Thermal power generating unit coal consumption prediction method based on function type data analysis
By using functional data analysis and linear programming optimization, the problem of efficiency decline in thermal power units under load changes was solved, and coal consumption was optimized and operating efficiency was improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-29
- Publication Date
- 2026-03-10
AI Technical Summary
Existing methods cannot effectively capture the temporal continuity of thermal power units, which cause the operating efficiency to decline under frequent load changes.
A functional data analysis approach was adopted, which involves extracting sample characteristic variables, smoothing, cluster analysis, and constructing a functional linear regression model. Combined with the multi-stage WRDD method, the adjustment range of controllable variables was optimized, and finally a linear programming model was established to minimize coal consumption.
It improves the operating efficiency of thermal power units under frequent load changes and reduces coal consumption.
Smart Images
Figure CN121638523A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system operation and energy-saving optimization technology, specifically a method for predicting coal consumption of thermal power units based on functional data analysis. Background Technology
[0002] With the rapid development of renewable energy sources such as wind power and photovoltaics, thermal power units are increasingly required to provide rapid response peak-shaving services in the power grid, and their operating characteristics are shifting from steady state to frequent dynamic changes.
[0003] Thermal power units are typically designed to achieve maximum energy efficiency under high load operation. Their operating efficiency often decreases with frequent load changes. Although various methods have been applied to predict the coal consumption of thermal power units, these methods are based on static or point-in-time data for modeling and cannot effectively capture the temporal continuity of thermal power unit changes. Summary of the Invention
[0004] The technical problem to be solved by this invention is to address the issue that the operating efficiency of thermal power units often decreases under frequent load changes, and to provide a method for predicting coal consumption of thermal power units based on functional data analysis.
[0005] The present invention solves the above-mentioned technical problems through the following technical solution:
[0006] This invention provides a method for predicting coal consumption of thermal power units based on functional data analysis. The method includes the following steps:
[0007] S1. Sample extraction: Based on the thermal cycle system, select feature variables in the thermal power unit dataset and extract samples.
[0008] S2. Smoothing of sample feature variable curves: Smoothing curves are applied to any given feature variable of the extracted sample to obtain the basis function coefficient vector.
[0009] S3. Cluster analysis and construction of functional linear regression model: Based on the coefficient vector of multiple feature variables of the sample, K-means algorithm is used to perform sample cluster analysis and construct functional linear regression model with coal consumption function as response variable;
[0010] S4. Obtaining the adjustment range of controllable variables: Use the multi-stage WRDD method to confirm the adjustment range of controllable variables;
[0011] S5. Coal consumption optimization and potential analysis: Establish a linear programming model to solve for the optimal coal consumption prediction of the sample.
[0012] The specific process in step S1 is as follows:
[0013] S11. Input historical data;
[0014] S12. Extract samples that meet the conditions, and extract corresponding samples based on the load data of the thermal power unit during operation.
[0015] The specific process in step S12 is as follows:
[0016] Step 1: Extraction of uncontrollable variables. Extract variables that affect the operation of thermal power units but cannot be controlled by humans.
[0017] Step 2: Extract process monitoring variables, and extract characteristic variables that reflect the operating status of thermal power units in detail.
[0018] Step 3: Extract controllable variables, extract the adjustable parameter variables during the operation of thermal power units.
[0019] The variables extracted in step one include unit load and blower inlet air temperature.
[0020] The variables extracted in step two include the oxygen content of flue gas before the air preheater, the oxygen content of flue gas after the air preheater, the condensate flow rate, the main steam pressure, the main steam temperature, the reheat steam temperature, the condenser circulating water inlet temperature, the condenser circulating water outlet temperature, and the return water current.
[0021] The variables extracted in step three include the unit's fuel supply, main feedwater flow rate, blower current, induced draft fan current, coal mill current, condenser vacuum, total air volume, and water supply current.
[0022] In step S2, based on the B-spline basis function, the given feature variables of any extracted sample are smoothed to obtain its corresponding coefficient vector. The obtained corresponding coefficient vectors are concatenated end to end to form the metric vector of the sample. Then, the distance between two samples is defined as the Euclidean distance between the two metric vectors.
[0023] The specific process in step S3 is as follows:
[0024] Step 1: Cluster analysis is performed under different unit load conditions. Based on the defined distance between samples, the K-means algorithm is used to perform cluster analysis on the sample sets under the load increase and load decrease scenarios, and the silhouette coefficient is used to determine the optimal number of clusters for each of the two scenarios.
[0025] Step 2: Construct functional linear regression models for different categories, using the coal consumption function as the response variable and the remaining 18 variables as independent variables. Construct functional regression models for each category and solve them. Construct functional linear regression models for each category and solve them.
[0026] The multi-stage WRDD method in S4 can be used to identify the improvement space of each stage in complex systems.
[0027] In step S5, the functional linear regression model established in step S3 and the adjustment range of the controllable variables obtained in step S4 are used to construct a corresponding linear programming model. The model is then optimized to minimize the predicted coal consumption value under the condition that the parameters of the controllable variables meet the threshold.
[0028] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of the present invention.
[0029] The positive and progressive effects of this invention are as follows:
[0030] This application represents the operating data of thermal power units as functional data. Functional data analysis methods can better analyze continuous data. By combining B-spline basis functions, function clustering, and variable selection methods, coal consumption can be predicted under a linear regression model. Furthermore, by combining a linear programming optimization model, the total coal consumption can be minimized under the constraints of adjustable operating parameters, thereby improving the operating efficiency of the unit under frequent load changes. Attached Figure Description
[0031] Figure 1 This is a schematic diagram of a coal consumption prediction method for thermal power units based on functional data analysis, according to an embodiment of the present invention.
[0032] Figure 2 This is a schematic diagram of the S1 process in an embodiment of the present invention.
[0033] Figure 3 This is a schematic diagram of the S12 process in an embodiment of the present invention.
[0034] Figure 4 This is a schematic diagram of the S3 process in an embodiment of the present invention. Figure 5 This is an example of the coal consumption prediction effect in an embodiment of the present invention. Detailed Implementation
[0035] The present invention will be further illustrated by way of embodiments below, but the present invention is not limited to the scope of the embodiments described herein.
[0036] A method for predicting coal consumption in thermal power units based on functional data analysis, comprising the following steps:
[0037] S1. Sample extraction: Based on the thermal cycle system, select feature variables in the thermal power unit dataset and extract samples.
[0038] S2. Smoothing of sample feature variable curves: Smoothing curves are applied to any given feature variable of the extracted sample to obtain the basis function coefficient vector.
[0039] S3. Cluster analysis and construction of functional linear regression model: Based on the coefficient vector of multiple feature variables of the sample, K-means algorithm is used to perform sample cluster analysis and construct functional linear regression model with coal consumption function as response variable;
[0040] S4. Obtaining the adjustment range of controllable variables: Use the multi-stage WRDD method to confirm the adjustment range of controllable variables;
[0041] S5. Coal consumption optimization and potential analysis: Establish a linear programming model to solve for the optimal coal consumption prediction of the sample.
[0042] The specific process in step S1 is as follows:
[0043] S11. Input historical data;
[0044] S12. Extract samples that meet the conditions, and extract corresponding samples based on the load data of the thermal power unit during operation.
[0045] Based on the distribution of historical operating data of thermal power units, continuous sequence data of unit load between 100MW and 130MW were first extracted as representative samples of the units under load increase and decrease scenarios. At the same time, in order to study the effect of thermal power units in the process of stable operation, data of unit load around 90MW, 100MW, 110MW and 120MW for a duration of more than 180 minutes were extracted as representative samples of stable operation of thermal power units.
[0046] The specific process in step S12 is as follows:
[0047] Step 1: Extraction of uncontrollable variables. Extract variables that affect the operation of thermal power units but cannot be controlled by humans.
[0048] Step 2: Extract process monitoring variables, and extract characteristic variables that reflect the operating status of thermal power units in detail.
[0049] Step 3: Extract controllable variables, extract the adjustable parameter variables during the operation of thermal power units.
[0050] The variables extracted in step one include unit load and blower inlet air temperature.
[0051] The variables extracted in step two include the oxygen content of flue gas before the air preheater, the oxygen content of flue gas after the air preheater, the condensate flow rate, the main steam pressure, the main steam temperature, the reheat steam temperature, the condenser circulating water inlet temperature, the condenser circulating water outlet temperature, and the return water current.
[0052] The variables extracted in step three include the unit's fuel supply, main feedwater flow rate, blower current, induced draft fan current, coal mill current, condenser vacuum, total air volume, and water supply current.
[0053] The amount of fuel supplied to the unit represents the amount of coal consumed.
[0054] In step S2, based on the B-spline basis function, the given feature variables of any extracted sample are smoothed to obtain its corresponding coefficient vector. The obtained corresponding coefficient vectors are concatenated end to end to form the metric vector of the sample. Then, the distance between two samples is defined as the Euclidean distance between the two metric vectors.
[0055] Each sample from a thermal power unit under different load scenarios contains 19 selected characteristic variables. Due to sensor sampling rate limitations, all characteristic variables in the raw data are characterized by discrete values. Based on B-spline basis functions, a smoothing curve is applied to the given characteristic variables of any sample to obtain the coefficients of the sample characteristic variables in the B-spline basis function space. This process can be expressed as:
[0056]
[0057] in, The characteristic variable representing any given sample This indicates the time point at which the data was collected, while This represents the curve expression obtained after smoothing. and (These represent the selected B-spline basis function vector and the corresponding coefficient vector, respectively). represent The second derivative, smoothing parameter This is used to control the smoothness (proportional) of the curve obtained from the processing. During implementation, the smoothing parameter... Finding the optimal value uses Values are performed, that is, within a given range, such that smallest Specifically, the value is achieved using the lambda2gcv function from the fda package in R, given... Value Value calculation, The range of values is set to The smoothing process is implemented using the smooth.basis function from the fda package in R.
[0058] Given a B-spline basis function, smoothing the feature variables yields their corresponding coefficient vectors. The first and last parts of the coefficient vectors are then concatenated to form the metric vector of the sample. The distance between two samples is then defined as the Euclidean distance between the two metric vectors.
[0059] The specific process in step S3 is as follows:
[0060] Step 1: Cluster analysis is performed under different unit load conditions. Based on the defined distance between samples, the K-means algorithm is used to perform cluster analysis on the sample sets under the load increase and load decrease scenarios, and the silhouette coefficient is used to determine the optimal number of clusters for each of the two scenarios.
[0061] The contour coefficient SC is:
[0062]
[0063] in, This represents the total number of samples participating in the clustering. Representing the The arithmetic mean of the distances between a sample and other samples in the same class. Then it represents the first The arithmetic mean of the distances between a sample and other samples in a different class. The range of values is The larger the value, the better the clustering effect under the partitioning method.
[0064] Given a number of clusters and a sample set, the results of the K-means algorithm are also affected by the selection of initial sample points in different clusters. Therefore, for each number of clusters, the range is set to... Repeat the K-means algorithm 25 times to find the largest value. The value is used as the number of clusters. Value, finally passed The value is selected to determine the optimal number of clusters for a given set of samples participating in the clustering. Specifically, K-means clustering is implemented using the `kmeans` function from the `stats` package in R, and the silhouette coefficients for a given classification are calculated using the `silhouette` function from the `cluster` package in R.
[0065] Step 2: Construct functional linear regression models for different categories, using the coal consumption function as the response variable and the remaining 18 variables as independent variables. Construct functional regression models for each category and solve them. Construct functional linear regression models for each category and solve them.
[0066] Using coal consumption function as response variable Construct the following functional linear regression model:
[0067]
[0068] in, Indicates the first One characteristic variable, It is its corresponding coefficient function.
[0069] In the model solution process, a feature selection method based on the Huber loss function and group Lasso regularization is used to extract key feature variables affecting coal consumption. Specifically, during the solution process, the Huber loss function and group Lasso regularization are used to transform the initial objective function into a globally non-differentiable but decomposable objective function. Then, the proximal gradient descent method can be used to solve it, automatically shrinking the coefficients of non-key feature variables to 0, thereby selecting key feature variables while ensuring computational efficiency. The above solution process is based on the implementation framework of the proximal gradient descent algorithm, combined with the proxgrad function from the proxgrad package in R.
[0070] The multi-stage WRDD method in S4 can be used to identify the improvement space of each stage in complex systems.
[0071] The multi-stage WRDD method, also known as the multi-stage weighted Russell directional distance method, is used in this application to determine the adjustable range of controllable variables. This avoids the adverse effects of subjectively assigning adjustment ranges to controllable variables, thereby ensuring the feasibility and safety of thermal power unit operation under subsequent optimization and adjustment. Based on 19 existing characteristic variables in the thermal power unit data, a two-stage WRDD method is used. The input variables of the first stage are all controllable variables, and the output variables are the main steam pressure and main steam temperature from the process monitoring variables. The input variables of the second stage are the output variables from the first stage, and the output variable is the unit load. In the implementation process, firstly, the actual values of the input and output of the first stage are input into the model to obtain the ideal value of the first stage output. Then, this is used as the input value of the second stage, and together with the actual output value of the second stage, it is input into the model to obtain the ideal value of the second stage output. Finally, the actual input value of the first stage and the ideal value of the second stage output are input into the model to obtain the ideal value of the first stage input, which is the lower limit of the adjustment range of all controllable variables. At the same time, the upper limit of its adjustment range is defined as the actual value in the historical data. The above process is based on the modeling concept of multi-stage data envelopment analysis and is implemented using the dea function in the dea package of the R language.
[0072] In step S5, the functional linear regression model established in step S3 and the adjustment range of the controllable variables obtained in step S4 are used to construct a corresponding linear programming model. The model is then optimized to minimize the predicted coal consumption value under the condition that the parameters of the controllable variables meet the threshold.
[0073] The corresponding linear programming model is:
[0074]
[0075]
[0076]
[0077] in, For the sample The included time points, This can be represented as the sum of two parts: the first part contains controllable characteristic variables, while the second part contains the remaining variables; that is, the first part can be optimized, while the second part has fixed values. The constraints only involve... The variables in the set, i.e., the optimization process, only apply to controllable variables, which involve... and These represent the lower and upper limits of the characteristic variable, respectively, which are the adjustment ranges of the controllable variables obtained in step 4.
[0078] The coal consumption improvement potential for each sample is defined as:
[0079]
[0080] in, For the sample The minimum coal consumption prediction value obtained from the above linear programming model is the coal consumption prediction value corresponding to the optimal value of the controllable variables. This is the predicted coal consumption value when the controllable variables for this sample are the actual values, that is, the predicted coal consumption value when all controllable variables are taken as the upper limit of the constraints.
[0081] The results obtained based on the above prediction methods are as follows:
[0082] (1) Coal consumption prediction effect:
[0084] (2) Coal consumption improvement potential: The improvement effect of a category is the arithmetic mean of the coal consumption improvement potential of each sample in that category.
[0085] category Coal consumption improvement effect Unit load is stable -1.46% Unit load increased by 1 -5.23% Unit load increased by 2 -6.57% Unit load reduced by 1 -1.91% Unit load reduced by 2 -0.51%
[0086] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but all such changes and modifications fall within the scope of protection of the present invention.
Claims
1. A method for predicting coal consumption of a thermal power unit based on functional data analysis, characterized in that: The function type data analysis coal consumption prediction method of the thermal power unit comprises the following operation steps: S1, extracting samples, selecting characteristic variables in the thermal power unit data set based on the heat cycle system, and extracting samples; S2, sample characteristic variable curve smoothing processing, performing smoothing curve processing on the given characteristic variable of any extracted sample to obtain a basis function coefficient vector; S3, clustering analysis and construction of a function type linear regression model, performing sample clustering analysis by using a K-means algorithm according to the coefficient vector of the multiple characteristic variables of the sample, and constructing a function type linear regression model with the coal consumption function as a response variable; S4, obtainable controllable variable adjustment range, confirming the adjustment range of the controllable variable by using a multi-stage WRDD method; S5, coal consumption optimization and potential analysis, establishing a linear programming model to solve the optimal coal consumption prediction of the sample.
2. The function-based data analysis method for predicting coal consumption of a thermal power unit according to claim 1, characterized in that: The specific process in the S1 step is as follows: S11, input historical data; S12, extract samples meeting the conditions, and extract corresponding samples according to the load data in the operation of the thermal power unit. 3.The function-based data analysis based coal consumption prediction method for thermal power generating units according to claim 2, characterized in that: The specific process in the S12 step is as follows: Step one: uncontrollable variable extraction, extracting variables that cannot be artificially controlled and affect the operation of the thermal power unit. Step two: process monitoring variable extraction, extracting characteristic variables that reflect the operating state of the thermal power unit in detail. Step three: controllable variable extraction, extracting parameter variables that can be adjusted in the operation of the thermal power unit.
4. The function-based data analysis method for predicting coal consumption of a thermal power unit according to claim 3, characterized in that: The variables extracted in the step one include unit load and air inlet temperature of the air supply fan.
5. The method of claim 3, wherein the method further comprises: The variables extracted in the step two include oxygen content of flue gas before the air preheater, oxygen content of flue gas after the air preheater, condensate flow, main steam pressure, main steam temperature, reheat steam temperature, condenser circulating water inlet temperature, condenser circulating water outlet temperature, and backwater current.
6. The function-based data analysis method for predicting coal consumption of a thermal power unit according to claim 3, characterized in that: The variables extracted in the step three include unit fuel supply, main feedwater flow, air supply fan current, induced draft fan current, coal mill current, condenser vacuum, total air volume, and water supply current.
7. The function-based data analysis method for predicting coal consumption of a thermal power unit according to claim 1, characterized in that: In the S2 step, the given characteristic variable of any extracted sample is subjected to smoothing curve processing based on a B-spline basis function to obtain a corresponding coefficient vector, the obtained corresponding coefficient vectors are spliced at the beginning and the end to form a metric vector of the sample, and then the distance between two samples is defined as the Euclidean distance between the two metric vectors. 8.The function-based data analysis based coal consumption prediction method for thermal power generating units according to claim 1, characterized in that: The specific process in the S3 step is as follows: Step one: clustering analysis under different unit load states, performing clustering analysis on the sample set under the load increase scenario and the load decrease scenario by using the K-means algorithm according to the defined distance between samples, and determining the optimal clustering number of each scenario by using a silhouette coefficient. Step two: constructing a function type linear regression model for different categories, constructing a function type regression model for solving by taking the coal consumption function as a response variable and the remaining 18 variables as independent variables, and constructing a function type linear regression model for solving. 9.The function-based data analysis based coal consumption prediction method for thermal power generating units according to claim 1, characterized in that: The multi-stage WRDD method in the S4 can be used to identify the improvement space of each stage in a complex system. 10.The function-based data analysis method for predicting the coal consumption of a thermal power unit according to claim 1, characterized in that: The function linear regression model established in S3 and the adjustment range of the controllable variable obtained in S4 are used to construct a corresponding linear programming model, and the controllable variable parameters are optimized to minimize the coal consumption prediction value under the condition that the controllable variable parameters meet the threshold value.