Method for constructing characteristic curve of water turbine under full guide vane opening
By using cyclic iteration and quantitative evaluation of a BP neural network, the sample data was expanded, and the characteristic curve under large guide vane opening was optimized. This solved the problem of insufficient data for the turbine model under large guide vane opening, realized the construction of characteristic curves under full guide vane opening, and improved the accuracy of hydropower station transient process calculation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-23
- Publication Date
- 2026-03-31
AI Technical Summary
The characteristic curves of existing turbine models lack data under large guide vane opening, making it difficult to meet the comprehensive needs of hydropower station transient process calculations. Neural network construction methods lack samples, making it difficult to extract key features.
By iteratively evaluating and quantizing the BP neural network, the sample data is expanded, the characteristic curve under large guide vane opening is optimized, and the characteristic curve under full guide vane opening is constructed by combining interpolation fitting and cubic B-spline smooth connection.
It achieves reasonable expansion of the characteristic curve under large guide vane opening and accurate prediction under small guide vane opening, improves the data volume and feature extraction capability of the neural network, and ensures that the output characteristic curve is more reasonable and accurate.
Smart Images
Figure CN121765868A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of water turbine units, specifically relating to a method for constructing characteristic curves of a water turbine under full guide vane opening. Background Technology
[0002] With the large-scale development and utilization of hydropower energy in my country, the demand for research on the safe and stable operation of hydropower stations is becoming increasingly urgent. In the actual operation of hydropower stations, the occurrence of operating condition switching and transient processes is very common. The core of hydropower station transient process calculation lies in the modeling and simulation of the hydraulic system, and a complete full characteristic curve of the turbine is an important foundation for unit simulation and calculation. However, the comprehensive characteristic curves of the turbine models provided by manufacturers usually only cover the characteristics near the high-efficiency region and the no-load runaway characteristics under large-opening conditions, which is insufficient to meet the comprehensive needs of transient process calculation. Therefore, it is necessary to construct the characteristic curves of the turbine under full guide vane opening.
[0003] Currently, the construction methods for turbine characteristic curves include using neural networks. For example, application number 2024108089661 discloses a method for extending the comprehensive characteristic curve of a turbine model based on a genetic algorithm-optimized BP neural network. However, the disadvantage of neural network construction is that it lacks a large number of samples, making it difficult to meet the requirements of neural network extraction of key features. The only effective solution that can be thought of is to increase the sample size. The above method does not indicate how to expand the sample size, but only inputs the expanded high-efficiency region data and the expanded runaway characteristic data into the neural network training model to construct the characteristic curve.
[0004] To address this, this proposal suggests a method for constructing the characteristic curve of a turbine at full guide vane opening. The aim is to generate a large number of reasonable samples through iterative BP neural network combined with quantitative evaluation, thereby optimizing the turbine characteristic curve at large guide vane opening. This allows the BP neural network to extract more key features when extending to the turbine characteristic curve at small guide vane opening. Summary of the Invention
[0005] This invention proposes a method for constructing characteristic curves of a water turbine under full guide vane opening. The method first fits the relationship between unit runaway speed and unit runaway flow rate under full guide vane opening, then constructs samples using a neural network, expands the samples using the neural network, optimizes the sample data after expansion, and then iterates again using the neural network to finally output the relationship between unit speed and unit flow rate under large guide vane opening. Finally, based on the unit runaway speed and unit runaway flow rate under full guide vane opening, the characteristic curve under full guide vane opening is finally fitted.
[0006] To achieve the above objectives, the present invention proposes the following technical content; A method for constructing characteristic curves of a water turbine at full guide vane opening includes the following steps: S1: Obtain the runaway point data of the turbine's large guide vane opening, and decompose it into a correspondence diagram of the large guide vane opening and the runaway point unit flow rate, and a correspondence diagram of the large guide vane opening and the runaway point unit speed; in the correspondence diagram of the large guide vane opening and the runaway point unit flow rate, interpolate to obtain the runaway point flow rate of the small guide vane opening, and fit to obtain the correspondence diagram of the full guide vane opening and the runaway point unit flow rate; in the correspondence diagram of the large guide vane opening and the runaway point unit speed, interpolate to obtain the runaway point speed of the small guide vane opening, fit the runaway point speed of the small guide vane opening and the runaway point speed of the large guide vane opening respectively, and then smoothly connect them to obtain the correspondence diagram of the full guide vane opening and the runaway point unit speed; S2: Combine the graphs showing the correspondence between the full guide vane opening and the unit flow rate at the escape point and the graphs showing the correspondence between the full guide vane opening and the unit speed at the escape point to obtain the correspondence between the unit speed at the escape point and the unit flow rate at the escape point when the guide vane is fully open. S3: Group by guide vane opening, read the high-efficiency zone characteristic data at large guide vane opening using image reading software, and then construct a sample set for the BP neural network based on the runaway point data obtained in S2. Fill the sample set into the sample matrix, input the sample matrix into the trained BP neural network, and the BP neural network outputs the predicted data for different guide vane openings according to the set rules. S4: For any guide vane opening, evaluate the predicted data under that guide vane opening, and divide the evaluated data into "excellent" and "poor" predicted data. S5: Replace the data in the sample matrix corresponding to the guide vane opening with the predicted data evaluated as "excellent"; do not replace the data in the sample matrix P corresponding to the guide vane opening with the predicted data evaluated as "poor" to obtain the updated sample matrix; then re-input the updated sample matrix into the BP neural network, and the BP neural network will output the corresponding predicted data again. Each time it outputs, the data in each row of the sample matrix will change. Repeat S4-S5 until the predicted data for each guide vane opening is evaluated as "excellent", and output the final sample matrix. S6: Final sample matrix The unit speed and unit flow rate at the runaway point under the small guide vane opening obtained in S2 are input into another trained BP neural network, which is based on the final sample matrix. Collect the characteristic curve variation feature vector under large guide vane opening, and then output the characteristic curve under full guide vane opening based on the runaway point unit speed and runaway point unit flow rate already obtained in S2.
[0007] Further, step S4 includes the following steps: S4.1: For guide vane opening of... g The predicted data at that time is used to calculate the absolute error between the neural network's predicted data and the data in the original sample:
[0008] In the formula, Indicates the guide vane opening is g At that time, the first i The predicted unit flow rate at the first adapted unit speed and the first i The absolute error of the actual unit flow rate at a unit speed of adaptation; Indicates the guide vane opening is g At that time, the first i Predicted unit flow rate at a unit speed of adaptation; Indicates the guide vane opening is g At that time, the first i The actual unit flow rate at a unit speed that is adapted to the unit speed; S4.2: Calculate the mean square error between the predicted data and the data in the original sample;
[0009] In the formula, n Indicates the amount of data to be adapted; S4.3: Calculate the monotonicity of the data predicted by the neural network under this guide vane opening;
[0010] In the formula, This indicates that when the guide vane opening is g, the neural network predicts the first... x Unit flow rate; This indicates that when the guide vane opening is g, the neural network predicts the first... x+ 1 unit flow rate; x ∈[1,27] and is an integer; S4.4: Constraints on low-speed and high-speed regions; when hour,
[0011] when hour,
[0012] S4.5: At this guide vane opening g If , and If all conditions are met and the speed range constraint of S4.4 is satisfied, then the predicted data under this guide vane opening is "excellent"; otherwise, it is "poor".
[0013] Furthermore, in step S4, since the unit flow rate of the turbine can remain basically balanced and changes little at small unit speeds, a small rate of change is designed. At large unit speeds, the flow rate decreases sharply as the speed increases, therefore a larger rate of change is designed. Designed for -9, Designed for -50.
[0014] Furthermore, in step S1, the runaway speeds of the small guide vane opening and the runaway speeds of the large guide vane opening are fitted separately and then smoothly connected using cubic B-splines.
[0015] Furthermore, in step S1, the correspondence between the full guide vane opening and the unit flow rate at the runaway point is as follows:
[0016] In the formula, and For coefficients; The fitting formula for the runaway speed at large guide vane opening is:
[0017] In the formula, , , , , For the corresponding coefficients; The fitting formula for the runaway speed of the small guide vane opening is:
[0018] In the formula, and For the corresponding coefficient.
[0019] The beneficial effects that can be achieved by adopting the above technologies are: 1. After neural network prediction, a quantitative evaluation method is used to evaluate the predicted data output by the neural network at each guide vane opening. Data evaluated as "excellent" replaces the data in the original sample matrix. In terms of sample size, the original sample matrix is limited to data in the high-efficiency range, such as 60-85 unit speed, with only 25 data points. However, this solution expands the data to 0-135 unit speed by using neural network prediction expansion, and samples are taken every 5 unit speeds, which can obtain 28 data points. The increase in data volume facilitates the subsequent extraction of key features by the neural network.
[0020] 2. This scheme adopts a cyclic iterative approach to ensure that each feature curve of the large guide vane opening is "excellent" data. Then, the neural network extracts key features and outputs the feature curve of the small guide vane opening. This achieves the purpose of first optimizing the feature data of the large guide vane opening and then predicting and outputting the feature data of the small guide vane opening, rather than directly using the original feature curve of the large guide vane opening to predict. Under this cyclic iterative design, the output full guide vane opening feature curve is more reasonable.
[0021] 3. The flyaway unit speed of the small guide vane opening is not directly used for interpolation fitting. Instead, the large guide vane opening is first fitted with the flyaway unit speed, and the small guide vane opening is fitted with the flyaway unit speed, respectively. Then, the two fitted curves are smoothly connected by cubic B-splines. This makes the flyaway unit speed of the small guide vane opening more accurate, which is more accurate when the neural network builds samples in the future. Attached Figure Description
[0022] Figure 1 This is a flowchart of the steps in this solution; Figure 2 These are the runaway point data at the large guide vane opening. Figure 3 This is a graph showing the relationship between guide vane opening and unit flow rate at the runaway point when the guide vane is fully open; Figure 4 This is a graph showing the relationship between the guide vane opening and the unit rotational speed at the runaway point when the guide vanes are fully open. Figure 5 These are the runaway point data at full guide vane opening; Figure 6 These are characteristic data and runaway point data graphs for the high-efficiency region at large guide vane opening. Figure 7 This is a comparison chart of the characteristic curves when they are excellent and poor under different guide vane openings; Figure 8 This is a graph of characteristic data when the guide vane opening is large after the final prediction by the BP neural network. Figure 9 This is the flow characteristic curve output by the BP neural network at full guide vane opening. Figure 10 This is the torque characteristic curve output by the BP neural network at full guide vane opening. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions 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, 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.
[0024] like Figure 1 The method for constructing the characteristic curve of a water turbine at full guide vane opening includes the following steps: S1: Obtain the runaway point data of the turbine's large guide vane opening, and decompose it into a graph showing the correspondence between the large guide vane opening and the runaway point per unit flow rate, and a graph showing the correspondence between the large guide vane opening and the runaway point per unit speed.
[0025] Specifically, assuming the turbine model used on-site is HLA1015a, the runaway point data at large guide vane opening is obtained using graph reading software based on the turbine efficiency and runaway test curves provided by the manufacturer. See details below. Figure 2 . Figure 2 The data on the runaway point at medium and large guide vane openings, that is, the corresponding relationship between the runaway point unit speed and the runaway point unit flow rate at the corresponding guide vane openings; the large guide vane openings obtained are 6°, 10°, 14°, 16°, 18°, 20°, 22°, 24°, 26°, 28° and 30°.
[0026] Based on the established boundary condition: when the guide vane opening is 0°, the unit flow rate is 0 regardless of the unit rotational speed. Figure 2 After disassembly, a graph showing the relationship between the large guide vane opening and the unit flow rate at the runaway point was obtained. From this graph, an interpolation fitting method was used to obtain a graph showing the relationship between the full guide vane opening and the unit flow rate at the runaway point. (See...) Figure 3 Similarly, Figure 2 After disassembly, a graph showing the correspondence between the large guide vane opening and the unit speed at the runaway point was obtained. In this graph, the relationship between the small guide vane opening and the unit speed at the runaway point was first obtained through interpolation. Then, the relationships between the small guide vane opening and the unit speed at the runaway point, and between the large guide vane opening and the unit speed at the runaway point, were fitted separately. Finally, the two fitted curves were smoothly connected using a cubic B-spline to obtain a graph showing the correspondence between the full guide vane opening and the unit speed at the runaway point. (See...) Figure 4 . Figure 3 and Figure 4 In the diagram, red dots represent data points obtained from the graph, while black dots represent interpolation points.
[0027] in, Figure 3 In the diagram, the relationship between the full guide vane opening and the unit flow rate at the escape point is as follows:
[0028] In the formula, a Indicates the guide vane opening. Q 11c This represents the unit flow rate at the escape point.
[0029] Figure 4 In the middle section, the formula for individually fitting the guide vane opening and the unit rotational speed at the runaway point at the large opening is:
[0030] In the formula, a Indicates the guide vane opening; N 11c This indicates the unit rotational speed at the escape point; The formula for fitting the guide vane opening and unit rotational speed at small opening is as follows:
[0031] In the formula, a Indicates the degree of openness; N 11c This indicates the unit rotational speed at the escape point; After smoothly connecting the two fitted curves using a cubic B-spline, we obtain... Figure 4 .
[0032] S2: Will Figure 3 and Figure 4 By combining these parameters, we obtain the corresponding relationship between the unit rotational speed and unit flow rate at the runaway point when the guide vane is fully open. Figure 5 . refer to Figure 5 It is possible to obtain the corresponding relationship between the unit speed at the runaway point and the unit flow rate at the runaway point under the guide vane opening.
[0033] S3: Group by guide vane opening, read the high-efficiency zone characteristic data at large guide vane opening using image reading software, and then construct a sample set for the BP neural network based on the runaway point data obtained in S2. Fill the sample set into the sample matrix, input the sample matrix into the trained BP neural network, and the BP neural network outputs the predicted data for different guide vane openings according to the set rules.
[0034] Specifically, the high-efficiency zone characteristic data is expressed as: the correspondence between unit speed and unit flow rate when the unit speed is in the range of 60-85; or as: the correspondence between unit speed and unit torque when the unit speed is in the range of 60-85.
[0035] During the data reading process, the software reads the large guide vane opening, unit speed, unit flow rate, and unit efficiency in the high-efficiency zone, and calculates the unit torque using an empirical formula: In the formula, M 11 Indicates unit torque; Q 11 Indicates unit flow rate; N 11 Indicates unit rotational speed; Indicates efficiency.
[0036] After obtaining the above data, two different sample sets were established. In one sample set, each sample includes guide vane opening, unit rotational speed, and unit flow rate; in the other sample set, each sample includes guide vane opening, unit rotational speed, and unit torque. These two different sample sets can be used to construct the corresponding characteristic curves under full guide vane opening, namely: the relationship between unit rotational speed and unit flow rate under full guide vane opening; and the relationship between unit rotational speed and unit torque under full guide vane opening.
[0037] The construction process for the two diagrams at full opening is exactly the same. To simplify the explanation, this method only uses the construction of the diagram of unit rotational speed versus unit flow rate at full guide vane opening as an example.
[0038] Specifically, the following steps are included: S3.1: Obtain the characteristic data of the high-efficiency region under large guide vane opening and the runaway point data obtained in S2, and construct a sample matrix.
[0039] See the characteristic data and runaway point data for the high-efficiency region. Figure 6 .
[0040] Specifically, each sample is constructed as follows: (guide vane opening, unit rotational speed in the high-efficiency zone, unit flow rate in the high-efficiency zone) or (guide vane opening, unit rotational speed at the escape point, unit flow rate at the escape point).
[0041] Taking a guide vane opening of 30° as an example, the sample that can be established is: , This indicates that when the guide vane opening is 30°, the first m Unit rotational speed in the high-efficiency zone; This indicates that when the guide vane opening is 30°, the first m Unit flow rate in each high-efficiency zone; This indicates the unit rotational speed of the flyback when the guide vane opening is 30°; This indicates the unit flow rate of the runoff when the guide vane opening is 30°.
[0042] Taking a guide vane opening of 28° as another example, the sample that can be established is: And so on... Different guide vane openings can establish corresponding samples. All samples constitute the sample set. The sample set is then filled into the sample matrix, and each row of the sample matrix represents a sample at the corresponding guide vane opening.
[0043] The sample matrix P has the following format:
[0044] The above sample matrix P The input is fed into the trained BP neural network, and the parameters of the BP neural network are set as follows: Maximum number of iterations: 2000; Target error: 1e-10; Learning rate: 0.001; S3.2: Sample Matrix P After being input into the BP neural network, the BP neural network outputs data according to the set rules, that is, it outputs the predicted data of the unit flow rate under the complete unit speed.
[0045] Specifically, the BP neural network, according to a set rule, outputs the unit flow rate corresponding to the unit rotational speed within the range of 0-135 unit rotational speeds (i.e., a complete unit rotational speed), with a unit rotational speed difference of 5. Taking a guide vane opening of 30° as an example, at this guide vane opening, the neural network outputs the unit flow rate corresponding to unit rotational speeds of 0, 5, 10, 15, ..., 135; data sampling for other guide vane openings is output in the same way.
[0046] S4: For any guide vane opening, evaluate the predicted data at that opening. This includes the following steps: S4.1: For guide vane opening of... g The predicted data at that time is used to calculate the absolute error between the data predicted by the neural network and the data in the original sample.
[0047]
[0048] In the formula, Indicates the guide vane opening is g At that time, the first i The predicted unit flow rate at the first adapted unit speed and the first i The absolute error of the actual unit flow rate at a unit speed of adaptation; Indicates the guide vane opening is g At that time, the first i The predicted unit flow rate at a unit speed that is adapted to the unit rotational speed. i ∈[1,6] and are integers, the first adapted unit speed represents a unit speed of 60, the second adapted unit speed represents a unit speed of 65, and so on...; Indicates the guide vane opening is g At that time, the first i The actual unit flow rate at a unit rotational speed, i.e., the unit flow rate in the sample in S3.1.
[0049] S4.2: Calculate the mean square error between the predicted data and the data in the original sample.
[0050]
[0051] In the formula, n This indicates the number of data items adapted, i.e.n =6.
[0052] S4.3: Calculate the monotonicity of the data predicted by the neural network under this guide vane opening.
[0053]
[0054] In the formula, This indicates that when the guide vane opening is g, the neural network predicts the first... x Unit flow rate; This indicates that when the guide vane opening is g, the neural network predicts the first... x+ 1 unit flow rate; x ∈[1,27] and is an integer; when x When =1, This represents the predicted unit flow rate when the guide vane opening is g and the unit rotational speed is 0; when x When =2, This represents the predicted unit flow rate when the guide vane opening is g and the unit rotational speed is 5; and so on... S4.4: Constraints are applied to the low-speed and high-speed ranges.
[0055] when hour, ; when hour,
[0056] The reason for adopting and This is because at low unit speeds, the unit flow rate of the turbine can remain basically balanced with minimal change, so a smaller rate of change is designed; at high unit speeds, the flow rate drops sharply as the speed increases, so a larger rate of change is designed.
[0057] S4.5: At this guide vane opening g If , and If all conditions are met and the speed range constraint of S4.4 is satisfied, then the predicted data under this guide vane opening is "excellent"; otherwise, it is "poor". R and K These are the set thresholds. Figure 7 The excellent curves are shown for judging the guide vane opening at 14° and 30° respectively.
[0058] S5: Replace the data in sample matrix P that is rated "excellent" for a certain guide vane opening in S4 with the data of the corresponding guide vane opening; do not replace the data of the corresponding guide vane opening in sample matrix P with the predicted data rated "poor". This will result in an updated sample matrix. Then, re-input the updated sample matrix into the BP neural network, and the BP neural network will output the corresponding predicted data again. Each time the data in each row of the sample matrix changes during the output, repeating S4-S5 until the predicted data for each guide vane opening is rated "excellent", and then output the final sample matrix.
[0059] Specifically, assuming that in the first evaluation, the predicted data is rated as "excellent" only when the guide vane opening is 30°, and the data at other guide vane openings are rated as "poor"; then the updated sample matrix P 新 for:
[0060] After loops S4-S5, until the predicted data for each guide vane opening is evaluated as "excellent", the final output sample matrix is:
[0061] When the final sample matrix is obtained At that time, characteristic curves for unit rotational speed and unit flow rate under large guide vane opening were constructed; such as Figure 8 .
[0062] S6: Final sample matrix The unit speed and unit flow rate at the runaway point under the small guide vane opening obtained in S2 are input into another trained BP neural network, which is based on the final sample matrix. The characteristic curve under large guide vane opening is collected as a feature vector. Then, based on the runaway point unit speed and runaway point unit flow rate already obtained in S2 under small guide vane opening, the BP neural network can output the characteristic curve under full guide vane opening, such as... Figure 9 .
[0063] Similarly, another BP neural network can output characteristic curves of unit rotational speed and unit torque at full guide vane opening, such as... Figure 10 .
[0064] The core design of this solution lies in: 1. After neural network prediction, a quantitative evaluation method is used to evaluate the predicted data output by the neural network at each guide vane opening. Data evaluated as "excellent" replaces the data in the original sample matrix. In terms of sample size, the original sample matrix is limited to data in the high-efficiency range, such as 60-85 unit speed, with only 25 data points. However, this solution expands the data to 0-135 unit speed by using neural network prediction expansion, and samples are taken every 5 unit speeds, which can obtain 28 data points. The increase in data volume facilitates the subsequent extraction of key features by the neural network.
[0065] 2. This scheme adopts a cyclic iterative approach to ensure that each feature curve of the large guide vane opening is "excellent" data. Then, the neural network extracts key features and outputs the feature curve of the small guide vane opening. This achieves the purpose of first optimizing the feature data of the large guide vane opening and then predicting and outputting the feature data of the small guide vane opening, rather than directly using the original feature curve of the large guide vane opening to predict. Under this cyclic iterative design, the output full guide vane opening feature curve is more reasonable.
[0066] 3. The flyaway unit speed of the small guide vane opening is not directly used for interpolation fitting. Instead, the large guide vane opening is first fitted with the flyaway unit speed, and the small guide vane opening is fitted with the flyaway unit speed, respectively. Then, the two fitted curves are smoothly connected by cubic B-splines. This makes the flyaway unit speed of the small guide vane opening more accurate, which is more accurate when the neural network builds samples in the future.
[0067] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.
Claims
1. A method for constructing characteristic curves of a water turbine at full guide vane opening, characterized in that, Includes the following steps: S1: Obtain the runaway point data of the turbine's large guide vane opening, and decompose it into a correspondence diagram of the large guide vane opening and the runaway point unit flow rate, and a correspondence diagram of the large guide vane opening and the runaway point unit speed; in the correspondence diagram of the large guide vane opening and the runaway point unit flow rate, interpolate to obtain the runaway point flow rate of the small guide vane opening, and fit to obtain the correspondence diagram of the full guide vane opening and the runaway point unit flow rate; In the graph showing the relationship between the large guide vane opening and the unit speed of the flyaway point, the flyaway point speed of the small guide vane opening is obtained by interpolation. The flyaway point speeds of the small guide vane opening and the large guide vane opening are fitted separately and then smoothly connected to obtain the graph showing the relationship between the full guide vane opening and the unit speed of the flyaway point. S2: Combine the graphs showing the correspondence between the full guide vane opening and the unit flow rate at the escape point and the graphs showing the correspondence between the full guide vane opening and the unit speed at the escape point to obtain the correspondence between the unit speed at the escape point and the unit flow rate at the escape point when the guide vane is fully open. S3: Group by guide vane opening, read the high-efficiency zone characteristic data at large guide vane opening using image reading software, and then construct a sample set for the BP neural network based on the runaway point data obtained in S2. Fill the sample set into the sample matrix, input the sample matrix into the trained BP neural network, and the BP neural network outputs the predicted data for different guide vane openings according to the set rules. S4: For any guide vane opening, evaluate the predicted data under that guide vane opening, and divide the evaluated data into "excellent" and "poor" predicted data. S5: Replace the data in the sample matrix corresponding to the guide vane opening with the predicted data evaluated as "excellent"; do not replace the data in the sample matrix P corresponding to the guide vane opening with the predicted data evaluated as "poor" to obtain the updated sample matrix; then re-input the updated sample matrix into the BP neural network, and the BP neural network will output the corresponding predicted data again. Each time it outputs, the data in each row of the sample matrix will change. Repeat S4-S5 until the predicted data for each guide vane opening is evaluated as "excellent", and output the final sample matrix. S6: Final sample matrix The unit speed and unit flow rate at the runaway point under the small guide vane opening obtained in S2 are input into another trained BP neural network, which is based on the final sample matrix. Collect the characteristic curve variation feature vector under large guide vane opening, and then output the characteristic curve under full guide vane opening based on the runaway point unit speed and runaway point unit flow rate already obtained in S2.
2. The method for constructing a characteristic curve of a water turbine at full guide vane opening according to claim 1, characterized in that, Step S4 includes the following steps: S4.1: For guide vane opening of... g The predicted data at that time is used to calculate the absolute error between the neural network's predicted data and the data in the original sample: ； In the formula, Indicates the guide vane opening is g At that time, the first i The predicted unit flow rate at the first adapted unit speed and the first i The absolute error of the actual unit flow rate at a unit speed of adaptation; Indicates the guide vane opening is g At that time, the first i Predicted unit flow rate at a unit speed of adaptation; Indicates the guide vane opening is g At that time, the first i The actual unit flow rate at a unit speed that is adapted to the unit speed; S4.2: Calculate the mean square error between the predicted data and the data in the original sample; ； In the formula, n Indicates the amount of data to be adapted; S4.3: Calculate the monotonicity of the data predicted by the neural network under this guide vane opening; ； In the formula, This indicates that when the guide vane opening is g, the neural network predicts the first... x Unit flow rate; This indicates that when the guide vane opening is g, the neural network predicts the first... x+ 1 unit flow rate; x ∈[1,27] and is an integer; S4.4: Constraints on low-speed and high-speed regions; when hour, ; when hour, ; S4.5: At this guide vane opening g If , and If all conditions are met and the speed range constraint of S4.4 is satisfied, then the predicted data under this guide vane opening is "excellent"; otherwise, it is "poor".
3. The method for constructing a characteristic curve of a water turbine at full guide vane opening according to claim 2, characterized in that, In step S4, since the unit flow rate of the turbine can remain basically balanced and changes little at small unit speeds, a small rate of change is designed. At large unit speeds, the flow rate drops sharply as the speed increases, so a larger rate of change is designed. Designed for -9, Designed for -50.
4. The method for constructing the characteristic curve of a water turbine at full guide vane opening according to claim 2, characterized in that, In step S1, the runaway speeds of the small guide vane opening and the large guide vane opening are fitted separately and then smoothly connected using cubic B-splines.
5. The method for constructing the characteristic curve of a water turbine at full guide vane opening according to claim 1, characterized in that, In step S1, the correspondence between the full guide vane opening and the unit flow rate at the runaway point is as follows: ； In the formula, and For coefficients; The fitting formula for the runaway speed at large guide vane opening is: ； In the formula, , , , , For the corresponding coefficients; The fitting formula for the runaway speed of the small guide vane opening is: ； In the formula, and For the corresponding coefficient.