An intelligent management and control system and a coordinated control method for an inter-cooling and air-cooling island

Abstract

The application discloses an intelligent management and control system for an air-cooled island and a coordination control method, and belongs to the technical field of intelligent operation and maintenance of equipment, which comprises the following steps: collecting key operation parameters in real time by building an intelligent sensing network, calculating fin heat exchange efficiency attenuation coefficient and back pressure early warning value; when the heat exchange efficiency attenuation coefficient is lower than a threshold value or the back pressure is out of limit, starting a linkage control mechanism of an intelligent cleaning system and a smart spraying system synchronously; constructing a control effect evaluation matrix based on the back pressure-attenuation coefficient coupling degree, dynamically adjusting the spraying intensity and the cleaning frequency, and switching to a normal monitoring mode after the parameters return to normal and continue to be stable; the application innovatively realizes the collaborative intelligent control of cleaning and spraying, improves the air-cooled island heat exchange efficiency through multi-parameter fusion evaluation and closed-loop feedback mechanism, reduces energy consumption, and simultaneously has self-adaptive abnormal monitoring and rapid response capability, thereby providing an efficient intelligent operation and maintenance solution for the air-cooled system of a thermal power generating unit.

Description

Technical Field

[0001] This invention belongs to the field of intelligent equipment operation and maintenance technology, specifically a smart management and control system and coordination control method for indirect air-cooled islands. Background Technology

[0002] Traditional air-cooled island control systems have significant limitations, primarily due to their reliance on single-parameter threshold triggering mechanisms, which fail to comprehensively reflect the actual operating status of the equipment. Furthermore, the cleaning and spraying systems operate independently, lacking coordinated control capabilities. Existing technologies employ fixed control strategies, which struggle to adapt to dynamically changing operating conditions, particularly under extreme weather conditions. Moreover, the lack of a closed-loop feedback mechanism prevents quantitative evaluation of control effectiveness, leading to increased energy waste and equipment wear. With the advancement of smart power plant construction, this extensive control model can no longer meet the demands of refined operation and maintenance, necessitating the development of a new control system capable of multi-parameter fusion and intelligent coordination. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes an intelligent management and control system and coordinated control method for indirect air-cooled islands. The system includes: real-time acquisition of key operating parameters through an intelligent sensing network to calculate the fin heat exchange efficiency attenuation coefficient and back pressure warning value; simultaneous activation of the linkage control mechanism between the intelligent cleaning system and the intelligent spraying system when the heat exchange efficiency attenuation coefficient is below the threshold or the back pressure exceeds the limit; construction of a control effect evaluation matrix based on the back pressure-attenuation coefficient coupling degree to dynamically adjust the spraying intensity and cleaning frequency until the parameters return to normal and remain stable, after which the system switches to conventional monitoring mode. This invention innovatively achieves coordinated intelligent control of cleaning and spraying, improving the heat exchange efficiency of the air-cooled island and reducing energy consumption through multi-parameter fusion evaluation and closed-loop feedback mechanisms. It also possesses adaptive anomaly monitoring and rapid response capabilities, providing an efficient and intelligent operation and maintenance solution for the air-cooled systems of thermal power generating units.

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

[0005] A coordinated control method for an intercooled air-cooled island includes:

[0006] S1: Build an intelligent sensing network to collect key operating parameters in real time, construct an evaluation model based on the key operating parameters, and calculate the fin heat exchange efficiency decay coefficient and back pressure warning value; the key operating parameters include fin temperature, back pressure, ambient temperature and contaminant thickness.

[0007] S2: When the fin heat exchange efficiency attenuation coefficient is lower than the preset value or the back pressure reaches the back pressure warning value, the linkage control mechanism of the intelligent cleaning system and the intelligent spraying system is activated simultaneously, and the initial cleaning and spraying control commands are output.

[0008] S3: Based on the current back pressure and fin heat exchange efficiency attenuation coefficient, construct a control effect evaluation matrix based on the coupling degree of back pressure-attenuation coefficient and evaluate the control effect level of cleaning and spraying. Adjust the spraying intensity or cleaning frequency according to the control effect level and output the adjusted control parameters.

[0009] S4: Until the ambient temperature and back pressure reach the preset normal conditions and remain so for a preset duration, output a control endpoint signal and switch to the normal monitoring mode based on the control endpoint signal; the normal monitoring mode maintains real-time acquisition and monitoring of the key operating parameters, and once an abnormal change in the key operating parameters is detected, immediately switch back to the intelligent control mode and restart the intelligent cleaning system or the intelligent spraying system.

[0010] Specifically, the evaluation model includes a parameter-attenuation coefficient mapping model and a back pressure warning value calculation model. The construction process of the evaluation model includes:

[0011] Acquire historical key operating parameters and perform noise filtering and timing alignment processing;

[0012] A multi-tree collaborative model was adopted, with fin temperature, ambient temperature, and contaminant thickness as input features, and the measured fin heat transfer efficiency attenuation coefficient as a label for training, to obtain a parameter-attenuation coefficient mapping relationship model.

[0013] Based on historical back pressure data, the mean and standard deviation of back pressure under different ambient temperatures are calculated using the sliding window method. Combined with the equipment safe operation threshold, a back pressure early warning value calculation model is constructed.

[0014] Specifically, the multi-tree collaborative model is trained using fin temperature, ambient temperature, and contaminant thickness as input features, and the measured fin heat transfer efficiency attenuation coefficient as a label, to obtain a parameter-attenuation coefficient mapping relationship model, including:

[0015] A1: Obtain fin temperature, ambient temperature, contaminant thickness, and measured fin heat transfer efficiency attenuation coefficient, and combine them to form an input parameter set. Among them, fin temperature, ambient temperature, and contaminant thickness are used as input features, and the measured fin heat transfer efficiency attenuation coefficient is used as a label.

[0016] A2: Define three types of nodes in the model, including root nodes, internal nodes, and leaf nodes; the root node is used to receive the input parameter set and initiate the first split; the internal nodes are used to store split features, split thresholds, and the link relationships between left and right child nodes for sample splitting; the leaf nodes are used to store the subset of samples that cannot be split further and the mean of the labels of that subset, which is output as the final predicted value.

[0017] A3: Set the node termination split condition; the node termination split condition is that the number of sample subsets of the current node is less than a preset number, or the variance of the labels in the sample subsets of the current node is less than a preset maximum variance, or the tree depth reaches a preset upper limit.

[0018] A4: Randomly select a subset of samples of the same size from the input parameter set as the sample subset of the first tree, and at the same time randomly select some features from the input features as candidate splitting features of the first tree; the sample subset of the first tree is allowed to be repeatedly extracted with the same parameters during the extraction process.

[0019] A5: Pass the sample subset of the first tree to the root node, and use the mean of all sample labels in the root node as the initial prediction value of the node.

[0020] A6: For the sample subset of the current node, if the node termination splitting condition is met, the current node becomes a leaf node, and its predicted value is the mean of all sample labels in the sample subset of the current node.

[0021] A7: For the current node that does not meet the termination condition, traverse the candidate splitting features of the first tree, sort the values ​​of each candidate splitting feature in ascending order, and take the midpoint of adjacent values ​​as the candidate threshold of the candidate splitting feature.

[0022] A8: For each candidate splitting feature and the corresponding candidate threshold, split the sample subset of the current node into a left subset and a right subset, calculate the variance of the label in the parent node, the variance of the label in the left subset, and the variance of the label in the right subset respectively, and then obtain the variance reduction by subtracting the weighted sum of the variances of the left and right subset labels from the variance of the label in the parent node.

[0023] A9: From all combinations of candidate splitting features and candidate thresholds, select the combination with the largest reduction in variance as the splitting feature and splitting threshold of the current node, i.e., the optimal splitting combination;

[0024] A10: Based on the optimal splitting combination, split the sample subset of the current node into a left subset and a right subset, generate a left child node and a right child node respectively, and pass the left subset and right subset into the two child nodes respectively. Then repeat A6-A9 for the left child node and the right child node until all child nodes meet the node termination splitting condition and become leaf nodes. The training of the first tree model is completed.

[0025] A11: Repeat steps A4 to A10 to generate a preset number of tree models, and optimize the structure of each tree model using a post-pruning strategy to obtain optimized multi-tree models;

[0026] A12: Integrate all optimized multi-tree models into a multi-tree collaborative model;

[0027] A13: When it is necessary to predict the fin heat transfer efficiency attenuation coefficient of a new sample, the input features of the new sample are fed into each tree model respectively. Each tree outputs a predicted value, which is the average label value of the leaf node to which the sample finally arrives. Then, the average of the predicted values ​​of all trees is taken as the final output of the multi-tree collaborative model, resulting in the parameter-attenuation coefficient mapping relationship model.

[0028] Specifically, the calculation method for the fin heat transfer efficiency attenuation coefficient is as follows:

[0029] The average fin heat exchange efficiency over 24 consecutive hours after the first operation of the new equipment or after complete cleaning is used as the benchmark value.

[0030] The current heat exchange efficiency is calculated by combining the temperature difference between the fins and the ambient temperature and the thickness of the contaminants collected in real time with Newton's cooling formula.

[0031] The ratio of the current heat exchange efficiency to the baseline value is used as the fin heat exchange efficiency decay coefficient.

[0032] Specifically, the linkage control mechanism for activating the intelligent cleaning system and the smart sprinkler system includes:

[0033] The intelligent cleaning system is activated, and a dynamic priority division method based on the coupling of contaminants and temperature is adopted. Priority areas are divided by combining the collected contaminant thickness and fin temperature and cleaned in sequence. The intelligent determination system for the cleaning endpoint is constructed based on the changes in the fin temperature field transmitted back in real time by the sensing network to determine whether the cleaning is completed.

[0034] The intelligent spray system is activated simultaneously, and a graded spraying strategy is implemented according to the fin temperature-divided areas. At the same time, the spray water pressure is monitored and adjusted and compensated, and the adjustment and compensation results are fed back to the intelligent sensing network in real time.

[0035] Specifically, the dynamic prioritization method for cleaning based on contaminant-temperature dual-parameter coupling includes:

[0036] The surface of the air-cooled island fins is divided into standard grid units. Each grid unit is equipped with an independent temperature sensor and laser contaminant detector to collect the fin temperature and ambient temperature of the grid unit in real time and calculate the contaminant thickness.

[0037] The contaminant thickness is divided into N grade ranges, and the fin temperature difference is divided into M grade ranges; the fin temperature difference is the temperature difference between the fin temperature and the ambient temperature.

[0038] Construct an N×M two-dimensional priority matrix; each element in the priority matrix corresponds to a cleaning priority for a combination of working conditions; the cleaning priority is divided into a first-level queue, a second-level queue, and a third-level queue;

[0039] All grid cells are scanned in real time. Based on the currently collected contaminant thickness and fin temperature difference, the grid cells are classified into a first-level queue, a second-level queue, or a third-level queue according to a dynamic scoring rule, and the spatial coordinates of the grid cells in each queue are recorded. The dynamic scoring rule is a predefined correspondence between the level combination of contaminant thickness and fin temperature difference and the cleaning priority.

[0040] An improved greedy algorithm is used to generate the shortest cleaning path covering all first-level queue grids.

[0041] Specifically, the process of generating the shortest cleaning path covering all first-level queue grids using an improved greedy algorithm includes:

[0042] Extract all grid cells classified into the first-level queue from the intelligent sensing network, record the physical coordinates of each cell, and read the contaminant thickness and fin temperature difference of each grid cell.

[0043] The contaminant thickness and fin temperature difference of all grid cells are normalized to the same scale, and their cleaning urgency is calculated according to a preset weight; the cleaning urgency is a comprehensive score of contaminant thickness and fin temperature difference.

[0044] Obtain the coordinates of the currently spraying wet area from the intelligent sprinkler system, mark it as a restricted area, and generate a restricted area list;

[0045] Using the current position of the cleaning robot as the starting point of the path, add all primary queue grids to the list to be cleaned, and at the same time, remove all grid cells that overlap with the restricted area;

[0046] For each remaining grid cell in the cleaning list, calculate the weighted sum of three indicators: Euclidean distance, cleaning urgency, and movement energy consumption, to obtain the comprehensive cost;

[0047] Based on the three indicators, the grid cell with the lowest overall cost is selected as the next target cleaning point. The robot is then controlled to move to the target cleaning point to perform the cleaning operation. After the cleaning is completed, the grid cell is removed from the list of cells to be cleaned.

[0048] If the spray system opens a new wet zone during the cleaning process, the no-entry list should be updated immediately.

[0049] Check the generated paths. If three detours are found, delete the intermediate target points. If the list to be cleaned is empty, it means that all first-level queue grid cells have been cleaned and the path planning is complete.

[0050] Specifically, the determination logic of the intelligent determination system for the cleaning endpoint based on the fin temperature field feedback is as follows:

[0051] The system monitors the rate of change and uniformity of the temperature field of the fins in real time during the cleaning process. When the rate of change of the temperature field is lower than the preset rate threshold and the uniformity of the temperature field distribution reaches the preset uniformity threshold, it determines that the cleaning of the area is complete and feeds the determination result back to the intelligent cleaning system to control it to switch to the cleaning operation of the next priority area.

[0052] Specifically, the process of constructing the control effect evaluation matrix based on the back pressure-attenuation coefficient coupling degree includes:

[0053] Using the back pressure change rate and the fin heat exchange efficiency attenuation coefficient change rate as two-dimensional coordinate axes, K evaluation regions are divided. Each evaluation region corresponds to a control effect level. By calculating the coupling degree index between back pressure and attenuation coefficient, the current control effect level is determined.

[0054] A smart management and control system for an inter-cooled air-cooled island includes: an intelligent sensing module, an evaluation and decision-making module, a linkage control module, an effect evaluation module, and a mode switching module;

[0055] The intelligent sensing module is used to collect key operating parameters of the air-cooled island in real time;

[0056] The evaluation and decision-making module is used to construct a multi-parameter fusion evaluation model and generate control strategies.

[0057] The linkage control module is used to perform coordinated control of intelligent cleaning and spraying;

[0058] The effect evaluation module is used to quantify the control effect and dynamically optimize the strategy;

[0059] The mode switching module is used to achieve seamless switching between intelligent control and routine monitoring.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] 1. This invention proposes an intelligent management and control system for intercooled and air-cooled islands, and optimizes and improves its architecture, operation steps and processes. The system has the advantages of simple process, low investment and operating costs and low production and working costs.

[0062] 2. This invention proposes a coordinated control method for air-cooled islands with indirect cooling. This invention monitors key operating parameters of the air-cooled island in real time through an intelligent sensing network, and dynamically calculates the heat exchange efficiency attenuation coefficient and back pressure warning value based on a multi-parameter fusion evaluation model. This achieves intelligent coordinated control of the cleaning and spraying systems. When performance degradation is detected, the system automatically triggers a linkage control mechanism. By dynamically optimizing the control strategy through a back pressure-attenuation coefficient coupled evaluation matrix, the accuracy and response speed of heat exchange efficiency maintenance are significantly improved.

[0063] 3. This invention proposes a coordinated control method for indirect-cooled air-cooled islands, innovatively constructing a closed-loop control system from anomaly detection and intelligent control to effect evaluation. It can not only quickly restore equipment performance to normal state, but also achieve continuous operation and maintenance through the automatic switching mechanism of conventional monitoring modes. Compared with traditional methods, this invention effectively reduces energy consumption, extends equipment life, and reduces manual intervention, providing an efficient and reliable fully automatic intelligent operation and maintenance solution for air-cooled island systems. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of a coordinated control method for an intercooled air-cooled island according to the present invention;

[0065] Figure 2 This is a flowchart illustrating the principle of a coordinated control method for an intercooled air-cooled island according to the present invention.

[0066] Figure 3 This is a diagram illustrating the architecture of an intelligent control system for an intercooled air-cooled island according to the present invention. Detailed Implementation

[0067] Example 1

[0068] Please see Figures 1-2 The present invention provides an embodiment of a coordinated control method for an intercooled air-cooled island, comprising the following steps:

[0069] S1: Build an intelligent sensing network to collect key operating parameters in real time, construct an evaluation model based on the key operating parameters, and calculate the fin heat exchange efficiency decay coefficient and back pressure warning value; the key operating parameters include fin temperature, back pressure, ambient temperature and contaminant thickness.

[0070] Furthermore, the construction of the intelligent sensing network includes: arranging temperature sensors, pressure sensors, environmental sensors, and thickness sensors in the fin area, back pressure monitoring point, environmental monitoring area, and pollutant accumulation area of ​​the indirect cooling island, respectively. Each sensor is connected to the central controller through a wireless communication module to form a distributed sensing network. The real-time collected data is preprocessed by edge computing nodes and then transmitted to the central controller to provide data support for the evaluation model.

[0071] Furthermore, the method for determining the back pressure warning value is as follows: based on the maximum allowable back pressure value in the equipment design parameters, combined with the influence coefficient of real-time ambient temperature on back pressure, the back pressure warning value is dynamically adjusted through the back pressure warning value calculation model. When the real-time back pressure reaches the warning value, the linkage control mechanism in S2 is triggered.

[0072] Furthermore, the influence coefficient of real-time ambient temperature on back pressure was gradually determined through a combination of data modeling and dynamic calibration, including:

[0073] First, we need to start from the operating principle of the equipment and clarify how ambient temperature affects back pressure. That is, changes in ambient temperature will change the density, viscosity and heat exchange efficiency of the air around the fins, thereby affecting the pollutant deposition rate and ultimately causing fluctuations in the resistance of airflow through the fins. Based on this mechanism, a basic dataset is constructed through experimental or historical operational data collection, covering back pressure values ​​corresponding to different ambient temperatures, back pressure change data under the same operating conditions, and the correspondence between the rate of temperature change and the rate of back pressure change. Next, the influence coefficient model is constructed, with the core objective of quantifying the relative rate of change in back pressure caused by a unit change in ambient temperature. First, an influence coefficient index is defined to represent the relative change in back pressure when the ambient temperature changes by 1°C. Then, regression analysis is used to fit the relationship between temperature and back pressure in the dataset, obtaining basic influence coefficient values ​​for different temperature ranges. This reflects the differentiated impact of different temperature ranges on back pressure; for example, the coefficients for high and low temperature ranges may show positive and negative differences. Finally, the basic influence coefficients are dynamically calibrated and updated in real time. By introducing correction factors and combining the deviation between the real-time collected ambient temperature and measured back pressure values ​​and the model's predicted values, the basic coefficients are adjusted to obtain real-time influence coefficients. This ensures that the real-time influence coefficients reflect the latest equipment operating status and avoids error accumulation caused by long-term use of fixed coefficients.

[0074] It should be explained that the construction of the intelligent sensing network enables comprehensive perception and real-time monitoring of the air-cooled island's operating status, solving the problems of incomplete data collection and poor timeliness in traditional methods. Through the deployment of high-precision sensors and the application of data preprocessing technology, the system can accurately capture subtle changes in the operation of the air-cooled island. At the same time, the real-time transmission and storage functions ensure the integrity and traceability of the data, providing support for system optimization and fault diagnosis.

[0075] S2: When the fin heat exchange efficiency attenuation coefficient is lower than the preset value or the back pressure reaches the back pressure warning value, the linkage control mechanism of the intelligent cleaning system and the intelligent spraying system is activated simultaneously, and the initial cleaning and spraying control commands are output.

[0076] Furthermore, the preset value mentioned in S2 is the critical threshold of the fin heat exchange efficiency decay coefficient. This threshold is determined by analyzing the energy efficiency curves of the equipment under different operating conditions. When the fin heat exchange efficiency decay coefficient is lower than the preset value, it indicates that the fin heat exchange efficiency has decreased to the extent that it affects the normal operation of the equipment, and the linkage control mechanism in S2 needs to be activated.

[0077] S3: Based on the current back pressure and fin heat exchange efficiency attenuation coefficient, construct a control effect evaluation matrix based on the coupling degree of back pressure-attenuation coefficient and evaluate the control effect level of cleaning and spraying. Adjust the spraying intensity or cleaning frequency according to the control effect level and output the adjusted control parameters.

[0078] Furthermore, the specific rules for adjusting the spray intensity or cleaning frequency based on the control effect level are as follows: when the control effect level is excellent, maintain the current parameters; when the control effect level is good, maintain the current parameters or make minor adjustments; when the control effect level is medium, increase the spray intensity or increase the cleaning frequency; when the control effect level is poor, increase the spray intensity or significantly increase the cleaning frequency, and the adjusted parameters are issued to the corresponding system as new control commands.

[0079] S4: Until the ambient temperature and back pressure reach the preset normal conditions and remain so for a preset duration, output a control endpoint signal and switch to the normal monitoring mode based on the control endpoint signal; the normal monitoring mode maintains real-time acquisition and monitoring of the key operating parameters, and once an abnormal change in the key operating parameters is detected, immediately switch back to the intelligent control mode and restart the intelligent cleaning system or the intelligent spraying system.

[0080] Furthermore, the preset normal conditions are: the ambient temperature is within the equipment's designed operating ambient temperature range, the back pressure is lower than the back pressure warning value, and the fin heat exchange efficiency attenuation coefficient is greater than or equal to the preset value. The preset duration is set according to the equipment's operational stability requirements, typically 30-60 minutes. When the above conditions are met, a control endpoint signal is output.

[0081] Furthermore, the conventional monitoring mode operates as follows: the intelligent sensing network maintains real-time acquisition of key operating parameters, but the acquisition frequency is lower than that of the intelligent control mode. The central controller performs routine analysis on the acquired data. When it detects that the key operating parameters exceed the normal fluctuation range, it is determined to be an abnormal change and the mode switching mechanism is immediately triggered.

[0082] Furthermore, the criteria for determining abnormal changes in the key operating parameters are: the sudden change in fin temperature exceeds the preset temperature difference threshold, the rate of back pressure increase exceeds the preset rate threshold, the ambient temperature exceeds the normal range, or the rate of increase in pollutant thickness is abnormal. When any of the above conditions are met, the system automatically switches back to the intelligent control mode and re-executes the operations from S2 to S4.

[0083] In summary, this application presents an innovative solution to address the problems of lagging efficiency decay monitoring, crude cleaning and spraying control, and insufficient energy efficiency optimization in the traditional operation and management of air-cooled islands. By building a high-precision intelligent sensing network and combining a multi-parameter fusion evaluation model with a dynamic linkage control mechanism, it achieves real-time perception, accurate evaluation, and intelligent control of the air-cooled island's operating status. Its core advantages are: 1) The system can collect key operating parameters such as fin temperature, back pressure, ambient temperature, and contaminant thickness in real time, ensuring the comprehensiveness and timeliness of data collection and providing a reliable basis for subsequent evaluation and control; 2) Based on historical data and real-time parameters, a calculation model for the fin heat exchange efficiency decay coefficient and back pressure early warning value is constructed to identify potential efficiency decay and safety risks in advance, providing a scientific basis for preventive maintenance; 3) Through the linkage control of the intelligent cleaning system and the intelligent spraying system, combined with the cleaning priority division and graded spraying strategy based on the coupling of contaminant and temperature dual parameters, the precise execution of cleaning and spraying is achieved, significantly improving control efficiency and energy efficiency; 4) Through the control effect evaluation matrix based on the coupling degree of back pressure and decay coefficient, the control strategy is dynamically optimized to ensure that the system can maintain the optimal operating state under different working conditions, while supporting seamless switching between intelligent control and conventional monitoring, improving system stability and reliability.

[0084] For example, the evaluation model includes a parameter-attenuation coefficient mapping model and a back pressure warning value calculation model, and the construction process of the evaluation model includes:

[0085] Acquire historical key operating parameters and perform noise filtering and timing alignment processing;

[0086] Furthermore, the specific process of noise filtering includes:

[0087] (1) Plot the trend of the original historical key operating parameters in time series and observe the data fluctuation characteristics: if the data shows irregular high-frequency small-amplitude oscillations, it is judged as random noise; if there is an isolated large jump, such as the fin temperature suddenly soaring and then falling back at a certain moment, it is judged as pulse noise; if the data as a whole shows a slow, non-physical regular upward or downward trend, such as the ambient temperature continuously rising by 0.5℃ per hour under stable operating conditions, exceeding the range of natural changes, it is judged as drift noise.

[0088] (2) Using 3 The criteria for identifying outliers corresponding to impulse noise include: first calculating the mean of historical data for any parameter. and standard deviation Set the outlier threshold range as follows: Data points exceeding the outlier threshold are marked as outliers. For marked outliers, a second verification is performed based on the physical operating characteristics of the equipment. For example, if the back pressure cannot jump from 0.2MPa to 0.8MPa instantaneously, it can be directly identified as an outlier. After confirmation, it is temporarily stored for further processing to obtain the operating data after removing outliers.

[0089] (3) For the running data after removing outliers, a sliding window mean filtering process is adopted, including: setting the window size; taking all data points in the window in sequence according to the time axis, calculating the arithmetic mean, and replacing the data point at the center of the window with the mean; the window slides point by point from the starting position of the data until it covers all data, and eliminates random noise by smoothing high-frequency fluctuations; the window size is determined according to the parameter characteristics, such as the window size is set to 5 sampling points when the fin temperature changes rapidly; the window size is set to 10 sampling points when the back pressure changes slowly.

[0090] (4) For the operational data after the above processing, establish a linear regression model according to the time series, and determine whether there is drift noise based on the slope of the linear regression model. The linear regression model is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0091] (5) If the absolute value of the slope exceeds the preset physical threshold, it is determined that there is drift noise. The corrected drift-free data is obtained by subtracting the linear model calculation value from the original running data.

[0092] (6) The outliers in (2) are filled by linear interpolation of adjacent valid data points to obtain the historical key operating parameters after noise filtering. The linear interpolation method is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0093] Furthermore, the specific process of performing timing alignment includes:

[0094] (1) Extract the original timestamps from the historical key operating parameters of each parameter and perform a unified conversion to ensure consistent time representation. For data without a clear timestamp, calculate the timestamp based on the start time and sampling interval.

[0095] (2) Statistically count the sampling frequency of all historical key operating parameters, select the highest sampling frequency among the historical key operating parameters as the reference frequency, and take the parameter time series corresponding to the reference frequency as the reference time axis. If there are at least two highest frequency parameters, take the parameter time axis with the largest data volume as the reference to ensure that the time range of all parameters is covered.

[0096] (3) For parameters whose sampling frequency is lower than the reference frequency, based on the reference time axis and reference frequency in (2), the sampling frequency is increased to the reference frequency by linear interpolation, including: on the reference time axis, find all reference time points between two adjacent original data points of the historical key operating parameter, calculate the interpolation value, so that the historical key operating parameter has a corresponding value at each time point on the reference time axis;

[0097] (4) For parameters whose sampling frequency is equal to the reference frequency, directly match the reference time axis;

[0098] (5) For the time offset caused by sensor response delay, the offset is calculated by cross-correlation analysis, including: taking the time series of two parameters, calculating the correlation coefficient under different time offsets, and taking the offset when the correlation coefficient is the largest as the correction value, and shifting the timestamp of the lag parameter forward by the corresponding time to achieve time synchronization. The cross-correlation analysis method is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0099] (6) After alignment, check whether there are missing values ​​on the reference time axis. If there are no parameter values ​​at any time point, use the mean of 10 valid data points before and after the parameter to fill in the missing values. Finally, by drawing a multi-parameter time series overlay plot, visually verify whether the values ​​of each parameter at the same time point match the physical logic, and ensure that the data logic is consistent after alignment.

[0100] A multi-tree collaborative model was adopted, with fin temperature, ambient temperature, and contaminant thickness as input features, and the measured fin heat transfer efficiency attenuation coefficient as a label for training, to obtain a parameter-attenuation coefficient mapping relationship model.

[0101] Based on historical back pressure data, the mean and standard deviation of back pressure under different ambient temperatures are calculated using the sliding window method. Combined with the equipment safe operation threshold, a back pressure early warning value calculation model is constructed.

[0102] Furthermore, the equipment safe operation threshold refers to the maximum allowable back pressure limit set during the design and actual operation of the equipment to ensure its stable, safe, and efficient operation. It can also be called the maximum allowable back pressure value or the safe back pressure upper limit. This is a core parameter determined by the equipment manufacturer based on factors such as hardware performance, operating principles, and lifespan degradation. It serves as a critical standard for judging whether the equipment is in a safe operating state. For example, taking the finned heat exchanger of an air conditioner outdoor unit as an example, the equipment safe operation threshold is set at 500 Pa. This means that when the back pressure exceeds 500 Pa, the fan load is too high, which may burn out the motor or lead to a decrease in heat exchange efficiency. Based on historical data, the average back pressure at different ambient temperatures is 200~400 Pa, with a standard deviation of 30~50 Pa. Therefore, the warning value calculation model uses 500 Pa as the safety baseline, and combines the influence coefficient, average, and standard deviation of ambient temperature on back pressure to dynamically set the warning value to 350~450 Pa, ensuring that the warning is triggered before the back pressure approaches the dangerous value. Here, Pa is a unit of pressure, representing Pascals.

[0103] The multi-tree collaborative model is trained using fin temperature, ambient temperature, and contaminant thickness as input features, and the measured fin heat transfer efficiency attenuation coefficient as a label, to obtain a parameter-attenuation coefficient mapping relationship model, including:

[0104] A1: Obtain fin temperature, ambient temperature, contaminant thickness, and measured fin heat transfer efficiency attenuation coefficient, and combine them to form an input parameter set. Among them, fin temperature, ambient temperature, and contaminant thickness are used as input features, and the measured fin heat transfer efficiency attenuation coefficient is used as a label.

[0105] A2: Define three types of nodes in the model, including root nodes, internal nodes, and leaf nodes; the root node is used to receive the input parameter set and initiate the first split; the internal nodes are used to store split features, split thresholds, and the link relationships between left and right child nodes for sample splitting; the leaf nodes are used to store the subset of samples that cannot be split further and the mean of the labels of that subset, which is output as the final predicted value.

[0106] A3: Set the node termination split condition; the node termination split condition is that the number of sample subsets of the current node is less than a preset number, or the variance of the labels in the sample subsets of the current node is less than a preset maximum variance, or the tree depth reaches a preset upper limit.

[0107] A4: Randomly select a subset of samples of the same size from the input parameter set as the sample subset of the first tree, and at the same time randomly select some features from the input features as candidate splitting features of the first tree; the sample subset of the first tree is allowed to be repeatedly extracted with the same parameters during the extraction process.

[0108] Furthermore, a subset of samples of the same size is randomly selected from the input parameter set as the sample subset of the first tree. Here, "same size" means that the number of samples in the sample subset is the same as the number of samples in the original input parameter set. Repeated selection is allowed, that is, some samples may be included in the subset multiple times, as long as the number is the same.

[0109] A5: Pass the sample subset of the first tree to the root node, and use the mean of all sample labels in the root node as the initial prediction value of the node.

[0110] A6: For the sample subset of the current node, if the node termination splitting condition is met, the current node becomes a leaf node, and its predicted value is the mean of all sample labels in the sample subset of the current node.

[0111] A7: For the current node that does not meet the termination condition, traverse the candidate splitting features of the first tree, sort the values ​​of each candidate splitting feature in ascending order, and take the midpoint of adjacent values ​​as the candidate threshold of the candidate splitting feature.

[0112] A8: For each candidate splitting feature and the corresponding candidate threshold, split the sample subset of the current node into a left subset and a right subset, calculate the variance of the label in the parent node, the variance of the label in the left subset, and the variance of the label in the right subset respectively, and then obtain the variance reduction by subtracting the weighted sum of the variances of the left and right subset labels from the variance of the label in the parent node.

[0113] A9: From all combinations of candidate splitting features and candidate thresholds, select the combination with the largest reduction in variance as the splitting feature and splitting threshold of the current node, i.e., the optimal splitting combination;

[0114] A10: Based on the optimal splitting combination, split the sample subset of the current node into a left subset and a right subset, generate a left child node and a right child node respectively, and pass the left subset and right subset into the two child nodes respectively. Then repeat A6-A9 for the left child node and the right child node until all child nodes meet the node termination splitting condition and become leaf nodes. The training of the first tree model is completed.

[0115] A11: Repeat steps A4 to A10 to generate a preset number of tree models, and optimize the structure of each tree model using a post-pruning strategy to obtain optimized multi-tree models;

[0116] Furthermore, a post-pruning strategy is applied to optimize the structure of each tree model, including:

[0117] Tracing back from the leaf node to the root node, for each internal node, two types of errors are calculated: one is the post-pruning error, which is the error when the node is converted into a leaf node and the mean of the sample labels of all its child nodes is used as the prediction value; the other is the pre-pruning error, which is the total error when the child nodes of the node are retained. If the post-pruning error is less than or equal to the pre-pruning error, the child nodes of the node are removed and the node is converted into a leaf node. This process is repeated until all pruning nodes have been processed, resulting in the optimized multi-tree model.

[0118] A12: Integrate all optimized multi-tree models into a multi-tree collaborative model;

[0119] A13: When it is necessary to predict the fin heat transfer efficiency attenuation coefficient of a new sample, the input features of the new sample are fed into each tree model respectively. Each tree outputs a predicted value, which is the average label value of the leaf node to which the sample finally arrives. Then, the average of the predicted values ​​of all trees is taken as the final output of the multi-tree collaborative model, resulting in the parameter-attenuation coefficient mapping relationship model.

[0120] As can be seen, the parameter-attenuation coefficient mapping model in this invention is a branch structure model that can autonomously learn the correlation between fin temperature, ambient temperature, pollutant thickness, and heat transfer efficiency attenuation coefficient. Its core consists of a unique branching network composed of three types of nodes. The root node is the starting point of the model and first collects all training samples. These samples contain the three feature parameters mentioned above and their corresponding measured attenuation coefficients. In the initial state, the mean of the attenuation coefficients of all samples is calculated as the initial prediction value. After the split is initiated, it becomes the core node that coordinates the branching direction. The branch nodes are the decision-making center of the model. Each node carries key attributes: first, the selected splitting feature; second, the corresponding critical value; and third, the link path between the left and right child nodes. Samples whose feature values ​​do not exceed the critical value will flow to the left child node, and those that exceed it will flow to the right child node, thus achieving accurate sample splitting. Finally, the terminal node is the output end of the model. When the sample subset meets the condition for stopping the split, it converges here. Its core attribute is all samples in the subset and their corresponding average attenuation coefficients. Finally, this average value is directly output as the prediction result.

[0121] Furthermore, in this invention, the root node first completes the initial split using the optimal splitting features and critical values, generating two branch nodes that respectively handle samples from different feature ranges. Each branch node repeats this process, splitting again based on the feature distribution of its own subset, until the number of samples in the subset is insufficient to support further subdivision, the fluctuation range of the attenuation coefficient reaches a preset purity anchor point (i.e., minimal fluctuation), or the branch depth reaches its upper limit. At this point, the branch node transforms into a terminal node, and the splitting stops. This progressive splitting logic allows the model to gradually capture the changing patterns of the attenuation coefficient under different feature combinations, forming a hierarchical interconnected network.

[0122] Furthermore, for the prediction of new samples, the multi-tree collaborative model initiates a path tracing mechanism: starting from the root node, based on the node's splitting characteristics and critical values, it determines the range of the new sample's feature values ​​and enters the corresponding branch node; after reaching the branch node, it continues to branch based on its features and critical values, progressing layer by layer until it reaches the terminal node, where the average attenuation coefficient is the final prediction result. This model is particularly well-suited to three features: for fin temperature, it can capture the feature sensitivity of a sharp drop in attenuation coefficient after the temperature exceeds any critical value; for ambient temperature, it can learn the difference in the response of the attenuation coefficient to fin temperature under different ambient temperatures; for pollutant thickness, it can quantify the gradient of attenuation coefficient decrease as the thickness increases by a certain amount, and finally, through a unique branching network, it constructs an accurate feature-attenuation coefficient mapping relationship.

[0123] The calculation method for the fin heat transfer efficiency attenuation coefficient is as follows:

[0124] The average fin heat exchange efficiency over 24 consecutive hours after the first operation of the new equipment or after complete cleaning is used as the benchmark value.

[0125] The current heat exchange efficiency is calculated by combining the temperature difference between the fins and the ambient temperature and the thickness of the contaminants collected in real time with Newton's cooling formula.

[0126] The ratio of the current heat exchange efficiency to the baseline value is used as the fin heat exchange efficiency decay coefficient.

[0127] It should be explained that the construction of the assessment model enables accurate assessment and early warning of the air-cooled island's operating status. The parameter-attenuation coefficient mapping model, through machine learning algorithms, reveals the complex relationship between fin temperature, ambient temperature, contaminant thickness, and fin heat exchange efficiency, providing a scientific basis for the quantitative assessment of system efficiency attenuation. The back pressure early warning value calculation model, based on historical data and statistical methods, sets back pressure safety thresholds at different ambient temperatures, providing strong protection for the safe operation of the system. The real-time calculation of the fin heat exchange efficiency attenuation coefficient enables the system to dynamically track changes in fin heat exchange efficiency, providing timely feedback for cleaning and spray control. The synergistic effect of these models improves the accuracy of system assessment and the timeliness of early warning.

[0128] The linkage control mechanism for activating the intelligent cleaning system and the smart sprinkler system includes:

[0129] The intelligent cleaning system is activated, and a dynamic priority division method based on the coupling of contaminants and temperature is adopted. Priority areas are divided by combining the collected contaminant thickness and fin temperature and cleaned in sequence. The intelligent determination system for the cleaning endpoint is constructed based on the changes in the fin temperature field transmitted back in real time by the sensing network to determine whether the cleaning is completed.

[0130] The intelligent spray system is activated simultaneously, and a graded spraying strategy is implemented according to the fin temperature-divided areas. At the same time, the spray water pressure is monitored and adjusted and compensated, and the adjustment and compensation results are fed back to the intelligent sensing network in real time.

[0131] Furthermore, the adjustment and compensation results refer to the specific operational results and data of the intelligent sprinkler system after detecting that the sprinkler water pressure deviates from the target value, and then using adjustment and control methods to restore the water pressure to the target range. These results are fed back to the intelligent sensing network in real time, providing a basis for subsequent control effect evaluation and parameter adjustment, including the operating parameters of water pressure adjustment, the actual value of water pressure after correction, the deviation correction status, and the status indicators of the adjustment process.

[0132] Furthermore, the step of implementing a graded spraying strategy based on the fin temperature-divided zones includes:

[0133] The temperature difference between the fins and the ambient temperature is divided into Q spray levels. The spray range is determined based on the area of ​​the fin region. The graded spraying is achieved by adjusting the number of spray heads, the water pressure, and the spraying time. Real-time data during the spraying process is fed back to the intelligent sensing network for evaluation of the control effect.

[0134] The dynamic prioritization method for cleaning based on the coupling of contaminants and temperature includes:

[0135] The surface of the air-cooled island fins is divided into standard grid units. Each grid unit is equipped with an independent temperature sensor and laser contaminant detector to collect the fin temperature and ambient temperature of the grid unit in real time and calculate the contaminant thickness.

[0136] The contaminant thickness is divided into N grade ranges, and the fin temperature difference is divided into M grade ranges; the fin temperature difference is the temperature difference between the fin temperature and the ambient temperature.

[0137] Construct an N×M two-dimensional priority matrix; each element in the priority matrix corresponds to a cleaning priority for a combination of working conditions; the cleaning priority is divided into a first-level queue, a second-level queue, and a third-level queue;

[0138] All grid cells are scanned in real time. Based on the currently collected contaminant thickness and fin temperature difference, the grid cells are classified into a first-level queue, a second-level queue, or a third-level queue according to a dynamic scoring rule, and the spatial coordinates of the grid cells in each queue are recorded. The dynamic scoring rule is a predefined correspondence between the level combination of contaminant thickness and fin temperature difference and the cleaning priority.

[0139] An improved greedy algorithm is used to generate the shortest cleaning path covering all first-level queue grids.

[0140] The process of generating the shortest cleaning path covering all primary queue grids using an improved greedy algorithm includes:

[0141] Extract all grid cells classified into the first-level queue from the intelligent sensing network, record the physical coordinates of each cell, and read the contaminant thickness and fin temperature difference of each grid cell.

[0142] The contaminant thickness and fin temperature difference of all grid cells are normalized to the same scale, and their cleaning urgency is calculated according to a preset weight; the cleaning urgency is a comprehensive score of contaminant thickness and fin temperature difference.

[0143] Obtain the coordinates of the currently spraying wet area from the intelligent sprinkler system, mark it as a restricted area, and generate a restricted area list;

[0144] Using the current position of the cleaning robot as the starting point of the path, add all primary queue grids to the list to be cleaned, and at the same time, remove all grid cells that overlap with the restricted area;

[0145] For each remaining grid cell in the cleaning list, the weighted sum of three indicators—Euclidean distance, cleaning urgency, and movement energy consumption—is calculated to obtain the comprehensive cost. The Euclidean distance calculation formula is existing technology in this field and is not an inventive solution of this application, so it will not be elaborated here.

[0146] Mobility energy consumption refers to the energy cost consumed by the intelligent cleaning equipment when it moves from its current grid cell to any target grid cell in the cleaning list. Specifically, after the intelligent cleaning equipment completes the cleaning of one grid cell, it needs to plan a movement path based on the relative positions of the two grid cells before moving to the next one. Mobility energy consumption is calculated based on the actual length of this path, combined with the average power and movement time of the equipment during the movement. For example, the greater the distance between two grid cells, the longer the time required for the equipment to move, and the higher the energy consumption. If there are obstacles in the path that need to be bypassed, the actual movement distance increases, and the energy consumption will also increase accordingly.

[0147] Based on the three indicators, the grid cell with the lowest overall cost is selected as the next target cleaning point. The robot is then controlled to move to the target cleaning point to perform the cleaning operation. After the cleaning is completed, the grid cell is removed from the list of cells to be cleaned.

[0148] If the spray system opens a new wet zone during the cleaning process, the no-entry list should be updated immediately.

[0149] Check the generated paths. If three detours are found, delete the intermediate target points. If the list to be cleaned is empty, it means that all first-level queue grid cells have been cleaned and the path planning is complete.

[0150] The determination logic of the intelligent determination system for the cleaning endpoint based on the fin temperature field feedback is as follows:

[0151] The system monitors the rate of change and uniformity of the temperature field of the fins in real time during the cleaning process. When the rate of change of the temperature field is lower than the preset rate threshold and the uniformity of the temperature field distribution reaches the preset uniformity threshold, it determines that the cleaning of the area is complete and feeds the determination result back to the intelligent cleaning system to control it to switch to the cleaning operation of the next priority area.

[0152] The process of constructing the control effect evaluation matrix based on back pressure-attenuation coefficient coupling degree includes:

[0153] Using the back pressure change rate and the fin heat exchange efficiency attenuation coefficient change rate as two-dimensional coordinate axes, K evaluation regions are divided. Each evaluation region corresponds to a control effect level. By calculating the coupling degree index between back pressure and attenuation coefficient, the current control effect level is determined.

[0154] Among them, the coupling degree index is a quantitative value that comprehensively measures the synergistic effect of changes in back pressure and changes in attenuation coefficient.

[0155] Furthermore, the process of constructing the two-dimensional coordinate axes includes: establishing a two-dimensional evaluation coordinate system with the rate of change of back pressure on the horizontal axis and the rate of change of the fin heat transfer efficiency attenuation coefficient on the vertical axis.

[0156] Horizontal axis direction: the left side is the area of ​​decreasing back pressure, and the right side is the area of ​​increasing back pressure;

[0157] Vertical axis: The upper part represents the region where the attenuation coefficient increases, and the lower part represents the region where the attenuation coefficient decreases;

[0158] The origin of the coordinate system corresponds to an initial state where the rate of change of back pressure is 0 and the rate of change of attenuation coefficient is 0, which means that the control measures have no effect.

[0159] Furthermore, the division of assessment areas and corresponding control effectiveness levels includes:

[0160] Based on the actual operational needs of the equipment and historical data patterns, K evaluation regions are divided in a two-dimensional coordinate system. Each region corresponds to a control effectiveness level, such as four levels: Excellent, Good, Average, and Poor, i.e., K=4. The division logic must reflect the principle that the more significant the decrease in back pressure and the more significant the increase in the attenuation coefficient, the better the control effect.

[0161] Superior region: Located in the upper left of the coordinate system, i.e., the rate of change of back pressure is negative and the rate of change of attenuation coefficient is positive;

[0162] Good grade region: located in the upper left of the coordinate system but close to the origin, that is, the rate of change of back pressure is a small negative value and the rate of change of attenuation coefficient is a small positive value;

[0163] Intermediate region: located in the lower right of the coordinate system or close to the origin, that is, the rate of change of back pressure is close to 0 or slightly positive and the rate of change of attenuation coefficient is close to 0;

[0164] Differential region: located in the lower right of the coordinate system, that is, the rate of change of back pressure is positive and the rate of change of attenuation coefficient is negative.

[0165] Furthermore, the process of calculating the coupling index between back pressure and attenuation coefficient includes:

[0166] (1) Determine the standardized contribution values ​​of the back pressure change rate and the attenuation coefficient change rate: take the absolute value of the back pressure change rate, since a negative value indicates improvement and the larger the absolute value, the greater the contribution; take the original value of the attenuation coefficient change rate directly, since a positive value indicates improvement and the larger the value, the greater the contribution.

[0167] (2) Assign different weights to the rate of change of back pressure and the rate of change of attenuation coefficient, such as a weight of 0.6 for back pressure and a weight of 0.4 for attenuation coefficient, with a total weight of 1. In this invention, the weights need to be determined based on the degree of influence of the two on the energy efficiency of the equipment in historical data.

[0168] (3) Calculate the final coupling index; the final coupling index is equal to the product of the standardized contribution value of the back pressure change rate and the back pressure weight, plus the sum of the product of the standardized contribution value of the attenuation coefficient change rate and the attenuation coefficient weight.

[0169] It should be noted that in this application, the design of the intelligent linkage control mechanism realizes the precise execution and efficient coordination of cleaning and spraying operations; the dynamic classification method of cleaning priority scientifically divides the priority of cleaning areas through two-parameter coupling analysis, ensuring priority cleaning of high-pollution and high-temperature difference areas, and improving the targeting and effectiveness of cleaning operations; the improved greedy algorithm path planning optimizes the movement path of the cleaning robot, reduces ineffective movement and energy consumption, and improves the efficiency of cleaning operations; the implementation of the graded spraying strategy flexibly adjusts the spraying parameters according to different fin temperature levels, ensuring the accuracy and adaptability of spraying operations; the introduction of the intelligent determination system for cleaning endpoint scientifically determines whether cleaning is complete by monitoring changes in the fin temperature field in real time, avoiding the problems of over-cleaning or under-cleaning, and improving the quality and reliability of cleaning operations. The synergistic effect of these mechanisms enhances the intelligence level of air-cooled island cleaning and spraying operations.

[0170] Example 2

[0171] Please see Figure 3 Another embodiment of the present invention provides: a smart control system for intercooled air-cooled islands, comprising:

[0172] Intelligent sensing module, evaluation and decision-making module, linkage control module, effect evaluation module, mode switching module;

[0173] The intelligent sensing module is used to collect key operating parameters of the air-cooled island in real time, providing data support for system decision-making;

[0174] The evaluation and decision-making module is used to build a multi-parameter fusion evaluation model and generate control strategies.

[0175] The linkage control module is used to perform coordinated control of intelligent cleaning and spraying;

[0176] The effect evaluation module is used to quantify the control effect and dynamically optimize the strategy;

[0177] The mode switching module is used to achieve seamless switching between intelligent control and routine monitoring.

[0178] The intelligent sensing module includes: a temperature monitoring unit, a back pressure monitoring unit, an environmental monitoring unit, and a dirt detection unit;

[0179] The temperature monitoring unit monitors the temperature distribution on the fin surface in real time by deploying an infrared sensor array;

[0180] The back pressure monitoring unit collects back pressure data at the inlet and outlet of the condenser via a pressure transmitter.

[0181] The environmental monitoring unit integrates temperature and humidity sensors and an anemometer to acquire ambient temperature and humidity data.

[0182] The dirt detection unit is used to measure the thickness of contaminant buildup on the fin surface using a laser thickness gauge.

[0183] The evaluation and decision-making module includes: a preprocessing unit, an efficiency calculation unit, an early warning analysis unit, and a decision generation unit;

[0184] The preprocessing unit is used to filter and normalize the raw data;

[0185] The efficiency calculation unit is used to calculate the fin heat transfer efficiency attenuation coefficient based on the heat transfer model.

[0186] The early warning analysis unit is used to dynamically generate back pressure early warning values ​​through machine learning algorithms;

[0187] The decision generation unit is used to trigger cleaning or spraying commands according to preset rules.

[0188] The linkage control module includes: a path planning unit, a sprinkler control unit, a control unit, and a safety obstacle avoidance unit;

[0189] The path planning unit is used to generate the optimal cleaning path using an improved greedy algorithm.

[0190] The spray control unit is used to control the start and stop of the spray heads and the water pressure according to the temperature difference zones.

[0191] The control unit is used to drive the cleaning robot arm to work along the planned path.

[0192] The safety obstacle avoidance unit is used to detect and avoid wet spray areas in real time using lidar.

[0193] The effect evaluation module includes: a coupling degree analysis unit, an effect grading unit, and a parameter adjustment unit;

[0194] The coupling analysis unit is used to construct a two-dimensional evaluation matrix of back pressure and attenuation coefficient;

[0195] Effect grading unit, used to classify control effects;

[0196] The parameter adjustment unit is used to automatically adjust the spray intensity or cleaning frequency according to the grade.

[0197] The mode switching module includes: a steady-state determination unit, an anomaly monitoring unit, and a mode switching unit;

[0198] The steady-state determination unit is used to detect the duration of stability of back pressure and ambient temperature;

[0199] Anomaly monitoring unit is used to identify parameter mutations;

[0200] The mode switching unit is used to automatically trigger the switching between intelligent control mode and conventional monitoring mode.

[0201] 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 present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A coordinated control method for an intercooled air-cooled island, characterized in that, include: S1: Build an intelligent sensing network to collect key operating parameters in real time, construct an evaluation model based on the key operating parameters, and calculate the fin heat exchange efficiency decay coefficient and back pressure warning value; the key operating parameters include fin temperature, back pressure, ambient temperature and contaminant thickness. S2: When the fin heat exchange efficiency attenuation coefficient is lower than the preset value or the back pressure reaches the back pressure warning value, the linkage control mechanism of the intelligent cleaning system and the intelligent spraying system is activated simultaneously, and the initial cleaning and spraying control commands are output. S3: Based on the current back pressure and fin heat exchange efficiency attenuation coefficient, construct a control effect evaluation matrix based on the coupling degree of back pressure-attenuation coefficient and evaluate the control effect level of cleaning and spraying. Adjust the spraying intensity or cleaning frequency according to the control effect level and output the adjusted control parameters. S4: Until the ambient temperature and back pressure reach the preset normal conditions and remain so for a preset duration, output a control endpoint signal and switch to the normal monitoring mode based on the control endpoint signal; the normal monitoring mode maintains real-time acquisition and monitoring of the key operating parameters, and once an abnormal change in the key operating parameters is detected, immediately switch back to the intelligent control mode and restart the intelligent cleaning system or the intelligent spraying system. The preset normal conditions are: the ambient temperature is within the equipment's designed operating ambient temperature range, the back pressure is lower than the back pressure warning value, and the fin heat exchange efficiency decay coefficient is greater than or equal to the preset value; the preset value is the critical threshold of the fin heat exchange efficiency decay coefficient.

2. The method for coordinated control of an intercooled air-cooled island as described in claim 1, characterized in that, The evaluation model includes a parameter-attenuation coefficient mapping model and a back pressure warning value calculation model. The construction process of the evaluation model includes: Acquire historical key operating parameters and perform noise filtering and timing alignment processing; A multi-tree collaborative model was adopted, with fin temperature, ambient temperature, and contaminant thickness as input features, and the measured fin heat transfer efficiency attenuation coefficient as a label for training, to obtain a parameter-attenuation coefficient mapping relationship model. Based on historical back pressure data, the mean and standard deviation of back pressure under different ambient temperatures are calculated using the sliding window method. Combined with the equipment safe operation threshold, a back pressure early warning value calculation model is constructed.

3. The method for coordinated control of an intercooled air-cooled island as described in claim 2, characterized in that, The multi-tree collaborative model is trained using fin temperature, ambient temperature, and contaminant thickness as input features, and the measured fin heat transfer efficiency attenuation coefficient as a label, to obtain a parameter-attenuation coefficient mapping relationship model, including: A1: Obtain fin temperature, ambient temperature, contaminant thickness, and measured fin heat transfer efficiency attenuation coefficient, and combine them to form an input parameter set. Among them, fin temperature, ambient temperature, and contaminant thickness are used as input features, and the measured fin heat transfer efficiency attenuation coefficient is used as a label. A2: Define three types of nodes in the model, including root nodes, internal nodes, and leaf nodes; the root node is used to receive the input parameter set and initiate the first split; the internal nodes are used to store split features, split thresholds, and the link relationships between left and right child nodes for sample splitting; the leaf nodes are used to store the subset of samples that cannot be split further and the mean of the labels of that subset, which is output as the final predicted value. A3: Set the node termination split condition; the node termination split condition is that the number of sample subsets of the current node is less than a preset number, or the variance of the labels in the sample subsets of the current node is less than a preset maximum variance, or the tree depth reaches a preset upper limit. A4: Randomly select a subset of samples of the same size from the input parameter set as the sample subset of the first tree, and at the same time randomly select some features from the input features as candidate splitting features of the first tree; the sample subset of the first tree is allowed to be repeatedly extracted with the same parameters during the extraction process. A5: Pass the sample subset of the first tree to the root node, and use the mean of all sample labels in the root node as the initial prediction value of the node. A6: For the sample subset of the current node, if the node termination splitting condition is met, the current node becomes a leaf node, and its predicted value is the mean of all sample labels in the sample subset of the current node. A7: For the current node that does not meet the termination condition, traverse the candidate splitting features of the first tree, sort the values ​​of each candidate splitting feature in ascending order, and take the midpoint of adjacent values ​​as the candidate threshold of the candidate splitting feature. A8: For each candidate splitting feature and the corresponding candidate threshold, split the sample subset of the current node into a left subset and a right subset, calculate the variance of the label in the parent node, the variance of the label in the left subset, and the variance of the label in the right subset respectively, and then obtain the variance reduction by subtracting the weighted sum of the variances of the left and right subset labels from the variance of the label in the parent node. A9: From all combinations of candidate splitting features and candidate thresholds, select the combination with the largest reduction in variance as the splitting feature and splitting threshold of the current node, i.e., the optimal splitting combination; A10: Based on the optimal splitting combination, split the sample subset of the current node into a left subset and a right subset, generate a left child node and a right child node respectively, and pass the left subset and right subset into the two child nodes respectively. Then repeat A6-A9 for the left child node and the right child node until all child nodes meet the node termination splitting condition and become leaf nodes. The training of the first tree model is completed. A11: Repeat steps A4 to A10 to generate a preset number of tree models, and optimize the structure of each tree model using a post-pruning strategy to obtain optimized multi-tree models; A12: Integrate all optimized multi-tree models into a multi-tree collaborative model; A13: When it is necessary to predict the fin heat transfer efficiency attenuation coefficient of a new sample, the input features of the new sample are fed into each tree model respectively. Each tree outputs a predicted value, which is the average label value of the leaf node to which the sample finally arrives. Then, the average of the predicted values ​​of all trees is taken as the final output of the multi-tree collaborative model, resulting in the parameter-attenuation coefficient mapping relationship model.

4. The method for coordinated control of an intercooled air-cooled island as described in claim 3, characterized in that, The calculation method for the fin heat transfer efficiency attenuation coefficient is as follows: The average fin heat exchange efficiency over 24 consecutive hours after the first operation of the new equipment or after complete cleaning is used as the benchmark value. The current heat exchange efficiency is calculated by combining the temperature difference between the fins and the ambient temperature and the thickness of the contaminants collected in real time with Newton's cooling formula. The ratio of the current heat exchange efficiency to the baseline value is used as the fin heat exchange efficiency decay coefficient.

5. The method for coordinated control of an intercooled air-cooled island as described in claim 1, characterized in that, The linkage control mechanism for activating the intelligent cleaning system and the smart sprinkler system includes: The intelligent cleaning system is activated, and a dynamic priority division method based on the coupling of contaminants and temperature is adopted. Priority areas are divided by combining the collected contaminant thickness and fin temperature and cleaned in sequence. The intelligent determination system for the cleaning endpoint is constructed based on the changes in the fin temperature field transmitted back in real time by the sensing network to determine whether the cleaning is completed. The intelligent spray system is activated simultaneously, and a graded spraying strategy is implemented according to the fin temperature-divided areas. At the same time, the spray water pressure is monitored and adjusted and compensated, and the adjustment and compensation results are fed back to the intelligent sensing network in real time.

6. The method for coordinated control of an intercooled air-cooled island as described in claim 5, characterized in that, The dynamic prioritization method for cleaning based on the coupling of contaminants and temperature includes: The surface of the air-cooled island fins is divided into standard grid units. Each grid unit is equipped with an independent temperature sensor and laser contaminant detector to collect the fin temperature and ambient temperature of the grid unit in real time and calculate the contaminant thickness. The contaminant thickness is divided into N level ranges, and the fin temperature difference is divided into M level ranges; the fin temperature difference is the temperature difference between the fin temperature and the ambient temperature. Construct an N×M two-dimensional priority matrix; each element in the priority matrix corresponds to a cleaning priority for a combination of working conditions; the cleaning priority is divided into a first-level queue, a second-level queue, and a third-level queue; All grid cells are scanned in real time. Based on the currently collected contaminant thickness and fin temperature difference, the grid cells are classified into a first-level queue, a second-level queue, or a third-level queue according to a dynamic scoring rule, and the spatial coordinates of the grid cells in each queue are recorded. The dynamic scoring rule is a predefined correspondence between the level combination of contaminant thickness and fin temperature difference and the cleaning priority. An improved greedy algorithm is used to generate the shortest cleaning path covering all first-level queue grids.

7. The method for coordinated control of an intercooled air-cooled island as described in claim 6, characterized in that, The process of generating the shortest cleaning path covering all primary queue grids using an improved greedy algorithm includes: Extract all grid cells classified into the first-level queue from the intelligent sensing network, record the physical coordinates of each cell, and read the contaminant thickness and fin temperature difference of each grid cell. The contaminant thickness and fin temperature difference of all grid cells are normalized to the same scale, and their cleaning urgency is calculated according to a preset weight; the cleaning urgency is a comprehensive score of contaminant thickness and fin temperature difference. Obtain the coordinates of the currently spraying wet area from the intelligent sprinkler system, mark it as a restricted area, and generate a restricted area list; Using the current position of the cleaning robot as the starting point of the path, add all primary queue grids to the list to be cleaned, and at the same time, remove all grid cells that overlap with the restricted area; For each remaining grid cell in the cleaning list, the weighted sum of three indicators—Euclidean distance, cleaning urgency, and movement energy consumption—is calculated to obtain the comprehensive cost; the movement energy consumption refers to the energy cost consumed by the intelligent cleaning device in the process of moving from its current grid cell to any target grid cell in the cleaning list. Based on the three indicators, the grid cell with the lowest overall cost is selected as the next target cleaning point. The robot is then controlled to move to the target cleaning point to perform the cleaning operation. After the cleaning is completed, the grid cell is removed from the list of cells to be cleaned. If the spray system opens a new wet zone during the cleaning process, the no-entry list should be updated immediately. Check the generated paths. If three detours are found, delete the intermediate target points. If the list to be cleaned is empty, it means that all first-level queue grid cells have been cleaned and the path planning is complete.

8. The method for coordinated control of an intercooled air-cooled island as described in claim 5, characterized in that, The determination logic of the intelligent determination system for the cleaning endpoint based on the fin temperature field feedback is as follows: The system monitors the rate of change and uniformity of the temperature field of the fins in real time during the cleaning process. When the rate of change of the temperature field is lower than the preset rate threshold and the uniformity of the temperature field distribution reaches the preset uniformity threshold, it determines that the cleaning of the area is complete and feeds the determination result back to the intelligent cleaning system to control it to switch to the cleaning operation of the next priority area.

9. The method for coordinated control of an intercooled air-cooled island as described in claim 1, characterized in that, The process of constructing the control effect evaluation matrix based on back pressure-attenuation coefficient coupling degree includes: Using the back pressure change rate and the fin heat transfer efficiency attenuation coefficient change rate as two-dimensional coordinate axes, K evaluation regions are divided. Each evaluation region corresponds to a control effect level. By calculating the coupling degree index between back pressure and attenuation coefficient, the current control effect level is determined. The process of calculating the coupling index between back pressure and attenuation coefficient includes: Determine the standardized contribution values ​​of the rate of change of back pressure and the rate of change of attenuation coefficient; Different weights are assigned to the rate of change of back pressure and the rate of change of attenuation coefficient, with the weight of back pressure being 0.6 and the weight of attenuation coefficient being 0.4, and the total weight being 1. Calculate the final coupling degree index; the final coupling degree index is equal to the product of the standardized contribution value of the back pressure change rate and the back pressure weight, plus the sum of the standardized contribution value of the attenuation coefficient change rate and the attenuation coefficient weight.

10. A smart control system for an intercooled air-cooled island, used to implement the coordinated control method for an intercooled air-cooled island as described in any one of claims 1-9, characterized in that, include: Intelligent sensing module, evaluation and decision-making module, linkage control module, effect evaluation module, mode switching module; The intelligent sensing module is used to collect key operating parameters of the air-cooled island in real time; The evaluation and decision-making module is used to construct a multi-parameter fusion evaluation model and generate control strategies. The linkage control module is used to perform coordinated control of intelligent cleaning and spraying; The effect evaluation module is used to quantify the control effect and dynamically optimize the strategy; The mode switching module is used to achieve seamless switching between intelligent control and routine monitoring.

Images