Performance comprehensive evaluation method and system for coordinated control system of high-alkali coal unit

Abstract

The invention relates to the technical field of high-alkali coal unit coordination control, and discloses a high-alkali coal unit coordination control system performance comprehensive evaluation method and system, and the method comprises the steps: collecting operation parameter data through a plurality of sensors disposed at key positions of a unit to generate a performance data set, and receiving real-time data through an evaluation server to construct a performance matrix; the performance matrix is processed, features are extracted in combination with a performance data set, performance is predicted according to the operation state gradient and control parameter information, performance index space-time distribution features are output through a comprehensive evaluation model, and the performance data set is updated; and dynamically adjusting unit coordination control according to the evaluation indexes, wherein the operations comprise valve opening, fuel supply and the like. The system comprises a plurality of sensors, an evaluation server, a performance evaluation index determination module and a dynamic adjustment module. According to the invention, comprehensive and dynamic evaluation and accurate control of the performance of the coordinated control system of the high-alkali coal unit are realized, and the safety, stability and economical efficiency of unit operation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of coordinated control of high-alkali coal units, and in particular to a comprehensive performance evaluation method and system for a coordinated control system of a high-alkali coal unit. Background Art

[0002] High-alkali coal, a special type of coal, is prone to slagging, fouling, and corrosion during combustion due to its high alkali metal content, seriously impacting the safe, stable, and economical operation of the unit. The coordinated control system for high-alkali coal-fired units requires precise control of multiple key parameters to ensure efficient and stable operation. However, current performance evaluation methods for coordinated control systems of high-alkali coal-fired units have numerous shortcomings and are difficult to meet practical application requirements.

[0003] From the perspective of data collection and processing, traditional performance evaluation methods typically rely on data collected by sensors in only a few key locations. This limited data coverage fails to fully reflect the operating status of all parts of the unit. Furthermore, the collected data lacks effective processing and analysis methods, making it difficult to extract key features that accurately characterize unit performance. For example, when processing parameters such as temperature and pressure, the temporal and spatial distribution characteristics of the data are not fully considered, resulting in an inaccurate assessment of the unit's operating status.

[0004] Existing methods for determining performance evaluation indicators mostly use a fixed index system, failing to fully consider the dynamic characteristics of high-alkali coal-fired units during operation. High-alkali coal-fired units experience significant variations in operating conditions under varying loads and coal qualities, and fixed evaluation indicators are unable to flexibly adapt to these variations, resulting in evaluation results that fail to truly reflect the unit's actual performance. Furthermore, traditional methods lack analysis of the spatiotemporal distribution of performance indicators, making it impossible to promptly detect abnormal operating conditions in localized areas of the unit, hindering comprehensive and dynamic evaluation of unit performance.

[0005] From the perspective of performance prediction and control adjustment, traditional methods lack performance prediction capabilities and are unable to accurately predict future unit performance changes based on historical and real-time operating data, resulting in a lack of foresight in control adjustments. When adjusting unit coordinated control, experience-based control strategies are typically used. These adjustments are less targeted and accurate, and are unable to provide refined control based on the specific performance of each unit. This can easily lead to over-control or under-control issues, impacting the unit's coordinated control effectiveness and operational efficiency.

[0006] With the increasing application of high-alkali coal units in power production, there is an urgent need for a method and system that can comprehensively and accurately evaluate the performance of the unit's coordinated control system and realize dynamic adjustment, so as to address the shortcomings of traditional methods and improve the safety, stability and economy of the unit. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and system for comprehensively evaluating the performance of a coordinated control system of a high-alkali coal unit, so as to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a comprehensive performance evaluation method for a high-alkali coal unit coordinated control system, which is applied to the coordinated control system of a high-alkali coal unit. The system includes multiple sensors deployed at key locations of the unit and an evaluation server connected to the sensors. Two adjacent sensors are separated by a set distance. The method includes: Collecting the unit's operating parameter data through the sensor to generate a performance data set; Receive real-time operating data of multiple sensors through the evaluation server to build a unit performance matrix; Determining a performance evaluation indicator based on the performance data set and the real-time data of each sensor, wherein determining the performance evaluation indicator includes: processing a performance matrix, extracting features based on the performance data set, and performing performance prediction based on an operating state gradient and control parameter information, outputting spatiotemporal distribution characteristics of the performance indicator through a comprehensive evaluation model, and updating the performance data set based on the spatiotemporal distribution characteristics; According to the evaluation indicators, the coordinated control of the unit is dynamically adjusted.

[0009] Preferably, determining the performance evaluation index includes: Processing the performance matrix to extract temperature distribution characteristics, pressure fluctuation characteristics, and performance change trends; Performing an operating state model on the performance matrix according to the temperature distribution characteristics and the pressure fluctuation characteristics, dividing the unit into a plurality of subunits and marking the unit identifiers, performing an association match with the performance data set according to the temperature distribution of the subunits, and marking the unit identifiers in the performance data set; Calculating an operating state gradient according to the sensor position, predicting the distribution of performance indicators according to the operating state gradient and the performance change trend, and calculating performance prediction information for each subunit; Constructing a comprehensive evaluation model, taking the performance prediction information as an input parameter of the comprehensive evaluation model, performing spatial correlation modeling on the performance prediction information through the comprehensive evaluation model, and outputting spatiotemporal distribution characteristics of performance indicators; The performance data set is updated according to the spatiotemporal distribution characteristics to obtain performance evaluation indicators.

[0010] Preferably, the processing of the performance matrix includes: Normalize the performance matrix, intercept the performance hotspot areas in the matrix through a sliding window, filter the noise in the hotspot areas, and calculate the performance change trend through the eigendecomposition algorithm; Calculating spatial correlation features of the performance matrix, calculating interference intensity, stability coefficient, and anomaly index between units based on the spatial correlation features, constructing a feature fusion network, and calculating pressure fluctuation features through the feature fusion network; The time domain features and frequency domain features collected by each sensor are extracted, and the sensor's operating feature vector is calculated based on the phase difference between the time domain features and the frequency domain features. Based on the operating feature vector, feature matching is performed on sensors at different positions, and the performance change trend is calculated.

[0011] Preferably, the performing operating state modeling on the performance matrix includes: According to the temperature distribution characteristics, temperature sampling points are extracted from each frame of data, and the sampling points are correlated and mapped with the pressure fluctuation characteristics to generate an operating state map. The operating state maps collected by multiple sensors are spatially aligned to calculate the operating state distribution of the unit; Set the interference threshold, locate the abnormal source based on the temperature distribution value of the multi-frame performance matrix, and calculate the interference intensity difference. If the interference intensity difference is greater than or equal to the interference threshold, it indicates that there is an operational abnormality in the unit. The control model constraint compensation is performed on the current unit. According to the thermodynamic model corresponding to the current unit, the operating state distribution of the current unit is iteratively corrected. The intensity compensation value of the abnormal unit is calculated based on the correction result. The performance matrix is ​​modeled for the operating state according to the operating state distribution, and the operating model of the unit is marked for fluctuations according to the pressure fluctuation characteristics.

[0012] Preferably, the calculating the operating state gradient according to the sensor position includes: Based on multiple sets of operating status data, temperature change points are extracted and mapped to a unified coordinate system based on the deployment location of the sensors. The change points are fitted using a spatial interpolation algorithm to generate the unit's operating field model. Performing sampling at equal intervals along the transmission path of the operating field model, calculating the thermal conductivity, fluctuation index, and state change slope of the path based on the sampling results, and calculating the state change parameter based on the thermal conductivity, fluctuation index, and state change slope; Based on the sensor deployment parameters and acquisition accuracy, the distribution characteristics of performance changes in each frame of data are projected onto the operational field model. The operational field model is partitioned along the transmission direction according to the number of sensors. The variation patterns of performance changes within the partitions are analyzed, and the performance distribution characteristics are calculated based on these variation patterns. An operating state gradient is calculated based on the state change parameter and the performance distribution characteristics. The calculation process of the operating state gradient includes: based on the position range from the first sensor to the last sensor, selecting spatial coordinate points in the sensor deployment direction, cumulatively calculating the product of the operating field strength characteristic weight value and the performance distribution characteristic weight value within the spatial resolution range, and superimposing the influence value of the sensor acquisition frequency on the operating state change rate.

[0013] Preferably, the calculating of the performance prediction information of each subunit includes: Using the main transmission path of the operating field model as a baseline, taking the peak position of the performance change in each frame of data as a reference point, calculating the performance offset, and drawing a performance distribution curve according to the coordinates; Correcting the growth rate and direction of the performance change trend according to the operating state gradient; Starting from the most recent performance distribution point, the distribution curve is continuously drawn based on the correction results of the growth rate and direction to generate the performance distribution points for the next period until the distribution points cover the entire target unit and generate performance prediction information.

[0014] Preferably, the building of a comprehensive evaluation model includes: The input layer is used to organize the performance prediction information into spatially distributed data and perform normalization processing; The feature fusion layer is used to extract unit-related features of performance by processing spatially distributed data and build dependency relationships between physical units; The resource allocation layer is used to integrate the correlation between performance indicators in spatial units and generate control parameter scheduling strategies.

[0015] Preferably, the obtaining of performance evaluation indicators includes: According to the spatiotemporal distribution characteristics of the performance indicators output by the comprehensive evaluation model, the identification of the sub-unit is matched with the spatiotemporal distribution characteristics; The unit data in the performance data set are reorganized according to the spatiotemporal characteristics to generate a unit distribution map sorted by the strength of the performance indicators; According to the reorganized unit distribution map, the optimized performance evaluation index is output.

[0016] Preferably, the dynamic adjustment of the coordinated control of the units includes: Map unit identifiers to areas of performance evaluation indicators one by one; According to the spatiotemporal distribution characteristics of performance indicators, the execution actions of coordinated control are controlled, including valve opening adjustment, fuel supply and flow distribution operations; Dynamically allocate control parameters to corresponding physical units based on the spatial distribution of performance indicators and preset adjustment strategies; The execution actions of the control coordination control include: When the performance index reaches a preset intensity threshold in the target unit, an opening adjustment instruction of the adjacent valve is triggered; Dynamically combine available resources according to the fuel supply strategy to generate a flow allocation vector; An actuator parameter of a target unit control node is adjusted based on the traffic distribution vector.

[0017] Preferably, the present invention also includes a comprehensive performance evaluation system for the coordinated control system of a high-alkali coal unit, which is applied to the comprehensive performance evaluation method for the coordinated control system of a high-alkali coal unit as described above. The system includes multiple sensors deployed at key positions of the unit and with a set distance between two adjacent sensors, and an evaluation server connected to the sensors, characterized in that: the multiple sensors are used to collect unit operating parameter data, and the operating parameter data include component data characterizing the metal content of high-alkali coal, temperature and pressure data characterizing combustion characteristics, and coking thickness data characterizing coking conditions, thereby generating a performance data set; the evaluation server is used to receive real-time operating data from multiple sensors and construct a unit performance matrix; the evaluation server is also provided with performance evaluation indicators A determination module is used to process the performance matrix and perform feature extraction in combination with the performance data set. The feature extraction includes extracting the distribution characteristics of metal elements in high-alkali coal, the abnormal temperature and pressure fluctuation characteristics caused by the characteristics of high-alkali coal during combustion, and the change characteristics of the degree of coking of high-alkali coal. Performance prediction is performed based on the operating state gradient and control parameter information. The spatiotemporal distribution characteristics of the performance indicators are output through a comprehensive evaluation model, and the performance data set is updated based on the spatiotemporal distribution characteristics. The system is also provided with a dynamic adjustment module for dynamically adjusting the coordinated control of the unit according to the evaluation indicators. The dynamic adjustment includes adjusting the fuel ratio according to the metal content of the high-alkali coal, optimizing the burner parameters according to the combustion characteristics, and controlling the soot blowing frequency and intensity according to the coking situation.

[0018] Compared with the prior art, the present invention has the following beneficial effects: In terms of data collection and processing, by deploying multiple sensors at key locations on the unit, with adjacent sensors spaced at a set distance, comprehensive, real-time collection of unit operating parameters is achieved, yielding a performance data set covering all key parts of the unit. The evaluation server receives real-time operating data from multiple sensors and constructs a unit performance matrix, providing a rich and comprehensive data foundation for subsequent performance analysis. By standardizing the performance matrix, capturing hotspots with a sliding window, filtering noise, and performing feature decomposition, key information such as temperature distribution characteristics, pressure fluctuation characteristics, and performance change trends can be effectively extracted. This improves the accuracy and effectiveness of data processing and lays the foundation for accurate unit performance evaluation.

[0019] In determining performance evaluation indicators, through in-depth processing and analysis of the performance matrix, combined with operating state gradients and control parameter information, performance prediction is performed. Using a comprehensive evaluation model to output the spatiotemporal distribution characteristics of performance indicators, the performance of each subunit of the unit can be comprehensively and dynamically assessed. Dividing the unit into multiple subunits and labeling them with their identifiers enables a refined assessment of unit performance and timely detection of abnormal operating conditions in local areas. Updating the performance dataset based on the spatiotemporal distribution characteristics allows evaluation indicators to reflect changes in the unit's operating status in real time, improving both their timeliness and accuracy.

[0020] In terms of performance prediction and control adjustment, by constructing an operating field model, calculating the operating state gradient and performance prediction information, it is possible to accurately predict the changing trend of unit performance, providing a forward-looking basis for control adjustment. The construction of a comprehensive evaluation model realizes the spatial correlation modeling of performance prediction information, can deeply analyze the dependencies between physical units, and generate a reasonable control parameter scheduling strategy. Dynamic adjustment of unit coordinated control based on performance evaluation indicators can accurately allocate control parameters to corresponding physical units, and realize refined control of execution actions such as valve opening, fuel supply and flow distribution. For example, when the performance indicator reaches the preset intensity threshold in the target unit, the opening adjustment instruction of the adjacent valve is triggered, which can respond to the abnormal state of the local area of ​​the unit in a timely manner and improve the safety, stability and operating efficiency of the unit.

[0021] In addition, the present invention forms a closed-loop control system through real-time monitoring, performance evaluation and dynamic adjustment of the unit's operating status, which can continuously optimize the unit's operating status and improve the overall performance of the high-alkali coal unit's coordinated control system. It effectively solves the problems of reduced safety, stability and economy caused by slagging, contamination, corrosion and other problems during the operation of the high-alkali coal unit, and has broad application prospects and significant economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a working principle diagram of the comprehensive performance evaluation method for the high-alkali coal unit coordinated control system according to the present invention; Figure 2 Flowchart for determining performance evaluation metrics; Figure 3 Flowchart for performance matrix processing; Figure 4 A flow chart that models operational states. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] See also Figures 1-4 The present invention relates to a comprehensive performance evaluation method for a high-alkali coal unit coordinated control system, applied to a system comprising multiple sensors and an evaluation server. The sensors are deployed at key locations within the unit, with the spacing between adjacent sensors set to a predetermined value. The evaluation server is connected to the sensors, and the specific implementation steps are as follows: Various sensors are used to collect real-time data on unit operating parameters. For example, alkali metal vapor concentration sensors are installed in areas prone to slagging, such as the furnace combustion area and superheater tube bundles, to monitor the spatial distribution and concentration of alkali metal volatiles such as Na and K in the furnace in real time. Slag thickness monitoring sensors are placed on heating surfaces such as water-cooled walls and superheaters, collecting real-time data on slagging layer thickness and contamination using infrared ranging or acoustic flaw detection technology. Online coal quality analyzers are deployed at the coal feeder inlet and pulverizer outlet to obtain characteristic parameters of the incoming coal, such as alkali metal (Na2O, KO) content, ash melting point, and ash viscosity. Flue gas composition sensors are installed at the air preheater outlet and at different height sections of the furnace to collect data such as O2 concentration, SOx, NOx, and alkali metal vapor partial pressure for analyzing the impact of the oxidizing atmosphere in the furnace on alkali metal volatilization. Operational status sensors are installed in the sootblower system to record data such as the purge frequency, purge pressure, and action time at each sootblowing point, which can be correlated with the effectiveness of slagging control. These sensors, deployed at set intervals, collect high-alkali coal-specific parameters such as alkali metal concentration distribution, slag thickness, and coal quality characteristics, along with conventional operating parameters (such as main steam temperature, furnace pressure, and feedwater flow) to the evaluation server, thereby constructing a multidimensional performance dataset encompassing the characteristics of high-alkali coal. The evaluation server receives real-time operating data from each sensor and structures it by sensor location and parameter type, constructing a performance matrix reflecting the overall operating status of the unit. Each element in the matrix corresponds to the parameter value of a specific sensor at a specific moment.

[0025] The performance matrix is ​​processed, features are extracted from the performance dataset, and performance predictions are made using operating state gradients and control parameter information. Ultimately, the comprehensive evaluation model outputs the spatiotemporal distribution characteristics of performance indicators, which are then used to update the performance dataset. This includes preprocessing the performance matrix through noise reduction and feature decomposition to extract features such as temperature distribution and pressure fluctuations; calculating operating state gradients based on sensor locations and analyzing spatial gradient differences in parameter changes; and using the comprehensive evaluation model to fuse features and predict the temporal and spatial distribution patterns of performance indicators.

[0026] Based on the determined performance evaluation indicators, the coordinated control strategy of the unit is dynamically adjusted. For example, based on the temporal and spatial distribution of the performance indicators, the valve opening, fuel supply, flow distribution and other execution actions are adjusted to achieve real-time optimization of the unit's operating status.

[0027] The present invention will be further described below in conjunction with Examples 1 to 5: Example 1: This embodiment refines the process of determining performance evaluation indicators. This process begins with performance matrix processing and implements multi-dimensional analysis and prediction of the unit's operating status through feature extraction, operating status modeling, operating status gradient calculation, and performance prediction information generation. The specific implementation is as follows: The performance matrix is ​​processed to extract essential features. The performance matrix is ​​constructed from real-time operating data from multiple sensors received by the evaluation server. It contains information on the spatiotemporal distribution of parameters such as temperature, pressure, and flow at key locations within the unit. During processing, the matrix is ​​first normalized. A normalization algorithm converts parameters of varying dimensions into a uniform range, eliminating the impact of dimensional differences on subsequent analysis. Next, a sliding window technique is used to locate performance hotspots within the matrix. These are areas where parameter fluctuations exceed normal fluctuations or exhibit abnormal coupling relationships, such as sudden temperature increases between different combustion zones in the furnace or sudden pressure changes in pipelines. A noise filtering algorithm is applied to these hotspots. By setting frequency or dynamic thresholds, high-frequency random noise or low-frequency drift signals is removed, retaining valid data features that reflect the actual operating status. Subsequently, an eigendecomposition algorithm is used to reduce the matrix's dimensionality. By extracting principal components or singular values, the high-dimensional parameter space is mapped to a low-dimensional feature space. Key eigenvectors that characterize performance trends are identified, such as the dominant conduction direction of the temperature field or the dominant mode of pressure fluctuation.

[0028] After extracting the temperature distribution characteristics, pressure fluctuation characteristics, and performance change trends, the performance matrix needs to be modeled for operating status. During the modeling process, the unit is divided into multiple subunits based on the temperature distribution characteristics. The division of subunits is determined based on the physical structure of the unit (such as furnace partitions, heating surface tube groups, steam-water system loops, etc.) and the sensor deployment location. Each subunit is assigned a unique unit identifier. By associating and matching the temperature distribution data of the subunits with the historical performance data set, marking the corresponding unit identifier in the performance data set, and establishing a spatial mapping relationship between real-time operating data and historical data. For example, the burner area on the front wall of the furnace is divided into subunit A, and its temperature distribution data is matched with the operating condition data of the start-stop phase, load change phase, etc. of the same area in the historical data, so as to facilitate subsequent analysis of the performance of the area under the current operating conditions.

[0029] Calculate the operating state gradient, which describes the rate and direction of spatial change of unit parameters. The specific steps are: Using multiple sets of continuously collected operating state data, identify points with significant temperature changes (such as temperature gradient abrupt changes or isothermal inflection points). Based on the three-dimensional sensor deployment locations, these change points are mapped to a unified Cartesian coordinate system to form a discrete set of spatial data points. A spatial interpolation algorithm is used to fit these discrete points to generate a continuous operating field model. This model simulates the spatial distribution and conduction patterns of parameters such as temperature and pressure. For example, it depicts the gradient field of decreasing furnace temperature from the burner to the water-cooled wall. Samples are taken at equal intervals along the main transmission paths of the operating field model (such as the flue gas flow direction and the working fluid flow direction) to obtain the thermal conductivity, fluctuation index, and state change slope at each sampling point. The thermal conductivity reflects the rate of temperature change per unit distance, the fluctuation index is calculated from the root mean square value of the pressure signal, and the state change slope is obtained by fitting the time-varying trend of the parameter. Combining these three parameters, a state change parameter is generated to characterize the spatial variation characteristics. At the same time, considering the sensor deployment parameters (such as spacing and angle) and acquisition accuracy, the distribution characteristics of performance changes in each frame of data (such as the expansion speed of high-temperature areas and the propagation range of pressure fluctuations) are projected onto the operating field model. The model is then divided into multiple analysis partitions along the transmission direction. By statistically analyzing the mean, variance, extreme values, and other characteristics of performance changes within each partition, the variation patterns are analyzed and the performance distribution characteristics are calculated, such as the periodicity of temperature changes or the attenuation characteristics of pressure fluctuations within a partition. Ultimately, the calculation of the operating state gradient comprehensively considers the state change parameters and performance distribution characteristics. A series of spatial coordinate points are selected in the sensor deployment direction. The product of the weighted value of the operating field strength characteristics and the weighted value of the performance distribution characteristics within the spatial resolution range of each coordinate point is cumulatively calculated. The influence of the sensor acquisition frequency on the operating state change rate is then added to obtain the operating state gradient that reflects the comprehensive intensity of parameter spatial changes.

[0030] Based on the operating state gradient and performance change trends, performance prediction information for each subunit is further calculated. First, using the main transmission path of the operating field model as the baseline, such as the furnace centerline or the main steam line centerline, the peak position of the performance change in each frame of data (such as the maximum temperature point or the minimum pressure point) is used as the reference point. The offset of the reference point relative to the baseline is calculated. This offset reflects the spatial position deviation of the performance anomaly. Based on the offset and the performance parameter value, a performance distribution curve is plotted in a coordinate system. The horizontal axis of the curve represents the spatial position and the vertical axis represents the parameter value, visually displaying the spatial distribution of the performance parameter at the current moment. The operating state gradient is then used to correct the growth rate and direction of the performance change trend. For example, if the operating state gradient indicates that the temperature change rate in a certain area is higher than the average, the growth rate of the performance distribution curve in that area is corrected to better reflect the actual change trend. If the gradient direction is opposite to the preset normal conduction direction, the extension direction of the curve is adjusted to reflect the possible presence of abnormal conduction paths. After the correction is complete, the distribution curve is drawn from the most recent performance distribution point, following the corrected growth rate and direction. Performance distribution points for the next time period are generated through iterative calculations until the distribution points cover the entire target subunit, forming a complete performance forecast. This information includes the predicted parameter values ​​for each spatial location of the target subunit in the future time period, providing input data for the comprehensive evaluation model and a predictive basis for dynamic adjustments to the unit's coordinated control.

[0031] Example 2: This embodiment details the performance matrix processing process. This process cleans, reduces dimensions, and extracts features from multi-source heterogeneous data through standardization, hotspot location, noise filtering, feature decomposition, spatial correlation analysis, and feature fusion. This provides high-quality feature input for subsequent operational status modeling and performance evaluation. The specific implementation is as follows: Standardize the performance matrix. The performance matrix consists of real-time data collected by multiple sensors deployed at key locations on the unit. It includes multiple parameters, such as temperature, pressure, flow rate, and speed. These parameters cannot be directly compared and analyzed due to their different dimensions (e.g., temperature in °C, pressure in MPa, and flow rate in t / h). Standardization uses a normalization or standardization algorithm to convert each dimensional parameter into a dimensionless value. The data is typically mapped to the [0, 1] interval or conforms to a standard normal distribution. For example, for temperature parameters, the minimum-maximum normalization method is used: the current value is subtracted from the historical minimum value and then divided by the difference between the historical maximum and minimum values ​​to obtain the normalized temperature value. For pressure parameters, the Z-score normalization method is used: the difference between the current value and the mean is calculated and divided by the standard deviation to obtain the normalized pressure value. Standardization eliminates the interference of dimensional differences on feature analysis and ensures the consistency of the weights of different parameters in subsequent calculations.

[0032] After standardization, the performance hotspot areas in the matrix are intercepted using the sliding window technology. A sliding window is an analysis window that slides in the time or space dimension, and its size and step size are set according to the sensor deployment density and the frequency of parameter changes. For example, in a furnace temperature monitoring scenario, if a sensor is deployed every 2 meters along the furnace height, the window size can be set to a spatial range that includes 5 adjacent sensors, with a step size of 1 sensor spacing, so that the window slides successively along the furnace height. During the window sliding process, statistics such as the variance and coefficient of variation of the parameters within the window are calculated. When the statistics exceed the preset threshold (such as the upper limit of the normal fluctuation range calculated based on historical data), the window area is determined to be a performance hotspot area, that is, an area with abnormal parameter fluctuations or potential failure risks. For example, a significant increase in the variance of the temperature parameters within a window may indicate that the combustion in the area is unstable or there is a slagging trend, which requires focused analysis.

[0033] To filter noise in hotspots, classic filtering algorithms from the field of signal processing are employed. Common filtering methods include median filtering, Gaussian filtering, and Butterworth filtering. The appropriate filtering method should be selected based on the noise type. If the noise is salt and pepper (manifested as random pulse interference in the data), a median filter is used. This sorts the data within a window and replaces the current value with the median value, effectively removing isolated noise points. If the noise is Gaussian (manifested as random fluctuations following a normal distribution), a Gaussian filter is used. This smoothes the data while preserving edge features by convolving a two-dimensional Gaussian kernel with the data. During the filtering process, the filter window size should be adjusted based on the sensor acquisition frequency and parameter change rate. For example, a smaller filter window can be used for high-frequency pressure data to preserve detailed features, while a larger window can be used for low-frequency temperature data to enhance smoothing. After filtering, the effective signal in the hotspot is highlighted, providing a clean data foundation for subsequent feature extraction.

[0034] Eigendecomposition algorithms are used to extract performance trends. Common methods include principal component analysis (PCA) and singular value decomposition (SVD). Taking PCA as an example, the covariance matrix of the performance matrix is ​​first calculated to determine the correlations between the various parameters. The eigenvalues ​​and eigenvectors of the covariance matrix are then calculated and sorted by eigenvalue size. The eigenvectors corresponding to the top k largest eigenvalues ​​are selected as principal components. These principal components can explain most of the variance in the original data (e.g., cumulative contributions exceeding 85%). By projecting the original data into the principal component space, data dimensionality reduction is achieved while retaining the main performance trends. For example, in the three-dimensional parameter space of temperature, pressure, and flow, PCA may extract a comprehensive eigenvector reflecting combustion efficiency. Its direction represents the trend of coordinated changes in temperature, pressure, and flow, and the eigenvalue size reflects the significance of this trend.

[0035] When calculating the spatial correlation characteristics of the performance matrix, a correlation coefficient matrix for the sensor parameters is first constructed. The elements in the matrix represent the Pearson correlation coefficient between any two sensor parameters, reflecting the degree of linear correlation between parameter changes. The correlation coefficient matrix is ​​used to identify strongly correlated sensor pairs (e.g., with an absolute value of the correlation coefficient greater than 0.8). The physical locations corresponding to these sensor pairs typically have direct thermal conduction, fluid coupling, or control relationships. Based on the correlation coefficient matrix, the interference intensity, stability coefficient, and anomaly index between units are calculated: the interference intensity is represented by the sum of the absolute values ​​of the correlation coefficients of the sensor parameters of adjacent subunits; a larger value indicates a stronger mutual influence between subunits; the stability coefficient is calculated as the inverse of the variance of the sensor parameters within a single subunit; a smaller variance indicates a higher stability coefficient, indicating a more stable operating state for the subunit; the anomaly index is calculated by the degree of deviation of the sensor parameters from the historical mean (e.g., absolute deviation or relative deviation), reflecting whether the parameters are outside the normal operating range.

[0036] When constructing a feature fusion network to calculate pressure fluctuation characteristics, deep learning architectures such as convolutional neural networks (CNNs) or graph neural networks (GNNs) are used. For two-dimensional performance matrices (such as the temperature distribution matrix of a furnace cross section), convolutional neural networks extract spatial local features through multiple layers of convolution kernels. For example, a 3×3 convolution kernel is used to capture the pressure fluctuation pattern between adjacent sensors. Pooling layers are then used to reduce the feature dimension and achieve multi-scale feature fusion. For three-dimensional spatiotemporal matrices (multi-frame performance matrices containing time series), a temporal convolutional network (TCN) or a long short-term memory network (LSTM) combined with spatial convolution can be used to simultaneously capture the spatial distribution characteristics and temporal evolution of pressure fluctuations. The input of the feature fusion network is a standardized pressure parameter matrix and spatial correlation features, and the output is a fused pressure fluctuation feature vector, which contains information such as the pressure coupling relationship at different spatial locations, the fluctuation frequency, and the propagation direction.

[0037] When extracting the time and frequency domain features of each sensor, time domain analysis focuses on the statistical characteristics of the parameters in the time dimension, including mean, variance, peak value, rise time, and fall time. For example, the time domain features of a temperature sensor can reflect the average temperature level, fluctuation amplitude, and sudden changes at that location. Frequency domain analysis converts the time domain signal into the frequency domain using a fast Fourier transform (FFT), extracting features such as the primary frequency, secondary frequency, and harmonic components to identify periodic interference or vibration sources. For example, if the frequency domain features of a pressure signal show a high-amplitude component at a fixed frequency, this may indicate mechanical vibration or fluid resonance. Calculating the phase difference between the time and frequency domain features generates a running feature vector containing spatiotemporal information. This vector, through the phase difference, reflects the time lag or lead relationship of parameter changes. For example, the phase difference between the peak value of a sensor's pressure fluctuation in the time domain and the primary frequency component in the frequency domain can reveal the temporal correlation between that fluctuation and fluctuations from other sensors.

[0038] When performing feature matching for sensors at different locations based on their operating feature vectors, metrics such as cosine similarity and Euclidean distance are used to calculate the similarity between vectors. Sensors with high similarity typically correspond to physical locations with similar operating states or coupling relationships. For example, pressure sensors upstream and downstream of the same pipeline should have highly similar operating feature vectors. Feature matching allows sensors to be clustered according to their operating modes, identifying abnormal sensors (such as nodes whose feature vectors differ significantly from similar sensors), while also providing a basis for analyzing performance trends. For example, if the operating feature vectors of multiple sensors in a certain area show a synchronous increase in phase difference, this may indicate abnormal airflow organization or uneven heat load distribution in that area, necessitating further analysis of performance trends.

[0039] The calculation of performance trends comprehensively considers the results of eigendecomposition, spatial correlation characteristics, and operational eigenvectors. The principal component score sequence obtained through principal component analysis can reflect the overall trend of performance changes. For example, an increase in principal component scores indicates that the unit is evolving toward a high-load or high-efficiency state. Spatial correlation characteristics can identify the dominant regions of trend change. For example, an increase in interference intensity in a subunit may be the source of the performance trend change. The temporal changes in the operational eigenvectors can track the propagation path and speed of the trend change. For example, the dynamic changes in phase difference reflect the propagation process of fluctuations in the sensor network. Combining all this information forms a multidimensional description of performance trends, providing a foundation for operational state modeling and performance prediction.

[0040] The entire performance matrix processing process runs through the entire process of data cleaning, feature extraction, dimensionality reduction and fusion. It eliminates dimensional differences through standardization, improves data quality through filtering and feature decomposition, mines implicit correlations between parameters through spatial correlation analysis and feature fusion networks, and captures the dynamic characteristics of the operating status through time-frequency domain feature analysis.

[0041] Example 3: This embodiment focuses on operating status modeling. Through the steps of correlation mapping of temperature and pressure characteristics, generating operating status maps, spatial alignment, abnormality source location, and model calibration, dynamic modeling of the unit operating status and abnormality identification are achieved. The specific implementation method is as follows: The basic information of the operating status map is extracted based on the temperature distribution characteristics. In each frame of performance matrix data, the sampling points of the temperature parameters are screened out. These sampling points correspond to the temperature values ​​at the sensor deployment locations, forming discrete temperature field data. The temperature sampling points are associated with the pressure fluctuation characteristics, that is, the temperature and pressure data of the same sensor are aligned in space to form operating status data points containing temperature-pressure two-dimensional parameters. The discrete data points are filled by interpolation algorithms (such as bilinear interpolation) to generate a continuous operating status map. The map displays the temperature field and pressure field distribution of the unit at a certain moment in the form of two-dimensional or three-dimensional graphics. For example, the temperature cloud map of the furnace cross section is superimposed with the pressure contour line, which intuitively presents the spatial coupling relationship between the high-temperature area and the pressure fluctuation area.

[0042] Perform spatial alignment on the operating status maps collected by multiple sensors. Since there are three-dimensional spatial differences in the deployment positions of different sensors (such as the coordinate offset of the sensors at the top and bottom of the furnace), a coordinate transformation algorithm is required to eliminate the spatial deviation. The specific operation includes: first determining a global reference coordinate system (such as the center of the unit furnace as the origin, the horizontal direction as the x-axis, the vertical direction as the y-axis, and the depth direction as the z-axis), obtaining the three-dimensional coordinates of each sensor ( ); Then, an affine transformation is performed on the operating status map generated by each sensor. Through translation, rotation, and scaling operations, the map coordinates are converted to coordinates in the global reference coordinate system, ensuring that the maps of different sensors are fully aligned in space. After spatial alignment, the map data of each sensor can be fused to form a complete unit operating status distribution. For example, the temperature maps of multiple furnace wall sensors can be fused into the overall furnace temperature field distribution, facilitating the analysis of spatial temperature continuity and gradient changes.

[0043] After generating the operating status distribution, abnormal source location and control model constraint compensation are required. First, an interference threshold is set. This threshold is determined based on statistical analysis of historical operating data. It is typically the average value of the inter-unit interference intensity under normal operating conditions plus a multiple of the standard deviation (e.g., mean + 3σ). This is used to distinguish normal fluctuations from abnormal interference. A time series analysis is performed on the temperature distribution values ​​of the multi-frame performance matrix. A sliding window method is used to calculate the mean, variance, and trend slope of the temperature change for each subunit in N consecutive frames of data. If the temperature change of a subunit exceeds the historical normal range (e.g., the temperature rise rate exceeds 5°C / min for three consecutive frames and the current temperature is higher than 110% of the rated value), the subunit is preliminarily identified as a candidate abnormal area.

[0044] The interference intensity difference between the abnormal candidate area and the adjacent sub-unit is further calculated. The interference intensity is calculated by the mutual correlation coefficient of the sensor parameters of the adjacent sub-units, and the formula is:

[0045] in, Represents a subunit and The interference intensity between For subunits At the moment The temperature value, For subunits The mean temperature, is the number of sampling cycles. Interference intensity difference ,in is the mean value of interference intensity under normal working conditions. If the value is greater than or equal to the interference threshold, the subunit is determined There are operational anomalies and control model constraint compensation is required.

[0046] The control model constraint compensation is based on the thermodynamic model of the subunit. The thermodynamic model usually includes heat transfer equations, fluid flow equations, etc. For example, for the furnace combustion area, the model can be expressed as:

[0047] in, is the temperature, For time, is the fluid velocity vector, is the thermal conductivity, is the density, is the specific heat capacity at constant pressure, is the intensity of the heat source. When an anomaly occurs, constraints are introduced into the model (such as limiting the upper temperature limit or pressure fluctuation range), and model parameters (such as heat source intensity) are adjusted through iterative calculations. or fluid velocity ), returning model output parameters such as temperature and pressure to normal ranges. After each iteration, the error between the model's predicted value and the actual sensor measurement is calculated. If the error exceeds a preset threshold (e.g., a temperature error exceeding 2°C), the parameters are adjusted until the error converges. This iterative correction generates an intensity compensation value for the abnormal cell. This value is used to adjust the control parameters of the coordinated control system, such as increasing air volume in the abnormal area to reduce temperature or adjusting the fuel injection angle to improve combustion distribution.

[0048] After locating the anomaly source and calibrating the model, the performance matrix is ​​modeled based on the operating state distribution. During modeling, the unit is divided into several physically meaningful operating units (such as the burner area, superheater area, economizer area, etc.), each corresponding to a set of sensor data and a thermodynamic model. By analyzing the temperature and pressure field distribution characteristics of each unit, an operating state model for the unit is established. Model parameters include temperature mean, pressure fluctuation amplitude, heat transfer rate, etc. For example, the operating state model of the burner area can be described as follows: the temperature reaches a peak at the center of the flame and is Gaussian distributed in all directions, and the pressure fluctuation is positively correlated with the fuel injection frequency.

[0049] The unit's operating model is marked with fluctuations based on the pressure fluctuation characteristics. Pressure fluctuation characteristics are extracted through time domain analysis (such as root mean square value, peak factor) and frequency domain analysis (such as main frequency, frequency harmonic components). For example, if the root mean square value of the pressure signal of a unit increases significantly, it indicates that the unit has strong pressure fluctuations; the frequency domain analysis shows that the main frequency is consistent with the burner vibration frequency, which may indicate the existence of combustion instability problems. According to the intensity and frequency of the pressure fluctuations, the fluctuation level (such as mild, moderate, severe) and fluctuation type (such as periodic fluctuation, random fluctuation) are marked in the operating model to provide clear status identification for the subsequent generation of performance evaluation indicators and the adjustment of coordinated control strategies. For example, units marked as "severe periodic fluctuations" need to prioritize combustion parameter adjustments to avoid fluctuations causing equipment failures or performance degradation.

[0050] The entire operating status modeling process achieves precise characterization of the unit's operating status and anomaly response by integrating spatial correlation of temperature and pressure data, anomaly detection algorithms, and thermodynamic models. Spatial alignment ensures consistency of multi-source data, interference intensity calculation and iterative correction of the thermodynamic model improve the accuracy of anomaly location and compensation, and fluctuation annotation provides an intuitive basis for visualizing operating status and formulating control strategies.

[0051] Example 4: Through a hierarchical design of the input layer, feature fusion layer, and resource allocation layer, spatial correlation modeling of performance prediction information and control strategy generation are achieved. The following is a detailed description based on specific application scenarios (such as performance evaluation of the furnace combustion area of ​​a high-alkali coal unit): The input layer of the comprehensive evaluation model receives and standardizes performance prediction information. This performance prediction information comes from the predicted values ​​of each subunit's parameters for future time periods generated in the previous step, such as the temperature and pressure forecasts for different burner zones (subunits A, B, and C) in the furnace. The input layer first organizes these predicted values ​​into spatially distributed data, specifically in the form of a two-dimensional matrix or a three-dimensional tensor. If the unit is divided into N×M grid cells based on the furnace cross-section, each grid cell corresponds to a subunit, and the matrix elements are the predicted temperature values ​​for that subunit. If the time dimension is considered, a three-dimensional tensor consisting of T time steps can be constructed to reflect the temporal evolution of each subunit's parameters.

[0052] During the standardization process, the input layer performs statistical analysis on the historical data of each sub-unit, calculates the mean and standard deviation, and converts the current predicted value into a standard score (Z-score). The formula is:

[0053] For example, if the historical mean temperature of subunit A is 1200°C with a standard deviation of 50°C, and the predicted value is 1280°C, the normalized value is (1280-1200) / 50=1.6, indicating that the predicted temperature in this area is 1.6 standard deviations higher than the historical mean, suggesting a potential overheating risk. Standardization makes parameters of different dimensions and fluctuation ranges (such as temperature, pressure, and flow) comparable, ensuring the stability of model training.

[0054] The feature fusion layer utilizes a graph neural network (GNN) architecture, designed to extract unit-related performance features and construct dependencies between physical units. For example, the furnace combustion area contains multiple burner subunits, each of which influences each other through physical processes such as flue gas flow and thermal radiation. The GNN abstracts each subunit as a node in a graph structure. Node features are standardized predicted parameters for that subunit (such as temperature, pressure, and fuel quantity). Edges between nodes represent the spatial proximity of subunits or the strength of physical coupling (such as the heat transfer coefficient between adjacent burners).

[0055] The computational process of the feature fusion layer is divided into two phases: message passing and node updating. During the message passing phase, each node sends its own feature information to adjacent nodes. The weight of the message passing is determined by the physical coupling strength of the edges. For example, if burner subunits A and B are adjacent and the heat transfer coefficient between them is high, subunit A will give subunit B a greater weight when passing messages. The message passing formula can be expressed as:

[0056] in, is the total number of messages received by node j, is the set of adjacent nodes of node j, is the edge weight between nodes i and j, is the feature vector of node i. In the node update phase, node j fuses its original features with the received message through an activation function (such as ReLU) to generate a new feature vector:

[0057] in, is the weight matrix, is the bias vector, is the activation function. Through multi-layer message passing and node updates, the feature fusion layer can capture indirect influences between distant subunits (such as the effect of adjusting the fuel quantity of a subunit in the burner area on the furnace outlet temperature), constructing a complex unit dependency network.

[0058] Taking furnace slagging prediction as an example, the feature fusion layer can learn from historical data to discover a delayed correlation between temperature increases in a burner region and pressure fluctuations in adjacent regions. This allows the model to establish a causal chain: "temperature increase → heat flow change → pressure fluctuations in adjacent regions." Extracting these correlation features helps to proactively identify the risk of global performance degradation caused by local parameter anomalies.

[0059] The core function of the resource allocation layer is to integrate the spatial correlations of performance indicators and generate control parameter scheduling strategies. This layer takes the node feature vectors output by the previous layer as input and, through a fully connected neural network or attention mechanism, calculates the control priority and resource demand weights of each subunit. For example, in a furnace combustion optimization scenario, the resource allocation layer analyzes the predicted temperature values ​​and associated characteristics of each burner subunit, identifies subunits with high overtemperature risk and significant impact on adjacent areas (such as subunit A), and assigns them higher control priority.

[0060] The generation of control parameter scheduling strategy needs to consider the physical constraints and optimization objectives of the unit operation. Physical constraints include valve opening range, fuel supply rate upper limit, equipment safety temperature threshold, etc.; optimization objectives can be set according to the current operating conditions, such as improving combustion efficiency, reducing NOx emissions, and inhibiting slagging. Taking fuel supply optimization as an example, the resource allocation layer generates a flow allocation vector based on the fuel demand weight of each subunit. ,in Represents the fuel supply of subunit i, which satisfies the requirement that the total fuel quantity is equal to the load demand of the unit and each The maximum supply rate of the equipment is not exceeded. At the same time, the fuel distribution ratio of adjacent sub-units is adjusted through the attention mechanism, such as reducing the fuel amount in the over-temperature area and increasing the fuel amount in its adjacent areas to balance the combustion distribution and reduce the local heat load.

[0061] In practical applications, the outputs of the resource allocation layer can be directly mapped to the actuators of the coordinated control system. For example, the flow distribution vector can be converted into opening instructions for each fuel control valve and sent to the on-site actuators via the DCS (distributed control system). For scenarios requiring coordinated adjustment of multiple parameters (such as the ratio of fuel to air volume), the resource allocation layer can generate a multidimensional control parameter vector, including fuel quantity, primary air volume, secondary air damper opening, and other parameters, ensuring that each parameter is adjusted in conjunction with the preset control logic.

[0062] The three-layer architecture of the comprehensive evaluation model forms a complete link from data input to strategy output: the input layer unifies the data format through standardization, the feature fusion layer mines the physical connections between units through graph structure modeling, and the resource allocation layer generates executable control strategies based on optimization objectives. Taking the typical operating conditions of a high-alkali coal unit as an example, when the temperature in a burner area abnormally rises due to changes in coal quality, the input layer detects a significant increase in the standardized temperature value of this area; the feature fusion layer, through graph neural network analysis, finds that this area has a strong correlation with the upstream coal blending unit and the downstream heating surface unit, and the abnormal temperature may cause fluctuations in the upstream coal feed or the risk of slagging downstream; the resource allocation layer generates control strategies based on this, including reducing the fuel supply in this area, increasing the secondary air ratio to enhance combustion, and adjusting the coal blending ratio of the upstream coal blending unit. Through multi-dimensional control actions, the abnormal state is suppressed.

[0063] Example 5: Through the mapping of unit identification and performance indicators, execution action triggering, resource allocation and dynamic parameter adjustment, real-time optimization of the unit's operating status is achieved. The following is a detailed description of the typical control scenarios of the steam-water system of a high-alkali coal unit (such as main steam temperature control and fuel-air ratio adjustment): Establish a one-to-one mapping relationship between unit identifiers and performance evaluation indicator areas. Taking the unit's steam-water system as an example, the superheater is divided into three sub-units: low-temperature superheater, platen superheater, and high-temperature superheater, and assigned unit identifiers S1, S2, and S3, respectively. Each unit corresponds to a set of performance evaluation indicators. For example, the indicators for S1 include tube wall temperature, steam flow rate, and desuperheating water opening. The evaluation server calculates the performance indicators of each unit in real time and generates spatiotemporal distribution characteristics that reflect its operating status, such as the tube wall temperature field distribution cloud map and steam pressure fluctuation curve in the S2 area. By binding the unit identifier to the indicator area, it ensures that the control action can be accurately located to the physical unit. For example, when the temperature indicator in the S2 area exceeds the preset threshold, the system automatically identifies the control node corresponding to the area (such as the desuperheating water regulating valve of the platen superheater).

[0064] Coordinated control actions are triggered based on the spatiotemporal distribution of performance indicators. For example, when the predicted outlet temperature of the high-temperature superheater unit (S3) approaches the rated upper limit (e.g., 540°C), the system first analyzes the temperature distribution. If the high-temperature region is concentrated in the front half of S3 (near the platen superheater outlet), the system determines that the heat exchange intensity in this area is too high, requiring cooling by adjusting the attemperating water flow rate. At this point, the system generates opening adjustment commands for adjacent valves, increasing the opening of the attemperating water control valve corresponding to the front half of S3 and increasing the water flow rate to absorb the heat of the superheated steam. During the adjustment process, the temperature change rate in the S3 region is monitored in real time. If the temperature decline trend does not meet expectations, the attemperating water valves in adjacent areas (e.g., the rear half of S2) are further adjusted, expanding the cooling range through multi-valve coordinated adjustment.

[0065] The dynamic adjustment of fuel supply and flow distribution is based on the preset fuel supply strategy. For example, when the unit load suddenly increases by 10%, the system needs to quickly increase the amount of fuel to maintain the steam pressure. At this time, the dynamic adjustment module allocates fuel to units with high combustion efficiency and low slagging risk based on the performance indicators of each burner unit (such as combustion efficiency and slagging risk index). Taking the four-corner tangential combustion furnace as an example, if the slagging risk index of the east burner unit (B1) is lower than that of the west unit (B2), then when the fuel is increased, the fuel supply increase of B1 is 10%-15% higher than that of B2. At the same time, a flow distribution vector is generated, which contains parameters such as the fuel amount, primary air volume, secondary air volume, etc. of each burner unit, for example The fuel-air ratio must meet the chemical equivalence ratio to ensure adequate combustion. Fuel supply is optimized by dynamically combining available resources (e.g., high-alkali coal and blended coal from different coal bunkers). For example, under high-load conditions, blended coal with higher calorific value is prioritized to reduce fuel consumption.

[0066] When adjusting the actuator parameters of the target unit's control node based on the flow distribution vector, the principle of "gradual response, near first, then far" must be followed. Taking furnace combustion adjustment as an example, when the performance indicators of a burner unit (B3) indicate insufficient combustion (e.g., CO concentration exceeds the standard), the system first adjusts the opening of the unit's secondary air damper to increase the oxygen supply to enhance combustion. If the indicators do not improve after adjustment, the fuel and air volume of adjacent units (B2 and B4) are further adjusted to enhance the disturbance by changing the airflow structure within the furnace. The amplitude and rate of actuator parameter adjustment are determined by the degree of deviation from the performance indicators. For example, if the CO concentration exceeds the standard by 10%, the secondary air damper opening is increased by 5%. If it exceeds the standard by 20%, in addition to increasing the air volume, the fuel supply to the unit is reduced by 5%, and the auxiliary burner of the adjacent unit is activated.

[0067] Control parameters are dynamically assigned to physical units based on the spatial distribution of performance indicators and preset adjustment strategies. For example, when the flue gas temperature at the furnace outlet is unevenly distributed (the left side is 30°C higher than the right), the preset strategy is to "adjust the fuel quantity and swirl intensity of the left burner to reduce the local heat load." Based on the flue gas temperature distribution characteristics, the system calculates the temperature deviation of each burner unit on the left and allocates fuel adjustments based on the size of the deviation: the unit with the largest deviation (B5) has its fuel supply reduced by 8%, and the adjacent unit (B6) by 5%. At the same time, the secondary air swirl intensity of B5 is increased to diffuse the flame and reduce the local temperature peak. After the adjustment, the flue gas temperature distribution is monitored in real time. If the deviation is reduced to within 10°C, the current parameters are maintained. If it does not meet expectations, the fuel quantity of the right burner is further adjusted. The flue gas temperature distribution is balanced through a two-way adjustment of "reducing the quantity on the left and increasing the quantity on the right."

[0068] Under complex operating conditions (such as changes in the high-alkali coal blend ratio and unit startup and shutdown phases), the dynamic adjustment module must integrate multi-dimensional performance indicators to generate a complex control strategy. For example, when the high-alkali coal blend ratio increases from 30% to 50%, the system anticipates an increased risk of slagging and proactively adjusts the following parameters: ① The primary air velocity of all burner units is increased by 5% to reduce coal dust deposition near the nozzles; ② The frequency of soot blowers in the superheater area is increased from every 8 hours to every 4 hours; and ③ The oxygen control target value for each unit is increased by 1% to enhance the oxidizing atmosphere and suppress alkali metal volatilization. These control actions are tied to physical locations through unit identifiers. For example, soot blower actions correspond to specific soot blowing points in superheater units S2 and S3, and oxygen adjustments correspond to the secondary air control loops of each burner unit.

[0069] The entire dynamic adjustment process forms a closed-loop feedback mechanism: sensors collect parameters in real time → the evaluation server generates performance indicators → the dynamic adjustment module triggers a control action → the actuator adjusts the parameters → the sensor collects parameters again, and the cycle repeats until the performance indicators return to the target range. Taking feedwater pump flow control as an example, when feedwater flow fluctuations cause the drum water level to deviate from the set value by ±50mm, the system first adjusts the speed of the main feedwater pump. If the water level fluctuations persist, it switches to the bypass control valve for adjustment. At the same time, it analyzes related indicators such as the upstream water tank level and downstream steam flow to determine whether there are underlying causes such as changes in pipeline resistance or sudden load changes. It then adjusts the number of operating feedwater pumps or coordinates the boiler combustion rate with the turbine throttle opening to achieve stable water level control.

[0070] For drum-less, once-through boiler feedwater pump flow control in boilers over 600 MW, multiple sensors are deployed at key locations, such as the economizer inlet, water wall inlet and outlet, and superheater sections. Adjacent sensors are spaced at a predetermined distance to collect real-time operating parameter data, including feedwater flow, working fluid temperature, pressure, and enthalpy in each section, to generate a performance dataset. An evaluation server receives real-time data from each sensor and constructs a performance matrix reflecting the overall operating status of the unit.

[0071] The performance matrix is ​​standardized, and hotspots such as abnormal flow fluctuations are captured using a sliding window for noise filtering. A feature decomposition algorithm is then used to extract flow rate trends and temperature distribution characteristics. Operating state gradients, such as the temperature gradient along the working fluid flow direction and the spatial propagation rate of flow fluctuations, are calculated based on sensor locations. The impact of feedwater flow changes on performance indicators such as superheated steam temperature and pressure is predicted in combination with control parameter information. The performance dataset is updated by comprehensively evaluating the spatiotemporal distribution characteristics of the model's output performance indicators.

[0072] When the comprehensive evaluation model shows that the superheated steam temperature deviates from the set value, and the analysis of the spatiotemporal distribution characteristics determines that it is caused by insufficient feedwater flow, the dynamic adjustment module maps the unit identifier to the performance evaluation indicator area to determine the feedwater pump and related control nodes that need to be adjusted. The feedwater pump speed adjustment instruction is triggered, and the fuel quantity is adjusted according to the fuel supply strategy to maintain an appropriate water-coal ratio. For example, if the superheated steam temperature rises, the feedwater pump speed is increased to increase the feedwater flow, and the fuel quantity is proportionally reduced to return the water-coal ratio to the optimized value. Based on the spatial distribution of performance indicators, the control parameters are dynamically allocated to the corresponding physical units, such as adjusting the opening of the economizer inlet feedwater regulating valve to ensure accurate adjustment of the feedwater flow, realize coordinated control of the direct current boiler feedwater system, and ensure safe, stable and economical operation of the unit.

[0073] This implementation ensures the targeted and effective coordinated control of units through precise mapping of unit identification and physical location, coordinated adjustment of multiple actuators, and dynamic generation of composite strategies. Whether local adjustment of a single parameter (such as a single valve opening) or global optimization of multiple systems (such as the linkage of fuel, air, and water), both are based on the spatiotemporal distribution of performance indicators, avoiding the lag and blindness of traditional control strategies. Furthermore, the combination of preset adjustment strategies and real-time data analysis enables the system to rapidly respond to the complex operating conditions of high-alkali coal units, providing a reliable control method for the safe and stable operation of units under a wide load range and variable coal quality conditions.

[0074] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0075] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A comprehensive performance evaluation method for a high-alkali coal unit coordinated control system is applied to the coordinated control system of a high-alkali coal unit. The system includes multiple sensors deployed at key locations of the unit and an evaluation server connected to the sensors. Two adjacent sensors are separated by a set distance. The method is characterized by: The method comprises: Collecting the unit's operating parameter data through the sensor to generate a performance data set; Receive real-time operating data of multiple sensors through the evaluation server to build a unit performance matrix; Determining a performance evaluation indicator based on the performance data set and the real-time data of each sensor, wherein determining the performance evaluation indicator includes: processing a performance matrix, extracting features based on the performance data set, and performing performance prediction based on an operating state gradient and control parameter information, outputting spatiotemporal distribution characteristics of the performance indicator through a comprehensive evaluation model, and updating the performance data set based on the spatiotemporal distribution characteristics; According to the evaluation indicators, the coordinated control of the unit is dynamically adjusted.

2. The method for comprehensive performance evaluation of the coordinated control system of a high-alkali coal unit according to claim 1 is characterized in that: Determining the performance evaluation index includes: Processing the performance matrix to extract temperature distribution characteristics, pressure fluctuation characteristics, and performance change trends; Performing an operating state model on the performance matrix according to the temperature distribution characteristics and the pressure fluctuation characteristics, dividing the unit into a plurality of subunits and marking the unit identifiers, performing an association match with the performance data set according to the temperature distribution of the subunits, and marking the unit identifiers in the performance data set; Calculating an operating state gradient according to the sensor position, predicting the distribution of performance indicators according to the operating state gradient and the performance change trend, and calculating performance prediction information for each subunit; Constructing a comprehensive evaluation model, taking the performance prediction information as an input parameter of the comprehensive evaluation model, performing spatial correlation modeling on the performance prediction information through the comprehensive evaluation model, and outputting spatiotemporal distribution characteristics of performance indicators; The performance data set is updated according to the spatiotemporal distribution characteristics to obtain performance evaluation indicators.

3. The method for comprehensive performance evaluation of the coordinated control system of a high-alkali coal unit according to claim 2 is characterized in that: The processing of the performance matrix includes: Normalize the performance matrix, intercept the performance hotspot areas in the matrix through a sliding window, filter the noise in the hotspot areas, and calculate the performance change trend through the eigendecomposition algorithm; Calculating spatial correlation features of the performance matrix, calculating interference intensity, stability coefficient, and anomaly index between units based on the spatial correlation features, constructing a feature fusion network, and calculating pressure fluctuation features through the feature fusion network; The time domain features and frequency domain features collected by each sensor are extracted, and the sensor's operating feature vector is calculated based on the phase difference between the time domain features and the frequency domain features. Based on the operating feature vector, feature matching is performed on sensors at different positions, and the performance change trend is calculated.

4. The method for comprehensive performance evaluation of the coordinated control system of a high-alkali coal unit according to claim 3 is characterized in that: The operating state modeling of the performance matrix includes: According to the temperature distribution characteristics, temperature sampling points are extracted from each frame of data, and the sampling points are correlated and mapped with the pressure fluctuation characteristics to generate an operating state map. The operating state maps collected by multiple sensors are spatially aligned to calculate the operating state distribution of the unit; Set the interference threshold, locate the abnormal source based on the temperature distribution value of the multi-frame performance matrix, and calculate the interference intensity difference. If the interference intensity difference is greater than or equal to the interference threshold, it indicates that there is an operational abnormality in the unit. The control model constraint compensation is performed on the current unit. According to the thermodynamic model corresponding to the current unit, the operating state distribution of the current unit is iteratively corrected. The intensity compensation value of the abnormal unit is calculated based on the correction result. The operating state modeling is performed on the performance matrix according to the operating state distribution, and the fluctuation of the operating model of the unit is marked according to the pressure fluctuation characteristics.

5. The method for comprehensive performance evaluation of the coordinated control system of a high-alkali coal unit according to claim 4 is characterized in that: The step of calculating the operating state gradient according to the sensor position includes: Based on multiple sets of operating status data, temperature change points are extracted and mapped to a unified coordinate system based on the deployment location of the sensors. The change points are fitted using a spatial interpolation algorithm to generate the unit's operating field model. Performing sampling at equal intervals along the transmission path of the operating field model, calculating the thermal conductivity, fluctuation index, and state change slope of the path based on the sampling results, and calculating the state change parameter based on the thermal conductivity, fluctuation index, and state change slope; Based on the sensor deployment parameters and acquisition accuracy, the distribution characteristics of performance changes in each frame of data are projected onto the operational field model. The operational field model is partitioned along the transmission direction according to the number of sensors. The variation patterns of performance changes within the partitions are analyzed, and the performance distribution characteristics are calculated based on these variation patterns. An operating state gradient is calculated based on the state change parameter and the performance distribution characteristics. The calculation process of the operating state gradient includes: based on the position range from the first sensor to the last sensor, selecting spatial coordinate points in the sensor deployment direction, cumulatively calculating the product of the operating field strength characteristic weight value and the performance distribution characteristic weight value within the spatial resolution range, and superimposing the influence value of the sensor acquisition frequency on the operating state change rate.

6. The method for comprehensive performance evaluation of a coordinated control system for a high-alkali coal unit according to claim 5, characterized in that: The calculating of the performance prediction information of each subunit includes: Using the main transmission path of the operating field model as a baseline, taking the peak position of the performance change in each frame of data as a reference point, calculating the performance offset, and drawing a performance distribution curve according to the coordinates; Correcting the growth rate and direction of the performance change trend according to the operating state gradient; Starting from the most recent performance distribution point, the distribution curve is continuously drawn based on the correction results of the growth rate and direction to generate the performance distribution points for the next period until the distribution points cover the entire target unit and generate performance prediction information.

7. The method for comprehensive performance evaluation of a coordinated control system for a high-alkali coal unit according to claim 2, characterized in that: The construction of the comprehensive evaluation model includes: The input layer is used to organize the performance prediction information into spatially distributed data and perform normalization processing; The feature fusion layer is used to extract unit-related features of performance by processing spatially distributed data and build dependency relationships between physical units; The resource allocation layer is used to integrate the correlation between performance indicators in spatial units and generate control parameter scheduling strategies.

8. The method for comprehensive performance evaluation of a coordinated control system for a high-alkali coal unit according to claim 2, characterized in that: The obtaining of performance evaluation indicators includes: According to the spatiotemporal distribution characteristics of the performance indicators output by the comprehensive evaluation model, the identification of the sub-unit is matched with the spatiotemporal distribution characteristics; The unit data in the performance data set are reorganized according to the spatiotemporal characteristics to generate a unit distribution map sorted by the strength of the performance indicators; According to the reorganized unit distribution map, the optimized performance evaluation index is output.

9. The method for comprehensive performance evaluation of a coordinated control system for a high-alkali coal unit according to claim 1, characterized in that: The dynamic adjustment of the coordinated control of the units includes: Map unit identifiers to areas of performance evaluation indicators one by one; According to the spatiotemporal distribution characteristics of performance indicators, the execution actions of coordinated control are controlled, including valve opening adjustment, fuel supply and flow distribution operations; Dynamically allocate control parameters to corresponding physical units based on the spatial distribution of performance indicators and preset adjustment strategies; The execution actions of the control coordination control include: When the performance index reaches a preset intensity threshold in the target unit, an opening adjustment instruction of the adjacent valve is triggered; Dynamically combine available resources according to the fuel supply strategy to generate a flow allocation vector; An actuator parameter of a target unit control node is adjusted based on the traffic distribution vector.

10. A system for comprehensively evaluating the performance of a coordinated control system for a high-alkali coal-fired power plant, applied to the method for comprehensively evaluating the performance of a coordinated control system for a high-alkali coal-fired power plant according to any one of claims 1 to 9, the system comprising a plurality of sensors deployed at key locations of the power plant, with adjacent sensors spaced a set distance apart, and an evaluation server connected to the sensors, characterized in that: The multiple sensors are used to collect unit operating parameter data, which include component data characterizing the metal content of high-alkali coal, temperature and pressure data characterizing combustion characteristics, and coking thickness data characterizing coking conditions, thereby generating a performance data set; the evaluation server is used to receive real-time operating data from multiple sensors and construct a unit performance matrix; the evaluation server is also provided with a performance evaluation index determination module for processing the performance matrix and performing feature extraction in combination with the performance data set, the feature extraction including extracting the distribution characteristics of metal elements in high-alkali coal, the abnormal temperature and pressure fluctuation characteristics caused by the characteristics of high-alkali coal during combustion, and the change characteristics of the degree of coking of high-alkali coal, and performing performance prediction based on the operating state gradient and control parameter information, outputting the spatiotemporal distribution characteristics of the performance indicators through a comprehensive evaluation model, and updating the performance data set based on the spatiotemporal distribution characteristics; the system is also provided with a dynamic adjustment module for dynamically adjusting the unit coordinated control according to the evaluation indicators, the dynamic adjustment including adjusting the fuel ratio according to the metal content of the high-alkali coal, optimizing the burner parameters according to the combustion characteristics, and controlling the soot blowing frequency and intensity according to the coking conditions.

Images