Adaptive modeling methods, devices, equipment, and storage media for thermal power plants

Abstract

This invention relates to the field of thermal power generation modeling technology, specifically disclosing a method, apparatus, equipment, and storage medium for adaptive modeling of thermal power plants. The method includes acquiring multiple key control parameters of the target unit; optimizing the key control parameters to obtain the target control parameters corresponding to the target unit; acquiring the optimization target of the target unit; dividing the target unit into multiple target operating conditions based on the target control parameters; determining multiple influencing characteristic variables under each target operating condition; normalizing the multiple influencing characteristic variables corresponding to each target operating condition within the [1,2] interval, performing high-order modeling, and then verifying the results to obtain a verified adaptive model of the thermal power plant. This scheme, by adapting the optimization target to divide the operating conditions, can obtain target unit models that match different optimization targets, exhibiting adaptability to thermal power plants with multiple target units, improving model adaptability, and enhancing model accuracy through optimization of key control parameters.

Description

Technical Field

[0001] This invention belongs to the field of thermal power generation modeling technology, specifically relating to adaptive modeling methods, devices, equipment, and storage media for thermal power plants. Background Technology

[0002] A thermal power plant, or coal-fired power plant for short, is a factory that uses combustible materials (such as coal) as fuel to produce electricity. Its basic production process is as follows: when fuel is burned, it heats water to generate steam, converting the chemical energy of the fuel into heat energy. The steam pressure drives the turbine to rotate, converting the heat energy into mechanical energy. Then, the turbine drives the generator to rotate, converting the mechanical energy into electrical energy.

[0003] In the thermal power generation sector, efficient and stable unit operation is crucial for energy utilization and power supply. With the increasing demands for energy conservation, emission reduction, and refined management in the power industry, establishing models adapted to unit characteristics is becoming a key direction for optimizing production processes and control strategies, and is thus essential for the industry's development. Summary of the Invention

[0004] To address the shortcomings of existing technologies, according to one aspect of this application, an adaptive modeling method for thermal power plants is disclosed, the method comprising:

[0005] Acquire multiple key control parameters of the target unit;

[0006] The key control parameters are optimized to obtain the target control parameters corresponding to the target unit;

[0007] Obtain the optimization objective of the target unit;

[0008] The target unit is divided into operating conditions based on the target control parameters to obtain multiple target operating conditions that meet the optimization objectives.

[0009] The optimization objective is determined under each target operating condition by identifying multiple influencing characteristic variables, wherein there is no coupling relationship between the multiple influencing characteristic variables;

[0010] For each target working condition, multiple characteristic variables are normalized to a range of 1-2.

[0011] After performing high-order modeling and validation on the normalized feature variables, the validated adaptive model of the thermal power plant is obtained; wherein, the adaptive model of the thermal power plant is characterized by the following formula:

[0012] ;

[0013] In the formula, To optimize the indicators;

[0014] The total number of feature variables corresponding to an optimization metric;

[0015] It is the highest order polynomial;

[0016] Single feature of Order coefficient;

[0017] Features , Interactive items , Order coefficient;

[0018] This is a constant term.

[0019] In some embodiments, optimizing the key control parameters to obtain the target control parameters corresponding to the target unit includes:

[0020] Identify downtime events, dead pixel events, and bottleneck events among the key control parameters;

[0021] Obtain the non-steady-state data of shutdown, non-steady-state data of bad pixels, and non-steady-state data of bottleneck caused by the shutdown event, the bad pixel event, and the bottleneck event, respectively;

[0022] Obtain the missing parameters of the key control parameters;

[0023] The missing parameters are filled in, and the shutdown unsteady-state data, bad point unsteady-state data, and bottleneck unsteady-state data are deleted from the key control parameters to obtain intermediate control parameters;

[0024] The intermediate control parameters are denoised to obtain the target control parameters.

[0025] In some embodiments, obtaining the non-steady-state data of shutdown, non-steady-state data of bad pixels, and non-steady-state data of bottleneck caused by the shutdown event, the bad pixel event, and the bottleneck event respectively includes:

[0026] Obtain the downtime period corresponding to the downtime event;

[0027] The corresponding shutdown-related key control parameters during the shutdown period are determined as shutdown unsteady-state data.

[0028] Determine the exact moment of the bad pixel event;

[0029] The key control parameters associated with bad points within the target time range at the specified time point are determined as bad point unsteady-state data.

[0030] The breakpoint of the bottleneck event is determined based on the push-pull principle;

[0031] Determine the past association key parameters of the preset past time period of the interruption point and the future association key parameters of the preset future time period of the interruption point;

[0032] The past correlation key parameters and the future correlation key parameters are determined as the bottleneck non-steady-state data.

[0033] In some embodiments, the step of dividing the target unit into operating conditions based on the target control parameters to obtain multiple target operating conditions that satisfy the optimization objective includes:

[0034] Obtain the target guidance parameter of the optimization target from the target control parameters;

[0035] The target unit is classified into operating conditions based on the target guidance parameters.

[0036] In some embodiments, determining the multiple influencing characteristic variables of the optimization objective under each target operating condition includes:

[0037] Based on the process mechanism, the necessary characteristic variables of the optimization objective under the target working condition are determined;

[0038] Based on the Pearson correlation formula, the marginal feature variables of the optimization target under the target working condition are determined;

[0039] Based on the necessary feature variables and the marginal feature variables, the influencing feature variables are determined;

[0040] The formula for the Pearson correlation coefficient is as follows:

[0041] ;

[0042] In the formula, the numerator is used to measure the covariance between the characteristic variable X and the optimization index Y (or the covariance between characteristic X1 and characteristic X2, reflecting the direction and strength of the linear association).

[0043] The denominator is the product of the standard deviations of X and Y (or feature X1 and feature X2), used for standardization to make the result in the interval [-1, 1].

[0044] In some embodiments, if there is a strong coupling relationship among multiple influencing feature variables, the method further includes:

[0045] Determine a set of mutually coupled characteristic variables from among the multiple influencing characteristic variables;

[0046] Based on the Pearson correlation coefficient formula, determine the reasonable coupled feature variables and redundant coupled feature variables in the coupled feature variable group;

[0047] The redundant coupled feature variables are removed from the coupled feature variable group to achieve decoupling of the coupled feature variable group.

[0048] In some embodiments, the method performs higher-order modeling and validation on the normalized feature variables, including:

[0049] Residual analysis, main effect analysis, and interaction effect analysis were performed on the adaptive model of the thermal power plant.

[0050] When the residual analysis, main effect analysis, and interaction effect analysis all meet their respective threshold conditions, the verification is deemed successful, and the verified adaptive model of the thermal power plant is obtained.

[0051] According to another aspect of this application, an adaptive modeling apparatus for thermal power plants is also disclosed, the apparatus comprising:

[0052] The key control parameter acquisition module is used to acquire multiple key control parameters of the target unit;

[0053] The target control parameter determination module is used to optimize the key control parameters to obtain the target control parameters corresponding to the target unit.

[0054] The optimization target acquisition module is used to acquire the optimization target of the target unit;

[0055] The operating condition division module is used to divide the target unit into operating conditions based on the target control parameters to obtain multiple target operating conditions that meet the optimization objectives.

[0056] The influencing characteristic variable determination module is used to determine multiple influencing characteristic variables of the optimization objective under each target working condition, wherein there is no coupling relationship between the multiple influencing characteristic variables;

[0057] The normalization module is used to normalize multiple feature variables corresponding to each target working condition in the interval [1,2].

[0058] The model determination module is used to perform high-order modeling and verification on the normalized feature variables to obtain the verified adaptive model of the thermal power plant; wherein, the adaptive model of the thermal power plant is characterized by the following formula:

[0059] ;

[0060] In the formula, To optimize the indicators;

[0061] The total number of feature variables corresponding to an optimization metric;

[0062] It is the highest order polynomial;

[0063] Single feature of Order coefficient;

[0064] Features , Interactive items , Order coefficient;

[0065] This is a constant term.

[0066] According to another aspect of this application, an electronic device is also disclosed, the electronic device including a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the electronic device to perform various steps of the adaptive modeling method for thermal power plants as described in any of the preceding claims.

[0067] According to another aspect of this application, a computer-readable storage medium is also disclosed, wherein instructions are stored on the computer-readable storage medium, characterized in that, when executed by a processor, the instructions implement the various steps of the adaptive modeling method for thermal power plants as described in any of the preceding claims.

[0068] The present invention includes, but is not limited to, the following beneficial effects: (1) This scheme divides the operating conditions by adapting to the optimization target, and can obtain target unit models that match different optimization targets. It has adaptability to thermal power plants with multiple target units, improves the model adaptation capability, and improves the model accuracy by optimizing key control parameters; (2) The influence feature variables in this scheme are uncoupled influence feature variables, which can avoid mutual interference between variables, reduce modeling errors, and allow the model to more accurately capture the independent and synergistic effects of each variable on the optimization target, thereby improving the accuracy of model prediction and control; (3) By normalizing the influence feature variables in the [1, 2] interval, the problem of reverse optimization results can be avoided; (4) This scheme can identify shutdown, bad points, bottleneck events and corresponding non-steady-state data, and selectively remove these interfering non-steady-state data, greatly improving the purity of modeling data and laying a high-quality data foundation for subsequent model construction; By acquiring and filling in missing parameters, the model training is not sufficient or biased due to missing data, ensuring the integrity of modeling data in the time and parameter dimensions, so that the model can fully learn the unit operation characteristics; By deleting Non-steady-state data, filling gaps, noise reduction and other processing, intermediate control parameters and target control parameters can more realistically reflect the unit's operating status. Inputting such data to train the model can reduce interference factors, make the model output more realistic, and high-quality, clean data can make the model training process more stable, reduce model oscillation or overfitting problems caused by abnormal data, and allow the model to maintain good performance under different operating conditions, improve the model's generalization ability and long-term operating reliability; (5) This scheme first determines the necessary feature variables through process mechanism to ensure that the features conform to physical laws, avoid data-driven model to select features irrationally, and then uses Pearson correlation coefficient to screen marginal feature variables, quantify the linear correlation between variables and optimization targets, supplement weakly correlated but valuable features not covered by the mechanism, make the feature set more complete and more in line with actual operating conditions, and improve model accuracy; furthermore, Pearson coefficient can identify strongly coupled influence feature variables, avoid redundant information between features from interfering with model training, and improve model accuracy; (6) This scheme decouples strongly coupled feature variables and deletes redundant feature variables to avoid prediction bias caused by repeated learning of the same influence, and improves model accuracy. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0070] Figure 1 This is a flowchart of an adaptive modeling method for thermal power plants according to an embodiment of this application;

[0071] Figure 2 This is another flowchart of the adaptive modeling method for thermal power plants according to the embodiments of this application;

[0072] Figure 3 This is another flowchart of the adaptive modeling method for thermal power plants according to the embodiments of this application;

[0073] Figure 4 This is another flowchart of the adaptive modeling method for thermal power plants according to the embodiments of this application;

[0074] Figure 5 This is another flowchart of the adaptive modeling method for thermal power plants according to the embodiments of this application;

[0075] Figure 6 This is a structural block diagram of the adaptive modeling device for thermal power plants according to an embodiment of this application;

[0076] Figure 7 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention;

[0077] Figure 8 This is a comparison chart of the predicted and actual values ​​of the model established in Example 1;

[0078] Figure 9 This is the residual analysis diagram from Example 1;

[0079] Figure 10 This is the main effect analysis diagram from Example 1;

[0080] Figure 11 This is the interaction effect analysis diagram from Example 1;

[0081] Figure 12 This is a comparison chart of the predicted values ​​and actual values ​​of the model established in Example 2;

[0082] Figure 13 This is the residual analysis diagram from Example 2;

[0083] Figure 14 This is the main effect analysis diagram from Example 2;

[0084] Figure 15 This is the interaction effect analysis diagram in Example 2. Detailed Implementation

[0085] This invention provides an adaptive modeling method for thermal power plants. The method includes: acquiring multiple key control parameters of the target unit; optimizing the key control parameters to obtain target control parameters corresponding to the target unit; acquiring the optimization objective of the target unit; dividing the target unit into operating conditions based on the target control parameters to obtain multiple target operating conditions that satisfy the optimization objective; determining multiple influencing feature variables of the optimization objective under each target operating condition, wherein there is no coupling relationship between the multiple influencing feature variables; normalizing the multiple feature variables corresponding to each target operating condition in the interval [1, 2]; and verifying the normalized influencing feature variables after high-order modeling to obtain a verified adaptive model of the thermal power plant. This scheme, by adapting the operating condition division to the optimization objective, can obtain target unit models that match different optimization objectives, has adaptability for thermal power plants with multiple target units, improves the model's adaptability, and improves the model's accuracy through the optimization of key control parameters.

[0086] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0087] For ease of understanding, the specific process of the embodiments of the present invention will be described below. Figure 1 This is a flowchart illustrating an adaptive modeling method for thermal power plants according to an embodiment of this application. Figure 1 As shown, it includes the following steps:

[0088] S100: Obtain multiple key control parameters of the target unit.

[0089] Specifically, key control parameters of the flue gas, steam-water, and pulverizing subsystems of the target unit within a target time period, such as within one year or six months, can be retrieved through the power plant's real-time monitoring system (SIS). Corresponding coal quality test data for the time period can be obtained from on-site reports. The target unit can refer to a specific generating unit in a thermal power plant that requires real-time optimization modeling and analysis, such as a 660MW supercritical generating unit. Key control parameters refer to the core parameters reflecting the unit's operating status retrieved from the plant-level real-time monitoring system (SIS) of a thermal power plant. These parameters may include, but are not limited to, the total coal feed rate, primary air volume, secondary air volume, furnace negative pressure, furnace oxygen content, air preheater outlet oxygen content, flue gas temperature, and induced draft fan current of the flue gas subsystem; feedwater flow rate, feedwater temperature, main steam flow rate, main steam pressure, main steam temperature, reheat steam temperature, reheat steam pressure, water-coal ratio, and superheat of the steam-water subsystem; operating parameters related to pulverized coal preparation in the pulverizing subsystem; and turbine-related parameters such as turbine load (active power), maximum turbine steam intake, condenser vacuum, circulating water feedwater temperature, and circulating water pump operating status. Specifically, the time granularity of these key control parameters is 1 minute, and coal quality test data is collected once per shift (8 hours). First, the timestamps and granularity of the control parameters and test data are aligned to achieve data merging.

[0090] S102. Optimize the key control parameters to obtain the target control parameters corresponding to the target unit.

[0091] Specifically, the optimization of key control parameters includes, but is not limited to, data cleaning and noise reduction. Data cleaning is used to remove some abnormal data, which may include, but is not limited to, duplicate data.

[0092] S104. Obtain the optimization target of the target unit.

[0093] The optimization objectives of the target unit may include, but are not limited to, combustion optimization and cold-end optimization.

[0094] S106. Divide the target unit into operating conditions based on the target control parameters to obtain multiple target operating conditions that meet the optimization objectives.

[0095] Specifically, different optimization objectives correspond to different operating conditions. Combustion optimization improves boiler efficiency (steam production per ton of coal, steam production per ton of standard coal) by adjusting air volume (primary air / secondary air). Cold-end optimization improves net power generation (power generation per ton of steam) by adjusting turbine back pressure (condenser vacuum). Net power generation refers to the difference between the active power of the turbine and the power consumption of the circulating water pumps. In one example, combustion optimization can divide operating conditions based on the actual unit load, i.e., active power, in the target control parameters. Depending on the amount of data, 4-8 operating conditions can be divided. Taking 4 operating conditions as an example, the first quartile, median, and third and fourth quartile values ​​of the active power dataset are used as the dividing boundaries. Cold-end optimization, in addition to dividing operating conditions based on the actual unit load, is also related to external environmental conditions, i.e., circulating water feedwater temperature. Furthermore, since most power plants' circulating water pumps are constant-speed pumps or dual-speed pumps switching between high and low speeds, the operating states of the circulating water pumps are discrete. Therefore, with sufficient data, the system can be divided into 4×4=16 operating conditions based on the main steam flow rate and circulating water feed temperature, and the optimal start-up and shutdown state of the circulating water pump can be found under each operating condition.

[0096] In another example, the target-oriented parameters of the optimization objective can be obtained first from the target control parameters; then, the target unit's operating conditions can be divided based on these parameters. Target-oriented parameters refer to those parameters in the target control parameters that are directly related to the optimization objective and can accurately reflect the direction and degree of influence of the target unit's operating state on the objective. For example, to optimize power generation efficiency, parameters that directly affect energy conversion, such as unit load, main steam pressure, and coal quality parameters (calorific value, volatile matter, etc.), would be selected as target-oriented parameters. Further, based on the different combinations and variation patterns of these parameters, the target unit's operating state can be divided into several operating conditions. For example, complex operating conditions can be categorized according to unit load range (low load, medium load, high load), coal quality type (high-quality coal, low-quality coal), or main steam pressure range, allowing the model to optimize separately for different operating conditions and improving adaptability.

[0097] S108. Determine the multiple influencing characteristic variables of the optimization objective under each target operating condition.

[0098] Among these, there is no coupling relationship between the multiple influencing characteristic variables. Specifically, for boiler combustion system modeling, the optimization objective y can be selected from main steam flow rate, gas production per ton of coal, steam production per ton of standard coal, and boiler efficiency. The dependent variable can be selected from multiple influencing characteristic variables x, such as total coal feed rate, feedwater flow rate, feedwater temperature, lower heating value of coal received basis, flue gas temperature, induced draft fan current, primary air volume, secondary air volume, furnace negative pressure, furnace oxygen content, air preheater outlet oxygen content, water-coal ratio, and superheat. Generally, the number of selected influencing characteristic variables is related to the amount of data and should not be excessive. For example, for each operating condition with more than 1000 sets of data, 3-6 influencing characteristic variables can be selected. In one example, such as... Figure 2 The diagram shown is another flowchart of the adaptive modeling method for thermal power plants according to the application embodiment. This flowchart is an exemplary illustration of step S108, which determines multiple influencing characteristic variables of the optimization objective under each target operating condition. Figure 2 As shown, it includes the following steps:

[0099] S200. Based on the process mechanism, determine the necessary characteristic variables of the optimization objective under the target operating condition.

[0100] Among them, process mechanism refers to the physical and chemical laws of thermal power production system and the operating principle of equipment. When selecting influencing characteristic variables, the necessary characteristic variables of the optimization objective under the target operating condition should be determined first. For example, in the combustion optimization process, the characteristics representing air volume and boiler load are necessary characteristic variables.

[0101] S202. Based on the Pearson correlation formula, determine the marginal characteristic variables of the optimization target under the target working condition.

[0102] Specifically, some features that cannot be directly determined, such as the water-to-coal ratio and furnace negative pressure, can be temporarily retained. Based on the Pearson correlation formula, correlation analysis is performed on the retained features x and the optimization objective y, and the Pearson correlation coefficient r is calculated. Special attention is paid to the correlation coefficient between the temporarily retained features x and y, and features with |r|>0.4 are retained. The formula is as follows:

[0103] ;

[0104] In the formula, the numerator is used to measure the covariance between the characteristic variable X and the optimization index Y (or the covariance between characteristic X1 and characteristic X2, reflecting the direction and strength of the linear association).

[0105] The denominator is the product of the standard deviations of X and Y (or feature X1 and feature X2), used for standardization to make the result in the interval [-1, 1].

[0106] S204. Based on the necessary feature variables and marginal feature variables, determine the influencing feature variables.

[0107] Specifically, necessary feature variables and marginal feature variables are used as influencing feature variables.

[0108] Understandably, the above method still retains multiple features x and multiple features y. Subsequently, multiple influencing feature variables can be matched with multiple optimization objectives through permutation and combination. However, the currently retained influencing feature variables may exhibit strong coupling between features and between features and objectives. For example, there is strong coupling between coal feed rate and water feed rate, water feed rate and main steam flow rate, and boiler efficiency and flue gas temperature. These need to be decoupled before modeling.

[0109] Specifically, in this example, such as Figure 3 As shown, another flowchart of the adaptive modeling method for thermal power plants is presented. (See attached document.) Figure 3 It includes the following steps:

[0110] S300. Identify the set of coupled characteristic variables from multiple influencing characteristic variables.

[0111] S302. Based on the Pearson correlation coefficient formula, determine the reasonable coupled feature variables and redundant coupled feature variables in the coupled feature variable group.

[0112] For example, based on the Pearson correlation coefficient formula, the coal feed rate and water feed rate are calculated to be r=0.85, indicating strong coupling. Furthermore, the coal feed rate is related to the optimization objective (e.g., boiler efficiency) with r=0.7, and the water feed rate is related to the optimization objective (boiler efficiency) with r=0.6. Therefore, the coal feed rate can be determined as a reasonably coupled characteristic variable, while the water feed rate is a redundant coupled characteristic variable.

[0113] S304. Remove redundant coupled feature variables from the coupled feature variable group to achieve decoupling of the coupled feature variable group.

[0114] Specifically, steps S302-S304 decouple strongly coupled influence feature variables. That is, one feature is retained while features strongly coupled with the target are deleted; otherwise, the influence trend of other features on the target would be greatly reduced or misjudged. This step also significantly reduces the number of permutations and combinations, thus reducing the modeling workload. For steam turbine modeling, the modeling target y can be selected as the unit's active power and power generation per ton of steam. The dependent variables can be selected from features x such as main steam flow rate, temperature, and pressure; reheat steam temperature and pressure; condenser vacuum degree; and circulating water feedwater temperature. The subsequent feature selection method is similar to that of combustion optimization.

[0115] In essence, this approach uses uncoupled influencing variables to avoid mutual interference, reduce modeling errors, and allow the model to more accurately capture the independent and synergistic effects of each variable on the optimization objective, thereby improving the accuracy of model prediction and control. Furthermore, this example approach first determines necessary feature variables through process mechanisms to ensure that features conform to physical laws and avoid data-driven models arbitrarily selecting features. Then, it uses the Pearson correlation coefficient to screen marginal feature variables, quantifying the linear relationship between variables and the optimization objective, and supplementing weakly correlated but valuable features not covered by the mechanism. This makes the feature set more complete and more closely reflects actual working conditions, improving model accuracy. Moreover, the Pearson coefficient can identify strongly coupled influencing feature variables, avoiding redundant information between features from interfering with model training and improving model accuracy.

[0116] S110. Normalize the multiple feature variables corresponding to each target working condition in the interval [1,2].

[0117] Specifically, to ensure that different feature variables have the same weight in the model and to avoid the situation where "large numbers eat up small numbers," each feature variable needs to be "normalized" before model training. The "normalization" process is as follows:

[0118] ;

[0119] In the formula, For feature predictors with dimensions;

[0120] for The maximum value in the array;

[0121] for The minimum value of the array.

[0122] The maximum and minimum values ​​can be defined by process engineers based on the actual process (if historical data includes extreme operating conditions, the upper limit of the historical data can be used as the upper limit and the lower limit as the lower limit). To ensure model accuracy, under normal factory operating conditions, the upper and lower limits of the actual feature variables should not exceed this range; this range should also not be set too large to avoid affecting the effect of the feature variable in the model. Through the above method, all feature variables are converted to values ​​between 1 and 2. The reason for not converting feature variables to the traditional normalization strategy of [0,1] is mainly because in actual operation, there will be situations where feature variables are less than their lower limit. In this case, the normalization result of the feature variable will be less than 0, causing the effect of the feature variable in the model to be opposite, leading to the problem of opposite optimization results. Specifically, by normalizing the influencing feature variables in the range of [1,2], the problem of opposite optimization results can be avoided.

[0123] S112. After performing high-order modeling on the normalized influence characteristic variables, the validated adaptive model of the thermal power plant is obtained.

[0124] Specifically, the adaptive model for thermal power plants is represented by the following formula:

[0125] ;

[0126] In the formula, To optimize the indicators;

[0127] The total number of feature variables corresponding to an optimization metric;

[0128] It is the highest order polynomial;

[0129] Single feature of Order coefficient;

[0130] Features , Interactive items , Order coefficient;

[0131] This is a constant term.

[0132] Based on the above formula, list the response y and the predictor variables. , , ...... of This method uses polynomial equations of order one. It aims to extract modeling information from the system's input and output data, and construct mathematical relationships between independent variables using the main factors influencing the output information. If the optimization objective y, obtained through feature selection, is steam production per ton of coal, then the influencing feature variable x is: total primary air volume. Total secondary air volume Main water supply flow Water-to-coal ratio ,when When =1, the first-order polynomial ,when When =2, the second-order polynomial .

[0133] The logistics of power plants typically exhibit clear causal relationships, such as the correlation between total coal feed and steam pressure, and between air temperature and combustion efficiency. Therefore, this invention samples first-order polynomials to construct independent models based on the main responses of the plant output and various constraints, retaining the cross terms in the high-dimensional representation, eliminating square terms without practical physical meaning, avoiding the risk of overfitting, and preserving the interpretability of the process mechanism.

[0134] Specifically, after feature selection, multiple optimization objectives y and multiple influencing feature variables x are obtained, thus enabling the construction of multiple polynomial models. Past experience typically involves selecting the model with the best accuracy based on regression model evaluation metrics such as mean squared error and mean absolute error. However, in actual industrial production, while model accuracy is important, the adjustment direction and magnitude of the feature variables are even more crucial. If the coefficients of the feature variables in the model have opposite signs, the actual adjustment direction is incorrect, leading to a decrease in unit efficiency. In this application, the polynomial model is validated by performing residual analysis, main effect analysis, and interaction effect analysis on the adaptive model of a thermal power plant. When the residual analysis, main effect analysis, and interaction effect analysis all meet their respective threshold conditions, the validation is deemed successful, resulting in a validated adaptive model for the thermal power plant.

[0135] (1) For residual analysis:

[0136] After performing multidimensional characterization, the actual and fitted values ​​of each optimization objective y can be obtained. The residuals between the actual and fitted values ​​are calculated as follows: Before performing residual analysis, the main steps are to create a normal probability plot of the residuals, a histogram of residual frequencies, a plot of residuals versus fitted values, and a residual order plot. During model residual analysis, three main aspects are analyzed:

[0137] The residuals of the model should roughly conform to a normal distribution: In a normal probability plot, ideally, the data points should be distributed approximately in a straight line along the diagonal, indicating that the residuals conform to the normality assumption. Check whether the frequency histogram shows a symmetrical bell-shaped distribution.

[0138] Residual and fitted value distribution analysis: Examine whether the residuals maintain equal variance, that is, the residuals are randomly dispersed around 0 values ​​without systematic changes, and should not have a funnel shape or a trumpet shape.

[0139] Residual order plot analysis: Examine whether the points in the scatter plot fluctuate randomly and irregularly above and below the horizontal axis.

[0140] (2) For main effects analysis:

[0141] The main effect refers to the average effect of different levels of a single independent variable (factor) on the dependent variable in a multifactor experiment or study, without considering the interaction effects of other variables.

[0142] The specific implementation method is as follows: After the model is trained, other feature variables need to be kept constant. The normalized mean of the data can be used, and only the single independent variable is varied. The result of this variation is then substituted into the model equation to observe its impact on the response. For example, when analyzing the impact of coal calorific value on steam production per ton of coal, the influence of variables such as boiler load and air volume on the system needs to be excluded before observation. Therefore, the normalized average of historical data for variables other than coal calorific value is used in the model, and the time-series data of normalized coal calorific value is used to observe the target trend. This trend should be consistent with the engineer's experience, such as the positive correlation between steam production per ton of coal and coal calorific value, the positive correlation between steam production per ton of coal and boiler load, and the positive correlation between air volume and outlet oxygen. Since this invention uses a first-order model, a main effect analysis is performed on each feature variable. The slope of the corresponding straight line can be used to determine the degree of influence of each feature on the target, and to observe whether the relationship between the change of a single physical quantity and the response conforms to the process logic and the actual production variation. Features that conform to the engineer's experience are retained and combined with the training model.

[0143] (3) For interaction effect analysis:

[0144] Interaction effects refer to the combined effect of two or more parameters on the final result, which is not simply the sum of the individual effects of each parameter. For example, the combination of total coal feed and total air volume in boiler combustion has a nonlinear effect on combustion efficiency, i.e., in a first-order model... This refers to the interactive items introduced.

[0145] The interaction effect is calculated as follows: If the characteristic variables are selected... , needs to be calculated and The interaction effect, after the model is trained, needs to be addressed by removing... and With the characteristic variables remaining unchanged, and taking the mean of the normalized data as well, first... Take the maximum value (which is 2 after normalization). Take the normalized average of its historical data and input it into the model to observe. The impact of changes on the optimization objective y; then Take the minimum value (which becomes 1 after normalization). Take the normalized average of its historical data and input it into the model to observe. The effect of changes on the optimization objective y, and vice versa. Take the maximum or minimum value and observe. The impact of changes on the target y. By constructing a lower triangular interaction effect analysis diagram, traversing all parameter combinations, identifying the impact of key interaction pairs, and retaining features that conform to engineer experience for selection and combination with its training model.

[0146] In one example, Figure 4 This is another flowchart of the adaptive modeling method for thermal power plants according to embodiments of this application. This flowchart is an exemplary illustration of step S102, which optimizes key control parameters to obtain target control parameters corresponding to the target unit. For details, please refer to [link to relevant documentation]. Figure 2 It includes the following steps:

[0147] S400: Identify downtime events, dead pixel events, and bottleneck events in the critical control parameters.

[0148] Specifically, a shutdown event refers to the process of a unit transitioning from an "operating state" to a "shutdown state," including planned shutdowns (such as maintenance and peak-shaving depth shutdowns) and unplanned shutdowns (such as fault trips and protection actions). A dead point event refers to an event where "unreasonable data fluctuations" occur due to sensor failures or transient equipment anomalies. A bottleneck event refers to an "operational bottleneck" caused by equipment hardware limitations or operating strategies, preventing the unit from continuing to optimize its objectives (such as efficiency or load), manifesting as an event where "parameters are adjusted to their limits but there is no response."

[0149] S402. Obtain the non-steady-state data of shutdown, non-steady-state data of bad pixels, and non-steady-state data of bottleneck caused by shutdown events, bad pixel events, and bottleneck events, respectively.

[0150] Specifically, Figure 5 This is another flowchart of the adaptive modeling method for thermal power plants according to embodiments of this application. This flowchart is an exemplary illustration of step S402, which obtains non-steady-state data related to shutdown events, bad pixel events, and bottleneck events, respectively. For details, please refer to... Figure 5 It includes the following steps:

[0151] S500: Obtain the downtime period corresponding to the downtime event.

[0152] In one example, the status signals of equipment related to shutdown (such as boilers and steam turbines) can be captured from a real-time database using the plant-level monitoring information system (SIS) and distributed control system (DCS) of a thermal power plant. When the equipment's operating status changes from "operating" to "shutdown" (e.g., the steam turbine speed drops to 0, the boiler's main burner stops completely, etc.), the timestamp at this point is recorded as the shutdown start point; when the equipment restarts and enters a stable operating state (speed reaches rated, boiler resumes stable steam production, etc.), the recovery timestamp is recorded as the shutdown end point. The shutdown period is determined by the shutdown start point and shutdown end point.

[0153] Another feasible approach is to combine logs to determine the downtime period.

[0154] S502. Determine the corresponding shutdown-related key control parameters during the shutdown period as shutdown unsteady-state data.

[0155] Specific shutdown-related key control parameters can refer to parameters that are directly affected by the shutdown action and are strongly correlated with the shutdown process. These can include equipment status parameters, process parameters, and auxiliary system parameters. Among them, equipment status parameters can include turbine speed, generator active power, and total boiler coal feed; process parameters can include main steam pressure and reheat steam temperature; and auxiliary system parameters can include circulating water pump frequency and induced draft fan current.

[0156] S504. Determine the timing of the bad pixel event.

[0157] Specifically, the time point when a sensor or other component malfunctions can be determined as the time point of the dead event.

[0158] S506. Determine the key control parameters associated with bad points within the target time range of the time point as bad point unsteady-state data.

[0159] It is understandable that fluctuations in thermal power plant parameters have "inertia," and sensor failures may be "gradual" (e.g., a response delay before drift), or bad spots may trigger "chain reactions" (e.g., false coal feed data leading to abnormal airflow regulation). Therefore, we can take the moment of the bad spot event as the center and extend forward or backward over a target time period to determine the corresponding bad spot-related key parameters within that time range as the bad spot's unsteady-state data. These bad spot-related key control parameters can refer to the physical mechanism, control logic, and equipment relationships of the sensors or hardware devices involved in the bad spot event.

[0160] S508. Determine the interruption point of the bottleneck event based on the push-pull principle.

[0161] S510 determines the past association key parameters of the preset past time period of the interruption point and the future association key parameters of the preset future time period of the interruption point.

[0162] S512. Determine the key parameters related to the past and the key parameters related to the future as bottleneck non-steady-state data.

[0163] Understandably, besides deleting non-steady-state data and bad point non-steady-state data caused by long-term downtime due to market changes, raw material supply, and equipment failures, production lines often have operational bottlenecks, which limit production capacity. An effective way to identify production bottlenecks is through the "push-pull rule": Before the bottleneck, upstream processes produce according to a planned or fixed rhythm, "pushing" materials downstream. If upstream capacity exceeds the bottleneck, continuing to push production will lead to intermediate products accumulating before the bottleneck. After the bottleneck, downstream processes have higher capacity than the bottleneck, but because the bottleneck restricts material supply, they can only passively wait (pull production). Operational bottlenecks cause raw material supply interruptions. With sufficient data, directly find the raw material supply interruption point, identify the key parameters related to the "past" and "future" time periods of this interruption as non-steady-state data, and delete them all. For power generation companies, find the interruption point in total coal supply, and look for non-steady-state data around that point, deleting all rows with the corresponding timestamps.

[0164] The preset past time period can refer to a period of time before the interruption point occurs, and the preset future time period can refer to a period of time after the interruption point occurs. Past key related parameters can refer to process parameters, equipment status parameters, and operational control parameters that are directly or indirectly related to the interruption point (such as equipment failure, data anomaly, or other critical time nodes) within a past historical time period, reflecting, influencing, or being affected by the interruption point. Future related key parameters can refer to process parameters, equipment status parameters, and operational control parameters that are directly or indirectly related to the interruption point (such as equipment failure, data anomaly, or other critical time nodes) within a future time period, reflecting, influencing, or being affected by the interruption point.

[0165] Understandably, this example solution can identify downtime, faulty nodes, bottleneck events, and corresponding non-steady-state data, and specifically remove these interfering non-steady-state data, significantly improving the purity of the modeling data and laying a high-quality data foundation for subsequent model construction. By acquiring and filling in missing parameters, it avoids insufficient model training or bias due to missing data, ensuring the integrity of the modeling data in both time and parameter dimensions, enabling the model to comprehensively learn the unit's operating characteristics. Through a series of processes such as deleting non-steady-state data, filling in gaps, and denoising, intermediate control parameters and target control parameters more accurately reflect the unit's operating status. Inputting such data to train the model can reduce interference factors, making the model output more realistic. Furthermore, high-quality, clean data makes the model training process more stable, reducing model oscillations or overfitting problems caused by abnormal data, allowing the model to maintain good performance under different operating conditions, and improving the model's generalization ability and long-term operational reliability.

[0166] S404. Obtain missing parameters for key control parameters.

[0167] S406. Complete the missing parameters and delete the shutdown unsteady-state data, bad point unsteady-state data and bottleneck unsteady-state data from the key control parameters to obtain intermediate control parameters.

[0168] S408. Denoise the intermediate control parameters to obtain the target control parameters.

[0169] Furthermore, to ensure the data integrity of the system, when there are many intermittent gaps in the collected data, data imputation methods are needed to complete the data. If the differences between the data points at the breakpoints are small, data from the previous time point can be used for imputation. If the differences between the data before and after the breakpoint are large, the average value of the data before and after the breakpoint can be used for imputation.

[0170] To prevent noise from affecting the data results, digital filtering methods are needed to remove noise and improve the representativeness of the data. Since the data sampling period is 1 minute, a large amount of data was deleted during data cleaning. This makes the existing data discontinuous. By adding a column to the merged table, rounding the original timestamps to 30 minutes, and applying a 30-minute time-area average filter to all parameters, the problem is solved.

[0171] For ease of understanding, this application provides an example using a 660MW supercritical thermal power generating unit:

[0172] Example 1: In the application of a 660MW supercritical thermal power generating unit, a boiler combustion system model was established. First, the key control parameters of the coordinated control, flue gas, steam-water, and pulverizing subsystems, as well as the turbine extraction, high-pressure heater, low-pressure heater, and cold terminal systems, were retrieved from the power plant's real-time monitoring system (SIS) for a target time period, such as July 1, 2023 to June 30, 2024. The time granularity was 1 minute, totaling approximately 530,000 data sets. Reports of on-site boiler coal quality tests, slag tests, and fly ash combustible material tests were obtained. Laboratory data was collected once per shift (8 hours), totaling approximately 1100 data sets.

[0173] First, the data from the control system and the testing system are aligned and merged based on the timestamps. Then, according to the data processing flow, non-steady-state data from shutdown, non-steady-state data with bad points, and non-steady-state data from bottlenecks are deleted. Further data filling and 30-minute data filtering are performed to obtain 12,600 sets of processed data for subsequent modeling.

[0174] Modeling of the boiler combustion system for a 660MW supercritical power plant:

[0175] Boiler combustion system data is based on the unit's power generation load P. gen As the basis for classifying operating conditions, a total of "P" is defined. gen <262MW”, 262<=P gen <287MW”, 287<=P gen <304MW”, 304<=P gen <360MW”, 360<=P gen <422MW”、、“422<=P gen <495MW”, 495<=P gen <547MW” and “ Pgen The ">=547MW" category includes 8 operating conditions, with approximately 1800 sets of data for each condition.

[0176] There are several optimization objectives for modeling, including main steam flow rate, steam production per ton of coal, steam production per ton of standard coal, boiler efficiency, and standard coal consumption for power generation. Main steam flow rate is obtained directly from data processing; steam production per ton of coal = main steam flow rate / total coal quantity; steam production per ton of standard coal = steam production per ton of coal × 29.307 / Qnet,ar, where Qnet,ar is the received lower heating value of the coal, in MJ / kg; boiler efficiency is calculated using the inverse balance method; standard coal consumption for power generation = total coal quantity × Qnet,ar / 29.307 / power generation.

[0177] Based on feature selection, six influencing variables were obtained and combined: total primary air volume, total secondary air volume, total coal feed rate, main feedwater flow rate, coal-water ratio, superheat, and lower heating value of the coal received. Since the total coal feed rate and main feedwater flow rate are highly coupled, they are not selected simultaneously. The main feedwater flow rate and the target main steam flow rate are also highly coupled; therefore, when the main steam flow rate is the target, it is best not to select the total coal feed rate and main feedwater flow rate as feature variables. Each combination has a maximum of four variables, and a multidimensional representation model of each feature variable and the target combination is constructed. Taking the operating condition "Pgen>=547MW", the target as steam production per ton of coal, and the feature variables as total primary air volume, total secondary air volume, main feedwater flow rate, coal-water ratio, and Qnet,ar, as an example, a comparison chart of the predicted and actual values ​​of the model is established as follows. Figure 8 As shown.

[0178] The established model was then analyzed step by step based on the process-oriented verification principle, and the residual analysis diagram is shown below. Figure 9 As shown, where, Figure 9 In the graph, a is the residual normal probability plot, b is the residual frequency histogram, c is the residual vs. fitted value plot, and d is the residual order plot. From the normal probability plot and histogram, it can be seen that the model's residuals should roughly conform to a normal distribution, and the residuals should maintain approximately equal variance with the fitted value distribution, without a funnel or trumpet shape. The residual order plot shows that the points fluctuate irregularly around the horizontal axis, indicating that the model has passed the residual analysis verification.

[0179] Further main effects analysis of the model was conducted, and the main effects analysis diagram is shown below. Figure 10 As shown, a larger main feedwater flow rate indicates a larger boiler load and more steam production per ton of coal, which conforms to the physical law that efficiency increases with increasing load and production experience. The lower heating value Q of the coal fed into the furnace is... net,ar The positive correlation between steam production per ton of coal and calorific value aligns with the physical law that higher calorific value results in greater steam production per ton of coal. Ignoring other uncertain variables that might influence the trend, we assume the model has passed main effect analysis. If this model is used for optimization, the main effect analysis indicates that reducing primary and secondary air volumes can increase steam production per ton of coal, thereby improving furnace efficiency. Figure 10In the diagram, e represents the main effect analysis of the total primary air volume, f represents the main effect analysis of the total secondary air volume, g represents the main effect analysis of the main feedwater flow rate, h represents the main effect analysis of the water-coal ratio, and i represents the main effect analysis of the lower heating value of the coal fed into the furnace.

[0180] Further interaction effect analysis was conducted, and the interaction effect analysis diagram is shown below. Figure 11 As shown, taking the interaction effect of water-coal ratio and total secondary air volume as an example, under high load, when the water-coal ratio reaches its maximum value, a decrease in total secondary air volume can increase steam production per ton of coal; when the water-coal ratio reaches its minimum value, the opposite is true, which is consistent with production experience; the synergistic effect between main feedwater flow rate and primary air volume is weak, and the water-coal ratio and Q... net,ar The synergistic effect is weak. The interaction effect analysis diagram does not show any obvious violations of physical laws and production experience, indicating that the model has been validated through interaction effect analysis. In summary, the feasibility of the model is proven.

[0181] Example 2:

[0182] In the application of a 660MW supercritical thermal power generating unit, boiler combustion system models and turbine system models are established. The turbine modeling does not consider cold-end optimization for the time being, focusing on the main steam flow rate. F vapor Based on this, the working conditions are divided into 8 categories. F vapor Divided into: F vapor < 731.24 t / h, 731.24 <=F vapor < 814.6 t / h, 814.6 <=F vapor < 868.38 t / h, 868.38 <=F vapor < 1031.1 t / h, 1031.1 <= F vapor <= 1219.0 t / h, 1219.0 <= F vapor < = 1426.7 t / h, 1426.7 <= F vapor <= 1578.8t / h and F vapor > = 1578.8t / h.

[0183] Steam turbine modeling objectivesy The unit's active power and power generation per ton of steam can be selected. Six characteristic variables are obtained through feature selection and combined: main steam flow rate, main steam pressure, main steam temperature, high-temperature reheat steam temperature, high-temperature reheat steam pressure, and condenser vacuum. Due to the high coupling between main steam flow rate and unit active power, main steam flow rate is not selected as a characteristic variable when the unit's active power is the target. Each combination has a maximum of four variables, constructing a multi-dimensional standard model for each characteristic variable and the target combination. The operating condition is "731.24". <=F vapor < Taking a model with a capacity of 814.6 t / h, a target power generation per ton of steam, and characteristic variables including main steam flow rate, main steam pressure, main steam temperature, high-temperature reheat steam temperature, and condenser vacuum as an example, a comparison chart of the predicted and actual values ​​of the model is shown below. Figure 12 As shown.

[0184] The established model was then analyzed step by step based on the process-oriented verification principle, and the residual analysis diagram is shown below. Figure 13 As shown, Figure 13 In the diagram, A is the residual normal probability plot, B is the residual frequency histogram, C is the residual and fitted value plot, and D is the residual order plot.

[0185] The normal probability plot and histogram show that the residuals of the model should roughly conform to a normal distribution. The residuals of the model should have roughly equal variance and no funnel or trumpet shape compared to the distribution of the fitted values. The residual order plot shows that the points fluctuate irregularly above and below the horizontal axis, indicating that the model has passed the residual analysis verification.

[0186] Furthermore, a main effects analysis diagram is generated, as shown in the diagram below. Figure 14 As shown in the figure, E represents the main steam flow rate, F represents the main steam header pressure, G represents the main steam header temperature, H represents the high-temperature reheat steam temperature, and I represents the condenser vacuum value. The power generation per ton of steam increases with increasing main steam pressure and temperature, and high-temperature reheat steam temperature, and increases with decreasing condenser back pressure, which conforms to the physical laws of steam turbine power generation, indicating that the model has passed the main effect analysis.

[0187] Furthermore, an interaction effect analysis was conducted, and the interaction effect analysis diagram is shown below. Figure 15As shown, taking main steam temperature and pressure as examples, ideally, simultaneously increasing both temperature and pressure would improve power generation efficiency. However, in actual systems, the effects of temperature and pressure changes on power generation are not linear and exhibit interactive effects. When the main steam temperature reaches its maximum value, changes in main steam pressure have little impact on the target, while when the main steam temperature reaches its minimum value, the power generation per ton of steam increases with increasing main steam pressure. This phenomenon is essentially due to the varying sensitivity of steam thermodynamic properties across different temperature ranges. At high temperatures, the marginal contribution of pressure to enthalpy decreases, while at low temperatures, increasing pressure effectively improves steam quality and work capacity. There are no significant synergistic effects between main steam flow rate and condenser vacuum, main steam temperature and high-temperature reheat steam temperature, or main steam temperature and condenser vacuum. This interaction effect analysis diagram does not significantly violate physical laws or production experience, indicating that the model has been validated through interaction effect analysis. In summary, the feasibility of this model is proven.

[0188] Furthermore, Figure 6 A structural block diagram of the adaptive modeling device for thermal power plants, such as... Figure 6 As shown, the device includes:

[0189] The key control parameter acquisition module is used to acquire multiple key control parameters of the target unit;

[0190] The target control parameter determination module is used to optimize key control parameters to obtain the target control parameters corresponding to the target unit.

[0191] The target acquisition module is optimized to acquire the optimization targets of the target unit.

[0192] The operating condition division module is used to divide the target unit into operating conditions based on the target control parameters in order to obtain multiple target operating conditions that meet the optimization objectives.

[0193] The module for determining influencing characteristic variables is used to determine multiple influencing characteristic variables of the optimization objective under each target working condition, wherein there is no coupling relationship between the multiple influencing characteristic variables;

[0194] The normalization module is used to normalize multiple feature variables corresponding to each target working condition in the interval [1,2].

[0195] The model determination module is used to perform high-order modeling and validation on the normalized feature variables to obtain a validated adaptive model for the thermal power plant; the adaptive model for the thermal power plant is characterized by the following formula:

[0196] ;

[0197] In the formula, To optimize the indicators;

[0198] The total number of feature variables corresponding to an optimization metric;

[0199] It is the highest order polynomial;

[0200] Single feature of Order coefficient;

[0201] Features , Interactive items , Order coefficient;

[0202] This is a constant term.

[0203] The application of the relevant modules of the device in this example can be referred to the relevant introduction of the method principle above, and will not be repeated here.

[0204] This scheme divides operating conditions by adapting to optimization objectives, resulting in target unit models that match different optimization objectives. It exhibits adaptability to multi-objective thermal power plants, improving model adaptability. Furthermore, optimization of key control parameters enhances model accuracy. Additionally, the scheme uses uncoupled influencing variables, avoiding mutual interference and reducing modeling errors. This allows the model to more accurately capture the independent and synergistic effects of each variable on the optimization objective, improving the accuracy of model prediction and control. Normalizing the influencing variables within the [1, 2] interval avoids the problem of contradictory optimization results. Furthermore, the scheme can identify shutdowns, faulty points, bottleneck events, and corresponding unsteady-state data, selectively removing these interfering unsteady-state data, significantly improving the purity of modeling data and laying a high-quality data foundation for subsequent model construction. By acquiring and supplementing missing parameters, it avoids insufficient model training or bias due to data loss, ensuring the integrity of modeling data in both time and parameter dimensions, enabling the model to comprehensively learn the unit's operating characteristics. By deleting unsteady-state data and supplementing... Through a series of processes such as gap filling and noise reduction, intermediate and target control parameters more accurately reflect the unit's operating status. Inputting such data to train the model reduces interference factors, making the model output more realistic. High-quality, clean data also ensures a more stable training process, reducing model oscillations or overfitting caused by abnormal data. This allows the model to maintain good performance under different operating conditions, improving its generalization ability and long-term reliability. Furthermore, this approach first determines necessary feature variables through process mechanisms, ensuring features conform to physical laws and avoiding arbitrary feature selection driven by data. Then, Pearson correlation coefficients are used to screen marginal feature variables, quantifying the linear relationship between variables and the optimization objective. This supplements weakly correlated but valuable features not covered by the mechanism, making the feature set more complete and closer to actual operating conditions, thus improving model accuracy. Additionally, Pearson coefficients can identify strongly coupled feature variables, avoiding redundant information between features from interfering with model training and improving model accuracy. After decoupling strongly coupled feature variables, redundant feature variables are removed, avoiding prediction bias caused by repeatedly learning the same influence and improving model accuracy.

[0205] above Figure 6 The apparatus in the embodiments of the present invention will be described in detail from the perspective of modular functional entities. The electronic device in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0206] Figure 7This is a schematic diagram of the structure of an electronic device 700 provided in an embodiment of the present invention. The electronic device 700 can vary significantly due to differences in configuration or performance, and may include one or more central processing units (CPUs) 710 (e.g., one or more processors) and a memory 720, and one or more storage media 730 (e.g., one or more mass storage devices) for storing application programs 733 or data 732. The memory 720 and storage media 730 can be temporary or persistent storage. The program stored in the storage media 730 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the electronic device 700. Furthermore, the processor 710 may be configured to communicate with the storage media 730 and execute the series of instruction operations in the storage media 730 on the electronic device 700.

[0207] Electronic device 600 may also include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input / output interfaces 760, and / or one or more operating systems 731, such as Windows Server, MacOSX, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 7 The illustrated electronic device structure does not constitute a limitation on electronic devices and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0208] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of an adaptive modeling method for thermal power plants.

[0209] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0210] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0211] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of adaptive modeling of a thermal power plant, characterized in that, The method comprises obtaining a plurality of key control parameters of a target unit; optimizing the key control parameters to obtain target control parameters corresponding to the target unit; obtaining an optimization target of the target unit; dividing the target unit into a plurality of target working conditions based on the target control parameters to obtain a plurality of target working conditions satisfying the optimization target; determining a plurality of influence characteristic variables of the optimization target under each target working condition, wherein there is no coupling relationship between the plurality of influence characteristic variables; normalizing the plurality of characteristic variables corresponding to each target working condition in the [1, 2] interval; performing high-order modeling and verification on the normalized influence characteristic variables to obtain a verified thermal power plant adaptive model; wherein the thermal power plant adaptive model is represented based on the following formula: wherein y is an optimization index; N is the total number of characteristic variables corresponding to an optimization index; K is the highest order of the polynomial; A i，k For a single feature x i of order k; B i，j，k，n For feature x i , x j k, n order coefficients of interaction term; C is a constant term.

2. The power plant adaptive model modeling method of claim 1, wherein, The optimization of the key control parameters to obtain the target control parameters corresponding to the target unit comprises: determining shutdown events, bad point events and bottleneck events in the key control parameters; obtaining shutdown non-steady-state data, bad point non-steady-state data and bottleneck non-steady-state data caused by the shutdown events, the bad point events and the bottleneck events respectively; obtaining missing parameters of the key control parameters; completing the missing parameters and deleting the shutdown non-steady-state data, the bad point non-steady-state data and the bottleneck non-steady-state data from the key control parameters to obtain intermediate control parameters; performing denoising processing on the intermediate control parameters to obtain the target control parameters.

3. The power plant adaptive model modeling method of claim 2, wherein, The obtaining of the shutdown non-steady-state data, the bad point non-steady-state data and the bottleneck non-steady-state data caused by the shutdown events, the bad point events and the bottleneck events respectively comprises: obtaining a shutdown time period corresponding to a shutdown event; determining the shutdown-associated key control parameters in the shutdown time period as the shutdown non-steady-state data; determining a time point of the bad point event; determining the bad point-associated key control parameters in a target time range of the time point as the bad point non-steady-state data; determining an interruption point of the bottleneck event based on a push-pull rule; determining past associated key parameters of a preset past time period of the interruption point and future associated key parameters of a preset future time period of the interruption point; determining the past associated key parameters and the future associated key parameters as the bottleneck non-steady-state data.

4. The power plant adaptive model modeling method of claim 1, wherein, The dividing of the target unit into a plurality of target working conditions based on the target control parameters to obtain a plurality of target working conditions satisfying the optimization target comprises: obtaining target-oriented parameters of the optimization target in the target control parameters; dividing the target unit into target working conditions based on the target-oriented parameters.

5. The power plant adaptive model modeling method of claim 1, wherein, The determination of a plurality of influence characteristic variables of the optimization target under each target working condition comprises: determining necessary characteristic variables of the optimization target under the target working condition based on a process mechanism; determining edge characteristic variables of the optimization target under the target working condition based on a Pearson correlation coefficient formula; determine the influence characteristic variable based on the necessary characteristic variable and the edge characteristic variable; wherein the Pearson correlation coefficient formula is: wherein the numerator is used to measure the covariance of the influence characteristic variable X and the optimization index Y, reflecting the linear correlation direction and strength; and the denominator is the product of the standard deviations of X and Y, used for standardization, so that the result is in the interval [-1, 1].

6. The power plant adaptive model modeling method of claim 1, wherein, If there is a strong coupling relationship between multiple influence characteristic variables, the method further comprises: determining a coupled characteristic variable group from the multiple influence characteristic variables; determining reasonable coupled characteristic variables and redundant coupled characteristic variables in the coupled characteristic variable group based on the Pearson correlation coefficient formula; deleting the redundant coupled characteristic variables from the coupled characteristic variable group to realize decoupling of the coupled characteristic variable group.

7. The power plant adaptive model modeling method of claim 1, wherein, The method comprises: performing residual analysis, main effect analysis, and interaction effect analysis on the thermal power plant adaptive model; when the residual analysis, main effect analysis, and interaction effect analysis all meet their respective threshold conditions, determining that the verification is passed, and obtaining the thermal power plant adaptive model that passes the verification.

8. A power plant adaptive model modeling apparatus characterized by comprising: The device comprises: a key control parameter acquisition module configured to acquire multiple key control parameters of a target unit; a target control parameter determination module configured to perform optimization processing on the key control parameters to obtain target control parameters corresponding to the target unit; an optimization target acquisition module configured to acquire an optimization target of the target unit; a working condition division module configured to divide working conditions of the target unit based on the target control parameters to obtain multiple target working conditions that meet the optimization target; an influence characteristic variable determination module configured to determine multiple influence characteristic variables of the optimization target under each target working condition, wherein there is no coupling relationship between the multiple influence characteristic variables; a normalization processing module configured to perform normalization processing on multiple characteristic variables corresponding to each target working condition in the interval [1, 2]; a model determination module configured to perform high-order modeling and verification on the normalized characteristic variables to obtain a thermal power plant adaptive model that passes the verification; wherein the thermal power plant adaptive model is represented based on the following formula: wherein y is an optimization index; N is the total number of characteristic variables corresponding to an optimization index; K is the highest order of the polynomial; A i，k is a single feature x i of order k; B i，j，k，n For feature x i , x j k, n order coefficients of interaction term; C is a constant term.

9. An electronic device, comprising: The electronic device comprises a memory and at least one processor, the memory having instructions stored therein; the at least one processor invokes the instructions in the memory to cause the electronic device to perform each step of the thermal power plant adaptive model modeling method according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon instructions, the instructions comprising, The instructions are executed by the processor to implement each step of the thermal power plant adaptive model modeling method according to any one of claims 1-7.

Images