A spline-based fuzzy cognitive map method for combined cycle power plant output power prediction
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTHEAST DIANLI UNIVERSITY
- Filing Date
- 2026-05-08
- Publication Date
- 2026-07-24
AI Technical Summary
Existing technologies cannot provide high-precision forecasts of combined cycle power plant output power, resulting in delays in power generation plans and an inability to meet electricity demand and environmental protection requirements.
A spline-based fuzzy cognitive graph approach is adopted, which replaces the connection weights of the traditional fuzzy cognitive graph by constructing a learnable B-spline function to establish an output power prediction model for combined cycle power plants, and uses historical operating data for training and prediction.
It achieves high-precision power plant output power prediction, improves the accuracy of power generation planning, reduces operating costs and environmental impact, and has interpretability and visualization effects.
Smart Images
Figure CN122456474A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of combined cycle power plant output power prediction technology, specifically relating to a method for predicting the output power of combined cycle power plants based on spline fuzzy cognitive graphs. Background Technology
[0002] Combined cycle power generation is a new form of power generation. Under conditions of rapid electricity demand growth, a crucial principle that combined cycle power plants must adhere to when performing tasks such as power generation and peak shaving is ensuring that power generation plans are completed on time. However, due to the lag in power generation plans, real-time information cannot be provided to the power plants. Therefore, forecasting power generation has become a problem that needs to be solved. Forecasting the power generation of power plants and its changing trends has always been an important reference for regulating electricity demand.
[0003] In most electricity markets, each power plant must submit its power production forecasts to the dispatch center. Furthermore, performance monitoring systems are used to ensure that power plants can achieve the optimal output-input ratio. For power plants, improving the accuracy of forecasts effectively reduces power generation operating costs and energy consumption, which in turn reduces environmental damage, ensuring that power plants can contribute to industrial restructuring and improve their economic viability. High-precision power plant output forecasting plays a crucial role in this process. Therefore, modern combined cycle power plants require a simple and efficient power forecasting method to meet their operational needs. Summary of the Invention
[0004] The purpose of this invention is to provide a method for predicting the output power of a combined cycle power plant based on spline fuzzy cognitive graphs, which can be directly applied to practical systems and has strong practicality.
[0005] To achieve the above objectives, the present invention provides the following solution: a method for predicting the output power of a combined cycle power plant based on spline fuzzy cognitive graphs, comprising the following steps: S1. Collect historical operating data of the combined cycle power plant under full load operation, and obtain the set of cause nodes and the set of result nodes based on the historical operating data; S2. Construct a spline fuzzy cognitive graph topology for the set of cause nodes and the set of result nodes; S3. The connection weights of the spline fuzzy cognitive graph topology are parameterized to obtain the output power prediction model of the combined cycle power plant; S4. Based on the output power prediction model, process the real-time operating data to obtain the power prediction results, and draw the spline function curves of each side to visualize the nonlinear relationship.
[0006] More preferably, the set of cause nodes includes: ambient temperature, exhaust vacuum, ambient pressure, and relative humidity; the set of result nodes includes: full-load power output.
[0007] More preferably, in S2, the process of constructing the topological structure of the fuzzy cognitive graph includes: S21. Construct the connection relationship between any two different cause nodes based on the set of cause nodes; S22. Establish a one-way connection from each cause node to the result node.
[0008] More preferably, in S3, the parameterized representation of the connection weights includes: ; In the formula, This represents the normalized activation value of the source node; Indicates the order of the spline; Indicates the number of control points; Indicates the control point coefficient; This indicates the calculation of the first term using the Cox-de-Boor recursive formula. One basis function.
[0009] More preferably, in S3, the loss function of the output power prediction model includes: ; In the formula, Indicates prediction error; Indicates reconstruction error; The weighting coefficients representing the causal reconstruction loss; This represents the L2 regularization coefficient.
[0010] More preferably, the constraints for constructing the topology of the fuzzy cognitive graph include: not establishing feedback connections from result nodes to cause nodes; not establishing self-loop connections within nodes themselves; and in S21, not establishing connections from internal state variables to external environmental meteorological variables.
[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention innovatively introduces the concept of Kolmogorov-Arnold Network (KAN), replacing the fixed scalar weights on the connecting edges of traditional fuzzy cognitive graphs with learnable B-spline functions, thereby constructing a "spline fuzzy cognitive graph." The B-spline function is used to nonlinearly fit the complex mapping relationships between "cause nodes" and "from cause to effect nodes," and the control point parameters of the spline function are optimized through training. Finally, the trained spline fuzzy cognitive graph is used for high-precision regression prediction of the output power of a power plant, and outputs the B-spline fitting function image of the coupling relationships between nodes, thus achieving interpretability and visualization of the prediction process. Attached Figure Description
[0012] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1 A schematic diagram of the overall process of a combined cycle power plant output power prediction method based on spline fuzzy cognitive graph provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the spline fuzzy cognitive graph topology provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a B-spline function provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a combined cycle power plant structure provided in an embodiment of the present invention; Figure 5 A visualization of the B-spline fitting function of "ambient temperature (AT)" versus "output power (PE)" provided in an embodiment of the present invention; Figure 6 A comparison chart of predicted and actual output power of a combined cycle power plant provided in an embodiment of the present invention. Detailed Implementation
[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0016] Example 1: like Figure 1 As shown, this embodiment provides a method for predicting the output power of a combined cycle power plant based on spline fuzzy cognitive graphs, including the following steps: S1, collecting historical operating data of the combined cycle power plant under full load operation, and obtaining a set of cause nodes and a set of result nodes based on the historical operating data.
[0017] Combination Figure 4 The provided schematic diagram of a combined cycle power plant shows that it mainly includes a gas turbine, a waste heat boiler, and a steam turbine. The physical characteristics of a combined cycle power plant mean that external environmental factors directly affect the thermodynamic efficiency of the gas turbine and the entire system. Therefore, this embodiment collected 9568 data samples from six years of historical operation data of the combined cycle power plant under full load conditions. The data samples were standardized to ensure their numerical distribution fits the effective domain of the B-spline function. Four environmental state variables were selected as inputs and defined as the set of cause nodes. The ambient temperature is AT, the exhaust vacuum is V, the ambient pressure is AP, and the relative humidity is RH. Target definition: The net energy output per hour is selected as the prediction target, defined as the set of result nodes. Then, normalization was performed to fit the domain of the B-spline function and accelerate model convergence. The dataset was then divided into training, validation, and test sets in an 8:1:1 ratio.
[0018] S2. Construct a spline fuzzy cognitive graph topology structure for the set of cause nodes and the set of result nodes.
[0019] like Figure 2 The diagram shown is a schematic representation of the spline fuzzy cognitive graph topology in this embodiment. A five-node network model (4 cause nodes and 1 result node) is constructed based on the physical causal relationships of the combined cycle power plant.
[0020] S21. Construct a connection relationship between any two distinct cause nodes based on the set of cause nodes; that is, any two distinct cause nodes... and There are unidirectional or bidirectional connections between them, which are used to capture the coupling relationships between factors affecting the power generation factor of a power plant.
[0021] S22. Establish a one-way connection from each cause node to the result node, and establish a connection from each cause node... Point to the result node The unidirectional connection is used to directly fit the nonlinear driving effect of environmental factors on power generation.
[0022] The constraints include: to reduce computational complexity and prevent logical infinite loops, feedback connections from result nodes to cause nodes are not established, nor are self-loop connections between nodes (including cause and result nodes). In S21, considering physical causal constraints, connections from internal state variables (exhaust vacuum level V) to external environmental meteorological variables (AT, AP, RH) are not established. After excluding self-loops, a total of 9 cause-effect connection edges are established. Figure 2 As shown by the curved arrows, there are a total of 13 directed edges. Each connection is no longer represented by a traditional single fixed value (scalar), but by an independent learnable function. express.
[0023] S3. The connection weights of the spline fuzzy cognitive graph topology are parameterized to obtain the output power prediction model of the combined cycle power plant.
[0024] Abandoning connection weights in traditional fuzzy cognitive graphs To maintain a fixed scalar value, this embodiment introduces the core idea of Kolmogorov-Arnold Networks (KANs), defining the weight of each connection as a learnable univariate nonlinear function. This function is parameterized using B-spline basis functions: ; In the formula, This represents the normalized activation value of the source node; Indicates the order of the spline; Indicates the number of control points; Indicates the control point coefficient; This indicates the calculation of the first term using the Cox-de-Boor recursive formula. One basis function.
[0025] in, ; In the formula, It is a node vector.
[0026] like Figure 3 As shown, B-spline function parameters are configured for each connecting edge in the spline fuzzy cognitive graph topology. Figure 3 As shown by the solid red dots, in this embodiment, 12 learnable control points are set for each edge. A 3rd-order B-spline function is used as the parameter, and the domain range is set to [-3.0, 3.0]. Weight initialization: To ensure stability in the early stages of training, the least squares method is used to fit the initial shape of all spline functions to a near-linear function. ,in, The slope of the near-linear initialization function, i.e., the linear weighting coefficient, is used to control the spline function during the initialization phase as it changes with the input. The intensity of the changing linear response; Indicates the first The spatial centroid coordinates of a B-spline basis function within its domain, and their relationship to the input variables of the B-spline function. The meanings are different. A small amount of random Gaussian noise (standard deviation 0.1) is added to break the parameter symmetry, facilitating gradient descent optimization. Figure 3 In the center, the red solid dots represent learnable control points. The black curve represents the B-spline basis function. White dots represent node vectors. The node positions in the diagram. and These are the left and right endpoints of the domain, respectively. Input variables The valid domain; , This indicates that the endpoint node is set repeatedly. "Indicates that the endpoint node is repeated" Second-rate. , Let these represent the first and last B-spline basis functions, respectively. , These represent the corresponding control points. (See diagram) This represents the normalized activation value of the source node and is the input variable of the B-spline function; Indicates the first The spatial centroid coordinates of the B-spline basis functions within their domain are used for control point initialization. This is achieved by initializing each control point... Training updates can extend the edge weights from the traditional fixed scalar form to a nonlinear mapping function that varies with the input variables.
[0027] S4. Based on the output power prediction model, process the real-time operating data to obtain the power prediction results, and draw the spline function curves of each side to visualize the nonlinear relationship.
[0028] Model training is performed using the AdamW optimizer, with forward inference: normalized cause node data... Enter the network. (Through) Figure 3 The B-spline functions on each connecting edge shown are used to calculate the power prediction value and the cause reconstruction value. The sum of all cause nodes pointing to the result node after their respective spline function mappings is calculated to obtain the power prediction value. Among them, subscript c Indicates the cause node reconstruction task; Indicates the cause node Point to output power node The corresponding connecting edge Spline mapping function. For each cause node. Calculate the sum of the effects of other cause nodes to obtain the cause reconstruction value: Multi-task learning is performed using a composite loss function: ;in, Indicates prediction error; Indicates reconstruction error; The weighting coefficients of the cause reconstruction loss are used to enhance the model's understanding of the dependencies between environmental factors; This is the L2 regularization coefficient, applied to the spline control point weights to prevent overfitting and function oscillations. The control point coefficients of all B-spline functions are iteratively updated using a gradient descent optimization algorithm. This allows the model to accurately reflect the physical constraints between environmental parameters while fitting the output power.
[0029] like Figure 5 The figure shown is a visualization of the B-spline fitting function of "ambient temperature (AT)" on "output power (PE)". Figure 5 The solid line represents the final learned nonlinear mapping relationship, and the dashed line represents the combination of various basis functions. From... Figure 5 It is evident that the relationship between ambient temperature and output power is not a simple linear one; its contribution to power varies dynamically across different temperature ranges (e.g., exhibiting distinct peaks or troughs). This visualization effectively breaks the "black box" nature of traditional deep learning neural networks, endowing the model with strong physical interpretability, enabling power plant operators to intuitively understand the specific impact trends of changes in environmental parameters on power generation. Model performance was evaluated on an independent test set. The prediction results were inversely normalized back to the original physical units (MW), and the root mean square error (RMSE) and coefficient of determination were calculated. ).like Figure 6 As shown, in this example, the output power prediction model on the test set... The value exceeds 0.94, and the RMSE under the original dimensions is 4.07, which proves the effectiveness of the method proposed in this invention in accurately predicting the output power of power plants and capturing the complex nonlinear characteristics of power plants.
[0030] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.
Claims
1. A method for predicting the output power of a combined cycle power plant based on spline fuzzy cognitive graphs, characterized in that, Includes the following steps: S1. Collect historical operating data of the combined cycle power plant under full load operation, and obtain the set of cause nodes and the set of result nodes based on the historical operating data; S2. Construct a spline fuzzy cognitive graph topology for the set of cause nodes and the set of result nodes; S3. The connection weights of the spline fuzzy cognitive graph topology are parameterized to obtain the output power prediction model of the combined cycle power plant; S4. Based on the output power prediction model, process the real-time operating data to obtain the power prediction results, and draw the spline function curves of each side to visualize the nonlinear relationship.
2. The method for predicting the output power of a combined cycle power plant based on spline fuzzy cognitive graphs according to claim 1, characterized in that, The set of cause nodes includes: ambient temperature, exhaust vacuum, ambient pressure, and relative humidity; the set of result nodes includes: full-load power output.
3. The method for predicting the output power of a combined cycle power plant based on spline fuzzy cognitive graphs according to claim 1, characterized in that, In S2, the process of constructing the topological structure of the fuzzy cognitive graph includes: S21. Construct the connection relationship between any two different cause nodes based on the set of cause nodes; S22. Establish a one-way connection from each cause node to the result node.
4. The method for predicting the output power of a combined cycle power plant based on spline fuzzy cognitive graphs according to claim 1, characterized in that, In S3, the parameterized representation of the connection weights includes: ； In the formula, This represents the normalized activation value of the source node; Indicates the order of the spline; Indicates the number of control points; Indicates the control point coefficient; This indicates the calculation of the first term using the Cox-de-Boor recursive formula. One basis function.
5. The method for predicting the output power of a combined cycle power plant based on spline fuzzy cognitive graphs according to claim 4, characterized in that, In S3, the loss function of the output power prediction model includes: ； In the formula, Indicates prediction error; Indicates reconstruction error; The weighting coefficients representing the causal reconstruction loss; This represents the L2 regularization coefficient.
6. The method for predicting the output power of a combined cycle power plant based on spline fuzzy cognitive graphs according to claim 1, characterized in that, The constraints for constructing the topology of the fuzzy cognitive graph include: not establishing feedback connections from result nodes to cause nodes; not establishing self-loop connections within nodes; and in S21, not establishing connections from internal state variables to external environmental meteorological variables.