abnormal data caused by sensor failure and data transmission interruption is removed by data cleaning algorithm. Then, the fan operation status information and temperature monitoring data are matched with a unified timestamp by time series alignment algorithm to ensure that the fan parameters at each moment correspond accurately with the corresponding temperature gradient information in the spatiotemporal dimension, and obtain a normalized historical associated dataset.

[0132] Based on the normalized historical association dataset, according to the spatial division rules of the inspection coordinate system, the fan status information of each location point at different times is extracted, including air volume parameters, wind speed parameters and wind direction parameters. At the same time, the spatial temperature gradient information of the corresponding location point is extracted, including the gradient value of the normal temperature gradient change area and the angle between the abnormal temperature gradient change area and the X-axis of the inspection coordinate system. The above fan status information and spatial temperature gradient information are used as initial feature variables. The difference in the dimensions between different variables is eliminated through feature standardization processing to form a spatial dimension feature set.

[0133] Based on the spatial dimension feature set, the three fan state variables of air volume parameter, wind speed parameter and wind direction parameter are reduced in dimension by factor analysis algorithm to eliminate multicollinearity among variables. The load coefficient of each fan state variable in the dimension of spatial temperature gradient change is calculated. Then, the load coefficient is converted into the adjustment contribution of each fan state parameter to the spatial temperature gradient change by normalization algorithm, and the influence weight of different fan parameters on the spatial temperature gradient is clarified.

[0134] Based on the spatial dimension feature set, the calculated adjustment contribution, and the wind turbine running time information in the historical correlation data, the air volume parameters, wind speed parameters, wind direction parameters, running time length at each moment are matched one-to-one with the spatial temperature gradient information at the corresponding location point through the sample matching algorithm, and a training sample set for the spatial temperature gradient inversion function is constructed. The wind turbine state parameters, running time length, and adjustment contribution are used as sample input features, and the spatial temperature gradient information is used as sample output labels. The training sample set is divided into a spatial dimension training subset and a spatial dimension validation subset using a stratified sampling algorithm.

[0135] Based on a spatial dimension training subset, a grid search algorithm is used to optimize the kernel function type, penalty coefficient, and gamma parameter of the support vector machine model to determine the model parameter combination suitable for spatial temperature gradient inversion. The input features of the spatial dimension training subset are then substituted into the support vector machine model for training. During training, weights are assigned to different input features based on their contribution, giving higher priority to wind turbine parameters that have a greater impact on spatial temperature gradients. After training, a spatial dimension validation subset is input into the model, and the model inversion accuracy is evaluated using mean squared error and mean absolute error indices. The model parameters are iteratively adjusted until the inversion accuracy meets a preset threshold, thus obtaining the spatial temperature gradient inversion function.

[0136] Based on the normalized historical correlation dataset, the fan status information of each consecutive time node is extracted, including air volume parameters, wind speed parameters, wind direction parameters and operating time length. At the same time, the time temperature gradient change information of the corresponding time node is extracted, including the temperature gradient change rate and the duration of the gradient change. The random fluctuations in the temperature gradient data are eliminated by the time series smoothing algorithm to form a time dimension feature set, ensuring that the data can reflect the continuous trend of temperature gradient change over time.

[0137] Based on the time dimension feature set, the wind volume parameters, wind speed parameters, wind direction parameters, running time length and corresponding time temperature gradient change information of each continuous time node are matched with the sample construction algorithm to construct the training sample set of the time temperature gradient inversion function. The fan state parameters and running time length are used as sample input features, and the time temperature gradient change information is used as sample output label. The time series segmentation algorithm is used to divide the training sample set into a time dimension training subset and a time dimension validation subset to ensure that the subset covers the time gradient change scenarios under different running lengths.

[0138] Based on the time-dimensional training subset, a parameter optimization method consistent with the construction of the spatial temperature gradient inversion function is adopted. The optimal kernel function and hyperparameters of the support vector machine model in the time dimension are determined by a grid search algorithm. The input features of the time-dimensional training subset are substituted into the model for training. During the training process, the impact of the running time on the change of the time-temperature gradient is focused on, and the model contribution of this parameter is enhanced by a dynamic weight adjustment algorithm. After training, the time-dimensional validation subset is input into the model to evaluate the inversion accuracy. The model parameters are iteratively optimized until the preset requirements are met, and the time-temperature gradient inversion function is obtained.

[0139] This process involves obtaining the spatial and temporal temperature gradient inversion functions. Historical data is regularized to ensure the accuracy and relevance of the basic data. Factor analysis is used to clarify the influence weights of fan parameters on the spatial temperature gradient. The nonlinear fitting capability of support vector machines is combined to achieve gradient inversion. Functions are constructed for both spatial and temporal dimensions, comprehensively covering the spatiotemporal variation characteristics of the air-cooled island's temperature gradient. The resulting inversion function can accurately invert temperature gradient changes based on the fan's operating status, providing precise quantitative basis for subsequent antifreeze control based on fan parameter adjustments. This effectively improves the targeting and real-time performance of antifreeze control for air-cooled islands, avoiding energy waste or antifreeze failure caused by blind regulation.

[0140] The temperature-backpressure causal distribution mapping field corresponding to the real-time target monitoring area is obtained. Combined with the preset temperature anomaly threshold value and the preset backpressure anomaly threshold value, the temperature anomaly time point and the backpressure anomaly time point are determined. The average of the temperature anomaly time point and the backpressure anomaly time point is used as the fan start time point.

[0141] It should be further explained that the specific steps for the wind turbine start-up time in this embodiment include:

[0142] Based on the real-time collected temperature and back pressure data of the air-cooled island target monitoring area, the two types of data are substituted into the verified temperature and back pressure causal distribution mapping field through the data synchronization algorithm, and the real-time spatiotemporal gradient information in the mapping field is updated. At the same time, the invalid spatiotemporal nodes caused by missing data in the mapping field are removed through the data integrity verification algorithm, so as to obtain the real-time updated temperature and back pressure causal distribution mapping field.

[0143] Based on the real-time updated temperature and back pressure causal distribution mapping field, the temperature data time series and back pressure data time series of all locations in the mapping field are extracted by the time dimension extraction algorithm according to the preset inspection coordinate system time scale, ensuring that each time node corresponds to complete global temperature and back pressure information, and obtaining temperature time series set and back pressure time series set;

[0144] Based on the preset temperature anomaly threshold, the anomaly detection algorithm compares the temperature value of each location in the temperature time series set with the threshold value at each time node. When the temperature value of a location exceeding a preset proportion at a certain time node is lower or higher than the temperature anomaly threshold, and the duration of this state reaches a preset threshold, the time node is marked as a temperature anomaly time node. This process is repeated to traverse all time nodes to obtain a set of temperature anomaly time nodes.

[0145] Based on the preset back pressure anomaly threshold, the same anomaly judgment algorithm as the temperature anomaly time point judgment is adopted. The back pressure value of each location point in the back pressure time series is compared with the threshold value at each time point. When the back pressure value of a location point at a certain time point exceeds the preset proportion of the back pressure anomaly threshold or is lower than the back pressure anomaly threshold, and the duration of this state reaches the preset threshold, the time point is marked as a back pressure anomaly time point. After traversing all time points, the set of back pressure anomaly time points is obtained.

[0146] Based on the time continuity filtering algorithm, the set of abnormal temperature time points and the set of abnormal back pressure time points are filtered separately. Time points that exist in isolation and have no adjacent abnormal time points are removed from the set, and the continuously distributed abnormal time point subsets are retained to obtain the filtered temperature abnormal time point subsets and back pressure abnormal time point subsets.

[0147] Based on the time format unification algorithm, all time nodes in the filtered subsets of temperature anomaly time points and back pressure anomaly time points are converted into a unified timestamp format. The arithmetic mean calculation algorithm is used to calculate the mean of all timestamps in the two subsets respectively, so as to obtain the mean of temperature anomaly time points and the mean of back pressure anomaly time points.

[0148] Based on the temperature and back pressure causal correlation verification algorithm, the mean values ​​of temperature and back pressure anomalies at certain time points are substituted into the real-time updated temperature and back pressure causal distribution mapping field to verify whether the causal correlation between temperature and back pressure before and after the two mean time points conforms to the preset abnormal causal logic. If it does, the arithmetic mean of the two means is determined as the candidate wind turbine start-up time point; if it does not, the subset of abnormal time points is re-selected and the mean calculation is repeated until the candidate time point satisfies the causal correlation verification.

[0149] Based on the time synchronization algorithm of the fan control system, the candidate fan start time is converted into a time command format that the fan control system can recognize. The validity of the time command is confirmed by the command verification algorithm. After confirmation, the time command is output as the final fan start time and synchronized to the air-cooled island fan control module.

[0150] Simultaneously, based on the temperature-back pressure causal distribution mapping field corresponding to the real-time target monitoring area and the anomaly type information matrix, the causal type of back pressure anomaly corresponding to the wind turbine start-up time point and the identification accuracy are predicted.

[0151] Simultaneously, based on the number of points at each time point that reach the critical value of temperature anomaly and the critical value of back pressure anomaly in the temperature-back pressure causal distribution mapping field, the area of ​​the temperature anomaly monitoring interval or the area of ​​the back pressure anomaly monitoring interval corresponding to each time point is determined.

[0152] Based on the ratio of the area of ​​the abnormal monitoring interval corresponding to each time point to the area of ​​the working area covered by the maximum rated power of each wind turbine, combined with the ratio of the back pressure gradient to the temperature gradient at the corresponding time point and the comprehensive fuzzy algorithm, the consistency anomaly level index at each time point is obtained.

[0153] It should be further explained that one implementation detail of the consistency anomaly level index at each time point in this embodiment includes:

[0154] Based on the real-time monitoring of the target monitoring area's temperature backpressure causal distribution mapping field and the constructed anomaly type information matrix, a data matching algorithm is used to compare the temperature backpressure gradient correlation features at each time point in the mapping field with the gradient features of each backpressure anomaly causal type in the anomaly type information matrix. Combined with a similarity calculation algorithm, the most matching backpressure anomaly causal type at each time point is determined. At the same time, a probability statistical algorithm is used to calculate the recognition accuracy of this causal type, thus obtaining the backpressure anomaly causal type and recognition accuracy data corresponding to each time point.

[0155] Based on preset temperature anomaly thresholds and back pressure anomaly thresholds, a spatial node traversal algorithm is used to check the temperature and back pressure values ​​of all locations in the patrol coordinate system at each time point in the temperature and back pressure causal distribution mapping field. The number of locations whose temperature values ​​reach or exceed the temperature anomaly threshold and the number of locations whose back pressure values ​​reach or exceed the back pressure anomaly threshold are counted. Combined with the spatial grid area parameters of the patrol coordinate system, a surface accumulation algorithm is used to calculate the total spatial area (i.e., temperature anomaly monitoring interval area) covered by temperature anomaly locations and the total spatial area (i.e., back pressure anomaly monitoring interval area) covered by back pressure anomaly locations at each time point, thus obtaining the temperature anomaly monitoring interval area and the back pressure anomaly monitoring interval area at each time point.

[0156] Based on the data of the coverage area of ​​each fan's maximum rated power stored in the air-cooled island fan parameter database, the area of ​​the temperature anomaly monitoring interval and the area of ​​the back pressure anomaly monitoring interval at each time point are matched with the area of ​​the coverage area of ​​each fan's maximum rated power through a data association algorithm. Combined with the ratio calculation algorithm, the ratio of the area of ​​the temperature anomaly monitoring interval to the coverage area of ​​the corresponding fan and the ratio of the area of ​​the back pressure anomaly monitoring interval to the coverage area of ​​the corresponding fan are calculated to obtain the two types of area ratio data at each time point.

[0157] Based on the backpressure gradient data and temperature gradient data at each location point at each time point in the temperature-backpressure causal distribution mapping field, the average backpressure gradient and the average temperature gradient at each location point in the whole domain at each time point are calculated by the gradient mean calculation algorithm. The ratio of the average backpressure gradient to the average temperature gradient is calculated by the ratio calculation algorithm to obtain the backpressure-temperature gradient ratio data at each time point. At the same time, the instantaneous fluctuations in the ratio data are eliminated by the data smoothing algorithm to ensure that the data reflects the overall trend of gradient correlation at each time point.

[0158] Based on the data standardization algorithm, the two types of area ratio data and back pressure temperature gradient ratio data at each time point are processed, and the three types of ratio data are uniformly converted into a preset standardized numerical range. This eliminates the problem of unbalanced evaluation weights caused by the difference in dimensions of different types of ratios, and obtains the standardized area ratio and standardized gradient ratio at each time point.

[0159] Based on the causal type and identification accuracy of the back pressure anomaly at each time point, the weight determination algorithm is used to analyze the influence weight of the area ratio and gradient ratio on the degree of anomaly under the causal type. If the identification accuracy is high and the causal type is a temperature-dominant anomaly (such as freezing due to low temperature), the weight of the area ratio related to the temperature anomaly is increased; if it is a back pressure-dominant anomaly, the weight of the back pressure-temperature gradient ratio is increased, thus obtaining the dynamic weight coefficients of the three standardized ratios at each time point.

[0160] A fuzzy evaluation model is constructed based on a comprehensive fuzzy algorithm. The standardized ratio and corresponding dynamic weight coefficient at each time point are input into the model. The membership degree of each standardized ratio in the preset abnormality level range (such as low, medium, high, and extremely high) is calculated by the fuzzy membership degree function. The fuzzy synthesis operator is used to weight and fuse each membership degree to obtain the fuzzy comprehensive evaluation result at each time point.

[0161] Based on defuzzification algorithms (such as the centroid method and the maximum membership method), the fuzzy comprehensive evaluation results at each time point are converted into specific level index values ​​or level labels. At the same time, the trend consistency verification algorithm is used to compare the level index with the anomaly monitoring area change trend and gradient ratio change trend at the corresponding time point to verify whether the level index conforms to the anomaly development law. If there is a deviation, the fuzzy evaluation model parameters are readjusted and the calculation is repeated to finally obtain the accurate and consistent anomaly level index at each time point.

[0162] This process integrates multi-dimensional information such as temperature and back pressure causal relationships, anomaly spatial coverage, fan control capabilities, and gradient synergy to avoid the one-sidedness of single-factor evaluation. It dynamically allocates weights based on the causal type of back pressure anomalies, making the evaluation results more consistent with actual fault characteristics. Standardization and fuzzy algorithms address the dimensional differences of different parameters, ensuring the scientific nature of the evaluation. The resulting consistent anomaly level index accurately quantifies the overall severity and synergy of air-cooled island anomalies at each time point, providing a quantitative basis for subsequent fan start-up timing and control intensity determination. This effectively improves the targeting and decision-making efficiency of air-cooled island antifreeze management, while reducing the risk of excessive energy consumption or antifreeze failure due to misjudgment of anomaly levels.

[0163] Based on the spatial and temporal gradient changes of temperature under the corresponding index level, and combined with the corresponding spatial temperature gradient inversion function and temporal gradient inversion function, the real-time air volume and wind speed of each fan in the corresponding working area are obtained.

[0164] At the fan start-up time, the air volume, air direction and wind speed are obtained in real time, and combined with the deep strategy algorithm, antifreeze control commands are generated for real-time antifreeze control.

[0165] Real-time monitoring of the adjusted temperature-backpressure causal distribution mapping field yields the ratio of the area of ​​the abnormal temperature monitoring interval at the corresponding time point after adjustment to the area of ​​the abnormal temperature monitoring interval at the corresponding time point predicted before adjustment. A temperature antifreeze regulation efficiency index is constructed, and the temperature spatial gradient field at the corresponding time point after adjustment is monitored in conjunction with a uniformity algorithm to calculate the temperature gradient change uniformity at the corresponding time point.

[0166] The temperature antifreeze regulation efficiency index and temperature gradient change uniformity at the corresponding time points are mapped to the HSV color space in real time for visualization processing, to obtain a temperature anomaly regulation efficiency uniformity comparison layer at each time point, and mapped to the temperature-backpressure causal distribution mapping field for three-dimensional spatial and temporal display.

[0167] The temperature antifreeze regulation efficiency index and the temperature gradient change uniformity at the corresponding time point are compared with the preset temperature antifreeze regulation efficiency index threshold and temperature gradient change uniformity threshold, respectively. If either the temperature antifreeze regulation efficiency index or the temperature gradient change uniformity at the corresponding time point does not meet the corresponding threshold, the corresponding temperature abnormal regulation efficiency comparison layer, the antifreeze regulation efficiency index, and the temperature gradient change uniformity at the corresponding time point are fed back to the deep strategy algorithm to adjust the antifreeze control command at the corresponding time point in real time.

[0168] Simultaneously, based on the currently predicted temperature antifreeze regulation efficiency index and the temperature gradient change uniformity at the corresponding time point, the temperature-backpressure gradient change mapping in the temperature-backpressure causal distribution mapping field is used to obtain the currently predicted backpressure antifreeze regulation efficiency index and the backpressure gradient change uniformity at the corresponding time point.

[0169] Based on the currently predicted back pressure antifreeze regulation efficiency index and the uniformity of back pressure gradient change at the corresponding time point, and the temperature antifreeze regulation efficiency index and the uniformity of temperature gradient change at the corresponding time point, the consistency anomaly level index of temperature and back pressure is adjusted to obtain the consistency anomaly level index of temperature and back pressure at the current moment. The anomaly type of the air-cooled island is judged in real time by the back pressure anomaly identification model that has been trained. When it is determined to be an antifreeze anomaly again, the above antifreeze control command adjustment process is repeated until the temperature antifreeze regulation efficiency index and the uniformity of temperature gradient change at the corresponding time point meet the corresponding threshold in real time.

[0170] When it is determined again that it is not an antifreeze anomaly, the temperature and back pressure consistency anomaly level index is combined with the temperature-back pressure causal distribution mapping field. The corresponding type is searched and located according to the direction of the increase in the deviation between the temperature gradient fluctuation and the back pressure gradient fluctuation. The corresponding non-antifreeze anomaly location point is obtained and mapped into the temperature-back pressure causal distribution mapping field for real-time display and early warning.

[0171] It should be further explained that the specific implementation details of the real-time antifreeze simulation control in this embodiment include:

[0172] Based on real-time collected actual operating parameters of the air-cooled island fan and theoretical fan parameters output by the spatial temperature gradient inversion function and the temporal temperature gradient inversion function, the absolute deviation between actual and theoretical air volume, the relative deviation between actual and theoretical wind speed, and the angular deviation between actual and theoretical wind direction are calculated using the term-by-term difference method. A weighted average method is then used to calculate the root mean square (RMS) value of each deviation, with the weighting coefficients determined based on the importance of each parameter to the temperature gradient. This RMS value is then compared with preset fan parameter deviation thresholds at multiple levels: if the RMS value is within the first-level threshold range, the fan operating parameters are considered to fully meet expectations; if it is within the second-level threshold range, it is considered to basically meet expectations but requires continuous monitoring; if it exceeds the second-level threshold, it is marked as a deviation from expectations, and in-depth diagnostic procedures such as impeller surface wear detection and inverter output characteristic testing are initiated.

[0173] Based on the preset dynamic lag correlation time in the causal distribution mapping field of temperature and back pressure, the sliding window cross-correlation method is used to accurately extract the starting time points of abnormal temperature and back pressure changes through time series analysis algorithms. The starting time point is defined as the time point in the data sequence that first continuously exceeds the abnormal threshold. The absolute value of the deviation between the actual lag time and the theoretical lag time is calculated using the absolute time difference method through the lag deviation calculation algorithm, and the relative deviation percentage is also calculated. The absolute deviation and relative deviation are jointly judged with the two thresholds: if both are less than or equal to the threshold, the causal correlation is determined to be completely in line with the dynamic law; if either exceeds the threshold, it is marked as abnormal, and analysis programs such as tube bundle flow section detection and fan response time testing are initiated.

[0174] Based on a database of successful historical anti-freezing regulation cases for air-cooled islands, a case matching algorithm is used to calculate the similarity between the current regulation parameters and historical cases using a multi-dimensional feature-weighted Euclidean distance method. Feature dimensions include temperature gradient change rate, back pressure stabilization time, and energy consumption indicators. A difference analysis algorithm is used to analyze the differences between the current data and the optimal matching case using a point-by-point comparison method, with the difference assessment using percentage deviation calculation. A three-level difference threshold system is established: if all difference items are within the first-level threshold, it is determined that the data fully conforms to successful experience; if some are within the second-level threshold, it is determined that the data basically conforms but requires optimization; if there are difference items exceeding the third-level threshold, a parameter correction procedure is initiated, including fan speed adjustment and wind direction optimization.

[0175] Based on the sub-region division of the inspection coordinate system, the Voronoi diagram method is used to ensure that each sub-region contains a complete tube bundle unit while guaranteeing that the fan coverage area does not overlap. The temperature gradient distribution uniformity within each sub-region is calculated using a local gradient calculation algorithm and the finite difference method. The uniformity difference between adjacent sub-regions is calculated using a synergy analysis algorithm and a region boundary gradient continuity analysis method. A dynamic synergy threshold system is established, which automatically adjusts the threshold size according to changes in ambient temperature: if the differences between all adjacent regions are within the threshold, the synergy is considered good; if a difference exceeding the threshold is detected, region-specific adjustments are initiated, including adjusting the fan tilt angle and modifying the airflow distribution.

[0176] Based on real-time external environmental parameters, a multiple linear regression model is used to quantify the influence coefficients of ambient temperature, wind speed, and humidity on efficiency indicators and uniformity. The correction calculation uses real-time parameters to be substituted into the correction model to obtain the indicator values ​​after environmental compensation. The threshold adjustment adopts an environmental parameter adaptive method, dynamically adjusting the evaluation criteria according to environmental changes: if the corrected indicators meet the adjusted thresholds, the environmental adaptability is judged to be good; if not, environmental compensation adjustment is initiated, including modifying the fan start-stop strategy and adjusting the steam flow rate.

[0177] Based on the tube bundle design parameters and real-time operating data, the heat flux density distribution at various points on the tube bundle surface is calculated using Fourier's law of thermal conductivity combined with the finite element discretization method. The critical heat flux density for antifreeze is determined using the tube bundle icing critical condition calculation method, taking into account tube material characteristics, medium parameters, etc. A heat flux density safety margin assessment system is established: if the actual heat flux density is higher than the critical value and maintains a safety margin, the heat exchange capacity is considered sufficient; if it is lower than the critical value, enhanced heat exchange measures are initiated, including increasing fan speed and optimizing steam distribution.

[0178] Based on continuous time series data, the cumulative effect of prediction error is calculated using the exponentially weighted moving average method; the model stability is assessed using the error trend analysis method to monitor the change pattern of error over time; a multi-level early warning mechanism is established: if the cumulative error is within the normal range, the model is determined to be stable; if the trend exceeds the threshold, the online learning program of the model is activated to update the model parameters using the latest data.

[0179] Based on the analysis of temperature change curves, the second derivative of the temperature curve at each point is calculated using numerical differentiation. Convergence is determined using a dual-condition system: both the second derivative must approach zero, and the temperature value must be within a standard range. A weighted proportional method is used statistically, assigning higher weights to key regions. A convergence evaluation system is established: if the convergence percentage is higher than a threshold and convergence is achieved in all major regions, the regulation is considered stable; if the criteria are not met, an enhanced regulation procedure is initiated, including extending the regulation time and increasing the regulation intensity.

[0180] The real-time control simulation system for antifreeze in air-cooled islands proposed in this application forms a fully intelligent solution from precise monitoring and scientific analysis to dynamic regulation through multi-module collaboration and multi-technology integration. Its beneficial effects can be gradually derived from the underlying technology to the system application layer. At the spatial positioning level, the system constructs a patrol coordinate system based on the three-dimensional structure of the air-cooled island heat sink tube bundle array and the fan operating characteristics. By establishing the association between tube bundle units and coordinate points through a topology network and combining it with densification processing for high-incidence anomaly areas, it not only ensures full coverage of the monitoring area but also improves the data acquisition accuracy of key areas. This solves the problems of coordinate chaos and insufficient monitoring of key areas in traditional monitoring, providing a spatial benchmark for subsequent precise analysis of the temperature field and back pressure field. At the data processing and causal correlation level, the system constructs spatiotemporal distribution fields of temperature and back pressure by extracting historical anomaly data. It then combines correlation analysis algorithms and radial kernel functions to construct a temperature-back pressure causal distribution mapping field. Bidirectional prediction verification enhances the accuracy of causal correlation, overcoming the limitation of traditional single-parameter monitoring in establishing dynamic correlations. This allows for precise capture of the coordinated changes in temperature and back pressure across the spatiotemporal dimensions, avoiding misjudgments caused by single-parameter anomalies and providing a scientific causal basis for anomaly identification. At the anomaly identification level, the system constructs an anomaly type information matrix based on features extracted from fault scenario texts. Combining random forest algorithms with simulation training, it forms a back pressure anomaly identification model. By fusing gradient correlation data from the temperature-back pressure causal distribution mapping field with a consistency anomaly level index, it achieves accurate differentiation of different anomaly types. Furthermore, incremental training addresses rare anomaly scenarios, solving the problems of weak generalization ability and easy confusion with similar anomalies in traditional identification models. This ensures accurate anomaly type determination under various operating conditions, providing guidance for targeted regulation. At the level of quantitative control parameters, the system constructs spatial temperature gradient inversion functions and temporal temperature gradient inversion functions based on historical wind turbine operation data and statistical algorithms. Through factor analysis, it clarifies the contribution of wind turbine parameters to the adjustment of temperature gradient, transforming wind turbine control from empirical operation to quantitative calculation. This avoids energy waste or antifreeze failure caused by blind parameter setting in traditional control. At the same time, combined with the scientific method of determining the wind turbine start-up time, the system verifies the rationality of the start-up timing by verifying the mean of abnormal time points and causal relationships. This ensures both the timeliness of the antifreeze response and avoids energy loss caused by starting too early or too late.At the real-time control closed-loop level, the system constructs a dynamic feedback mechanism by monitoring regulation efficiency and temperature gradient uniformity. Combined with HSV visualization, it achieves an intuitive presentation of the regulation effect. At the same time, it introduces multi-dimensional supplementary judgments such as fan parameter deviation judgment, dynamic verification of causal relationships, historical case matching, sub-regional synergy analysis, environmental adaptability correction, heat flux density assessment, model stability monitoring, and regulation convergence judgment. This forms a closed-loop regulation optimization process that can promptly detect deviations in the regulation process, such as fan parameters deviating from expectations, sub-regional regulation imbalance, and insufficient environmental adaptability. It also adjusts control commands in real time through deep strategy algorithms to ensure that the regulation effect continuously meets the antifreeze requirements. In addition, it can accurately locate non-antifreeze faults and issue warnings when there are non-antifreeze anomalies, preventing the fault from escalating. From the perspective of overall application effect, the system has realized the transformation of air-cooled island antifreeze from manual dependence to intelligent autonomy. Through precise monitoring, it has reduced the risk of anomaly omissions and misjudgments, ensuring the safe operation of the unit; through quantitative control, it has reduced energy consumption such as fans and water resources, achieving energy conservation and consumption reduction; through dynamic closed-loop control, it has improved the adaptability of antifreeze regulation, and can cope with different environmental conditions and changes in unit load; through intelligent operation, it has reduced manual intervention, lowered operation and maintenance costs and safety hazards, and ultimately provided comprehensive support for the efficient, safe and economical operation of the air-cooled island, which meets the development needs of intelligent manufacturing and energy conservation and consumption reduction.

[0181] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

Claims

1. A real-time control simulation system for anti-freezing of an air-cooling island, characterized in that, include: Monitoring module, analysis module, inversion module, and response simulation module; The monitoring module is used to monitor the first temperature field information, back pressure distribution field and air-cooled island fan operation status information in the target area in real time. The analysis module is used to calculate the temperature spatiotemporal distribution field and the back pressure spatiotemporal distribution field representing the target area based on the first temperature field information and the back pressure distribution field, and to calculate the temperature-back pressure causal distribution mapping field representing the real-time state of the target area based on the temperature spatiotemporal distribution field and the back pressure spatiotemporal distribution field combined with the correlation analysis algorithm. The inversion module is used to determine the type of back pressure anomaly based on the temperature-back pressure causal distribution mapping field of the target area combined with the back pressure anomaly identification model. When the anomaly is determined to be antifreeze anomaly, the inversion mapping space is calculated based on the temperature-back pressure causal distribution mapping field of the target area combined with the operation inversion function of the air-cooled island fan. The inversion mapping space includes at least the warning level index, fan start time, air volume, wind direction and wind speed. The operation inversion function of the air-cooled island fan is used to obtain the corresponding fan start time, running time, wind direction and wind speed based on the maximum working area of ​​each fan and the current temperature and back pressure change trend, so that the temperature and back pressure of the target area are kept in the standard temperature range and the standard back pressure range in real time. The response simulation module is used to perform real-time antifreeze simulation control based on the real-time inversion mapping space combined with the preset enhancement adjustment strategy and simulation algorithm, so that the real-time temperature and back pressure of the monitored target area meet the preset temperature standard range and back pressure standard range.

2. The real-time control simulation system for anti-freezing of an air cooling island according to claim 1, wherein The process of constructing the temperature-backpressure causal distribution mapping field includes: Based on the topological relationships of the target monitoring area, determine the patrol coordinate system of the target monitoring area; Obtain the temperature information of each location point corresponding to each anomaly type in the historical inspection coordinate system, and determine the temperature spatial gradient field of adjacent location points.

3. The real-time control simulation system for anti-freezing of an air cooling island according to claim 2, wherein The process of constructing the temperature-backpressure causal distribution mapping field also includes: Obtain temperature trend information at each location point at different time points, and obtain the temperature time gradient field and temperature change curve at each location point; The temperature time gradient field and temperature change curve at each location point are mapped to the corresponding location point in the temperature spatial gradient field to obtain the temperature spatiotemporal distribution field. For the temperature change curves at locations with spatial gradient anomalies, chromaticity and luminance are marked according to the magnitude of the spatial temperature gradient anomaly deviation and the HSV color space.

4. The real-time control simulation system for anti-freezing of an air cooling island according to claim 3, wherein The process of constructing the temperature-backpressure causal distribution mapping field also includes: Obtain the back pressure information of each location point under the corresponding anomaly type in the patrol coordinate system, and obtain the back pressure spatiotemporal distribution field in the same way as the temperature spatiotemporal distribution field. The spatial coordinates of the back pressure spatiotemporal distribution field and the temperature spatiotemporal distribution field are aligned to obtain the aligned temperature-back pressure spatial gradient pair and temperature-back pressure temporal gradient pair. Based on the aligned temperature-backpressure spatial gradient pairs, the correlation degree of temperature-backpressure spatial gradient changes is obtained through correlation analysis algorithms.

5. A real-time control simulation system for anti-freezing of an air-cooled island as claimed in claim 4, wherein The process of constructing the temperature-backpressure causal distribution mapping field also includes: Based on the aligned temperature-back pressure time gradient pair, obtain the temperature-back pressure time gradient change correlation; The temperature-back pressure time gradient change correlation and the spatial gradient change correlation are combined as weights to construct a temperature-back pressure corresponding time and space gradient change mapping, and the established temperature-back pressure corresponding time and space gradient change mapping is embedded into the temperature spatiotemporal distribution field to obtain a temperature-back pressure causal distribution mapping field. Based on the temperature-back pressure causal distribution mapping field combined with the real-time collected temperature gradient data and back pressure gradient data, bidirectional prediction verification is performed through a causal inference algorithm to obtain a temperature-back pressure causal distribution mapping field for each abnormal type after verification.

6. A real-time control simulation system for anti-freezing of an air-cooled island as claimed in claim 5 wherein, The specific construction process of the back pressure abnormality identification model includes: Obtain the actual operation fault scene text information of the air cooling island, extract the corresponding fault type and the temperature and back pressure corresponding spatial gradient change state and time gradient change state under the corresponding fault type, and construct an abnormal type information matrix; Based on the abnormal type information matrix combined with a statistical analysis algorithm, obtain the interval of the consistency abnormality level index of the corresponding temperature and back pressure under each abnormal type; the consistency abnormality level index of the temperature and back pressure is obtained by multiplying the ratio of the spatial gradient change state of the temperature and back pressure at each time point by the ratio of the abnormal temperature area to the abnormal back pressure area at the corresponding time point; Based on the abnormal type information matrix and the consistency abnormality level index of the temperature and back pressure, a random forest tree algorithm, a verified temperature-back pressure causal distribution mapping field, and a simulation algorithm are combined for simulation training to obtain a trained back pressure abnormality identification model.

7. A real-time control simulation system for anti-freezing of an air-cooled island as claimed in claim 5 wherein, The specific process of real-time anti-freezing simulation control includes: Obtain the historical air cooling island fan operation state information and combine a statistical algorithm to extract the corresponding air volume, air speed, air direction state information at different times, and the spatial temperature gradient information and time temperature gradient change information under the corresponding operating state; Based on the air volume, air speed, air direction state information and the spatial temperature gradient information under the corresponding operating state, a factor analysis algorithm is used to obtain the adjustment contribution degree of air volume, air speed and air direction to the spatial temperature gradient change; Based on the current time air volume, running time length, air speed, air direction state information, corresponding spatial adjustment contribution degree and spatial temperature gradient information under the operating state, a support vector machine is used to construct a spatial temperature gradient inversion function. Similarly, the air volume, running time length, air speed, air direction state information and time temperature gradient change information under the corresponding operating state are used to construct a time gradient inversion function.

8. The real-time control simulation system for anti-freezing of an air-cooled island according to claim 7, wherein The specific process of real-time anti-freezing simulation control further includes: Obtain the temperature-back pressure causal distribution mapping field corresponding to the real-time target monitoring area, combine the preset temperature abnormality threshold and the preset back pressure abnormality threshold, determine the temperature abnormality time point and the back pressure abnormality time point, and take the mean value of the temperature abnormality time point and the back pressure abnormality time point as the fan start time point; Meanwhile, based on the temperature-back pressure causal distribution mapping field corresponding to the real-time target monitoring area and the abnormal type information matrix, a back pressure abnormal causal type corresponding to a fan starting time point and an identification accuracy are predicted; Meanwhile, based on the number of arrival temperature abnormal critical value and back pressure abnormal critical value position points of each time point of the temperature-back pressure causal distribution mapping field, a temperature abnormal monitoring interval area or a back pressure abnormal monitoring interval area corresponding to each time point is determined; Based on a ratio of an abnormal monitoring interval area corresponding to each time point to an area of a maximum rated power corresponding to each fan, a ratio of a back pressure gradient to a temperature gradient corresponding to each time point, and a comprehensive fuzzy algorithm, a consistency abnormality level index of each time point is obtained.

9. The real-time control simulation system for anti-freezing of an air-cooled island according to claim 8, wherein The specific process of the real-time anti-freezing simulation control further includes: Based on the temperature spatial gradient change value and the time gradient change value under the corresponding index level, the corresponding spatial temperature gradient inversion function and the time gradient inversion function are combined to obtain the real-time air volume and air speed of each fan in the corresponding working area; At the fan starting time point, the air volume, air direction and air speed are obtained in real time, and a depth strategy algorithm is combined to generate an anti-freezing control instruction for real-time anti-freezing control; The temperature anti-freezing regulation efficiency index is constructed by obtaining a ratio of an adjusted abnormal temperature monitoring interval area corresponding to each time point to a predicted abnormal temperature monitoring interval area corresponding to each time point before adjustment, and monitoring the temperature spatial gradient field under the corresponding time point after adjustment and combining a uniformity algorithm to calculate the temperature gradient change uniformity under the corresponding time point. The temperature anti-freezing regulation efficiency index and the temperature gradient change uniformity of the corresponding time point are mapped to the HSV color space in real time for visual processing to obtain a temperature abnormality regulation efficiency uniformity comparison layer of each time point, and are mapped into the temperature-back pressure causal distribution mapping field for three-dimensional space and time display.

10. The real-time control simulation system for anti-freezing of an air-cooled island according to claim 9, wherein The specific process of the real-time anti-freezing simulation control further includes: The temperature anti-freezing regulation efficiency index and the temperature gradient change uniformity of the corresponding time point are compared with the preset temperature anti-freezing regulation efficiency index threshold and the temperature gradient change uniformity threshold, respectively. If either of the temperature anti-freezing regulation efficiency index and the temperature gradient change uniformity of the corresponding time point does not satisfy the corresponding threshold, the corresponding temperature abnormality regulation efficiency comparison layer, the anti-freezing regulation efficiency index and the temperature gradient change uniformity under the corresponding time point are fed back to the depth strategy algorithm to adjust the anti-freezing control instruction of the corresponding time point in real time; Meanwhile, according to the current predicted temperature anti-freezing regulation efficiency index and the temperature gradient change uniformity of the corresponding time point, the temperature-back pressure corresponding time and space gradient change mapping in the temperature-back pressure causal distribution mapping field is obtained to obtain the current predicted back pressure anti-freezing regulation efficiency index of the corresponding time point and the back pressure gradient change uniformity of the corresponding time point. Based on the current predicted back pressure anti-freezing regulation efficiency index and the corresponding time point back pressure gradient change uniformity and the temperature anti-freezing regulation efficiency index and the corresponding time point temperature gradient change uniformity, the temperature and back pressure consistency anomaly level index is adjusted, the temperature and back pressure consistency anomaly level index corresponding to the current time is obtained, and the trained back pressure anomaly recognition model is used for real-time air cooling island anomaly type judgment; when the anti-freezing anomaly is judged again, the above anti-freezing control instruction adjustment process is repeated until the temperature anti-freezing regulation efficiency index and the temperature gradient change uniformity of the corresponding time point satisfy the corresponding threshold in real time; When it is judged again that it is not an anti-freezing anomaly type, the temperature and back pressure consistency anomaly level index is combined with the temperature-back pressure causal distribution mapping field, the corresponding type is searched and positioned according to the direction of the increase of the temperature gradient fluctuation and the back pressure gradient fluctuation deviation, the corresponding non-anti-freezing anomaly position point is obtained, and the corresponding non-anti-freezing anomaly position point is mapped into the temperature-back pressure causal distribution mapping field for real-time display and early warning.

Images