Water turbine modeling method combining mucous algorithm and BP neural network

Abstract

The invention discloses a water turbine modeling method combining a mucus algorithm and a BP neural network. The water turbine modeling method comprises the steps of obtaining test data of a prototype water turbine; perfecting opening characteristic and efficiency characteristic data; mixing and recombining the improved opening characteristic and efficiency characteristic data to obtain flow characteristic and torque characteristic sample data; combining the complete moment characteristic data with a BP (Back Propagation) neural network to construct an opening characteristic neural network model taking the unit rotating speed and the unit moment as input and the guide vane opening as output; correcting the guide vane opening in the test data of the water turbine based on a mucus algorithm; combining the corrected water turbine test data with a BP neural network to construct a water turbine flow characteristic and torque characteristic neural network model; the problem of limitation of a water turbine modeling method in the prior art is solved, so that the subsequent stability of the hydroelectric generating set and the reliability of optimization control research are ensured, and long-term efficient and stable operation of the hydroelectric generating set is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of the safe and stable operation of hydropower units, and particularly relates to a method for modeling a water turbine by combining a slime mold algorithm and a BP neural network. Background Art

[0002] To ensure safety, reduce test costs, and improve efficiency, the work of parameter adjustment and optimization design of the hydropower unit regulation system usually relies on a virtual simulation platform. Constructing a high-precision water turbine model is the basis for studying the stability and adaptive control of the water turbine regulation system. However, water turbine modeling generally relies on idealized assumptions, such as ideal fluids, perfect impellers, etc. These assumptions often do not hold in actual engineering, resulting in a deviation between the model results and the actual situation. The actual water turbine is affected by various factors, such as water quality changes, mechanical wear, structural deformation, etc. The uncertainty and variability of these parameters restrict the modeling accuracy. Model correction is a necessary step to improve the accuracy and reliability of the water turbine model after the model is constructed. The corrected water turbine model can better reflect the dynamic response characteristics of the water turbine under different working conditions, such as starting, stopping, and load changes. Conventional model correction inevitably ignores the nonlinearity of the water turbine, and the way of correcting the model input or output increases the complexity of the hydropower unit regulation system simulation platform.

[0003] In view of the limitations of the traditional water turbine modeling method, it is necessary to design a method for modeling a water turbine by combining a slime mold algorithm and a BP neural network to solve the above technical problems. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for modeling a water turbine by combining a slime mold algorithm and a BP neural network, which is used to solve the limitations of the existing water turbine modeling method, so as to ensure the reliability of the subsequent research on the stability and optimal control of the hydropower unit, and realize the long-term efficient and stable operation of the hydropower unit.

[0005] To achieve the above technical effects, the technical solution adopted by the present invention is as follows: A method for modeling a water turbine by combining a slime mold algorithm and a BP neural network, comprising the following steps: S1, obtaining the test data of the prototype water turbine; S2, improving the data of the opening characteristic and the efficiency characteristic; S3, mixing and recombining the improved data of the opening characteristic and the efficiency characteristic to obtain the sample data of the flow characteristic and the torque characteristic; S4, combining the complete torque characteristic data and the BP neural network to construct an opening characteristic neural network model with the unit speed and the unit torque as the input and the guide vane opening as the output; S5, correcting the guide vane opening in the water turbine test data based on the slime mold algorithm; S6. Combine the corrected turbine test data and the BP neural network to construct a neural network model for the flow characteristics and torque characteristics of the turbine.

[0006] Preferably, in step S1, a graph-reading software is used to obtain the prototype turbine test data. The specific method is as follows: Convert the prototype turbine comprehensive characteristic curve graph containing the prototype turbine test data into the fig format; Import the prototype turbine comprehensive characteristic curve graph in the fig format into the "GetData Graph Digitizer2.24" software; Select the x-y coordinate system of the prototype turbine comprehensive characteristic curve graph in "GetData Graph Digitizer 2.24"; Read the efficiency characteristic data on the constant efficiency line and the opening characteristic data on the constant opening line in sequence; Export the read data in the Excel format.

[0007] Preferably, the specific method of step S2 is as follows: Determine the number of neurons in the hidden layer of the BP neural network; Unit speed n 11 and unit torque M 11 are used as inputs, the guide vane opening Y and efficiency η are used as outputs respectively for the opening characteristic neural network model OCNN, Y = Y ( n 11 , M 11 ) and the efficiency characteristic neural network model ECNN, η = η ( n 11 , M 11 ); where n 11 satisfies equation (1): ; (1) Substitute the efficiency characteristic data into the opening characteristic neural network model to obtain the corresponding guide vane opening Y， i.e., obtain the complete efficiency characteristic data; Substitute the opening characteristic data into the efficiency characteristic neural network model to obtain the corresponding efficiency, and calculate the unit flow rate according to equation (2) Q 11 , i.e., obtain the complete opening characteristic data: ; (2) In the formula, γ = ρg , ρ is the density of water, g is the acceleration of gravity; D is the runner diameter; H is the working head; X is the rotational speed.

[0008] Preferably, the number of neurons in the hidden layer of the BP neural network can be determined according to the empirical formula: ; (3) In the formula, N i is the number of neurons in the input layer, N o is the number of neurons in the output layer, N h is the number of neurons in the hidden layer, Z is a constant from 0 to 10.

[0009] Preferably, in the process of building a model based on the BP neural network, the mean square error MSE given by formula (4) is selected as the fitness function, and the weights and thresholds of the BP neural network are continuously adjusted by the gradient descent method to make the model output close to the expected output: ; (4) In the formula, k represents the number of sample data, represents the predicted data, D i represents the sample data.

[0010] Furthermore, to prevent the network from blindly pursuing accuracy requirements and overfitting, the maximum number of iterations is set to 1000, and the number of verification checks is set to 6, that is, if the network passes 6 consecutive verification checks or the number of iterations reaches 1000, the network reaches the termination condition and stops learning in a timely manner.

[0011] Preferably, the specific method of step S3 is: Mix and reorganize the complete opening characteristics and efficiency characteristics data to obtain the sample data of flow characteristics and torque characteristics; Mix and reorganize the opening characteristics and efficiency characteristics data including unit rotational speed, guide vane opening, unit flow rate, and unit torque to obtain the sample data of flow characteristics and torque characteristics with a larger sample space.

[0012] Preferably, the specific method of step S4 is: Determine the number of neurons in the hidden layer of the BP neural network; Construct an opening characteristic neural network model with unit speed and unit torque as inputs and guide vane opening as the output by combining the complete torque characteristic data and the BP neural network.

[0013] Preferably, the specific method of step S5 is as follows: Select a suitable correction function, and combine the slime mold algorithm SMA and the actual operation data of the water turbine to obtain the coefficients in the correction function, that is, realize the correction of the guide vane opening in the water turbine test data. Preferably, based on the fact that the modeling samples come from the prototype water turbine test data and the actual operation data Y There is a linear relationship with its mapped value in the opening characteristic neural network model The correction function adopts a first-order function: ; (5) In the formula, p 0 is the correction coefficient.

[0014] Preferably, the specific method of step S6 is as follows: Determine the number of neurons in the hidden layer of the BP neural network; combine the corrected prototype water turbine test data and the BP neural network to construct a water turbine flow characteristic neural network model DCNN with unit speed and guide vane opening as inputs and unit flow rate as the output, and a water turbine torque characteristic neural network model TCNN with unit speed and guide vane opening as inputs and unit torque as the output.

[0015] The beneficial effects of the present invention are as follows: 1. The water turbine modeling method combining the slime mold algorithm and the BP neural network designed by the present invention can fully consider the nonlinearity of the water turbine and the model accuracy, and effectively ensure the safe and stable operation of the hydropower unit.

[0016] 2. The high-precision nonlinear water turbine model obtained by the present invention can ensure the reliability of the subsequent research on the stability and optimal control of the hydropower unit, thereby effectively ensuring the safe and stable operation of the hydropower unit. Description of the Drawings

[0017] Figure 1 Is the design step of the method of the present invention; Figure 2 Is the schematic diagram of the correction of the water turbine test data based on the slime mold algorithm, BP neural network and actual operation data of the present invention; Figure 3 Is the schematic diagram of the nonlinear water turbine model based on the BP neural network constructed by the present invention. Detailed Embodiments

[0018] Example 1: As Figure 1As shown in the figure, a method for modeling a hydraulic turbine by combining the slime mold algorithm and the BP neural network includes the following steps: S1. Obtain the test data of the prototype hydraulic turbine; S2. Improve the data of the opening characteristic and the efficiency characteristic; S3. Mix and recombine the improved data of the opening characteristic and the efficiency characteristic to obtain the sample data of the flow characteristic and the torque characteristic; S4. Combine the complete torque characteristic data and the BP neural network to construct an opening characteristic neural network model with the unit speed and the unit torque as the inputs and the guide vane opening as the output; S5. Modify the guide vane opening in the test data of the hydraulic turbine based on the slime mold algorithm; S6. Combine the modified test data of the hydraulic turbine and the BP neural network to construct a neural network model for the flow characteristic and the torque characteristic of the hydraulic turbine.

[0019] Preferably, in step S1, a graph reading software is used to obtain the test data of the prototype hydraulic turbine, and the specific method is as follows: Convert the comprehensive characteristic curve graph of the prototype hydraulic turbine containing the test data of the prototype hydraulic turbine into the fig format; Import the comprehensive characteristic curve graph of the prototype hydraulic turbine in the fig format into the "GetData Graph Digitizer2.24" software; Select the x-y coordinate system of the comprehensive characteristic curve graph of the prototype hydraulic turbine in "GetData Graph Digitizer 2.24"; Read the efficiency characteristic data on the constant efficiency line and the opening characteristic data on the constant opening line in sequence; Export the read data in the Excel format.

[0020] Preferably, the specific method of step S2 is as follows: Determine the number of neurons in the hidden layer of the BP neural network; Unit speed n 11 And unit torque M 11 As the inputs, and the guide vane opening Y And efficiency η Respectively as the outputs of the opening characteristic neural network model OCNN, Y = Y ( n 11 , M 11 ) and the efficiency characteristic neural network model ECNN, η = η ( n 11 , M11 ) where n 11 satisfies Equation (1): ; (1) Substitute the efficiency characteristic data into the opening characteristic neural network model to obtain the corresponding guide vane opening Y， That is, the complete efficiency characteristic data is obtained; Substitute the opening characteristic data into the efficiency characteristic neural network model to obtain the corresponding efficiency, and calculate the unit flow rate according to Equation (2) Q 11 , that is, the complete opening characteristic data is obtained: ; (2) In the formula, γ = ρg , ρ is the density of water, g is the acceleration due to gravity; D is the runner diameter; H is the working head; X is the rotational speed.

[0021] Preferably, the number of neurons in the hidden layer of the BP neural network can be determined according to the empirical formula: ; (3) In the formula, N i is the number of neurons in the input layer, N o is the number of neurons in the output layer, N h is the number of neurons in the hidden layer, Z is a constant from 0 to 10.

[0022] Preferably, in the modeling process based on the BP neural network, select the mean square error given by Equation (4) MSE as the fitness function, and continuously adjust the weights and thresholds of the BP neural network through the gradient descent method to make the model output close to the expected output: ; (4) In the formula, k represents the number of sample data, represents the predicted data, D i represents the sample data.

[0023] Furthermore, to prevent the network from blindly pursuing accuracy requirements and overfitting, set the maximum number of iterations to 1000 and the number of validation checks to 6, that is, if the network passes 6 consecutive validation checks or the number of iterations reaches 1000, the network reaches the termination condition and stops learning in a timely manner.

[0024] Preferably, the specific method of step S3 is as follows: Mix and reorganize the complete opening characteristics and efficiency characteristic data to obtain the sample data of the flow characteristics and torque characteristics; Mix and reorganize the opening characteristics and efficiency characteristic data including unit speed, guide vane opening, unit flow rate, and unit torque to obtain the sample data of the flow characteristics and torque characteristics with a larger sample space.

[0025] As Figure 3 shown, preferably, the specific method of step S4 is as follows: Determine the number of neurons in the hidden layer of the BP neural network; Combine the complete torque characteristic data and the BP neural network to construct an opening characteristic neural network model with unit speed and unit torque as inputs and guide vane opening as the output.

[0026] Preferably, the specific method of step S5 is as follows: Select an appropriate correction function, and combine the slime mold algorithm SMA and the actual operation data of the water turbine to obtain the coefficients in the correction function, that is, to correct the guide vane opening in the water turbine test data. Preferably, based on the fact that the modeling samples come from the prototype water turbine test data and the actual operation data Y and there is a linear relationship with their mapped values in the opening characteristic neural network model The correction function adopts a first-order function: ; (5) where p 0 is the correction coefficient.

[0027] Preferably, the specific method of step S6 is as follows: Determine the number of neurons in the hidden layer of the BP neural network; combine the corrected prototype water turbine test data and the BP neural network to construct a water turbine flow characteristic neural network model DCNN with unit speed and guide vane opening as inputs and unit flow rate as the output, and a water turbine torque characteristic neural network model TCNN with unit speed and guide vane opening as inputs and unit torque as the output.

[0028] Embodiment 2: As Figure 2As shown, the Slime Mould Algorithm (SMA) adopted is a bio-inspired optimization algorithm proposed based on the oscillatory foraging behavior of slime moulds, featuring few parameters and strong optimization performance. Slime moulds are a type of protozoa with unique movement and aggregation behaviors, capable of finding the shortest path and effective network connections in complex environments. This behavior makes the slime mould algorithm an efficient method for solving complex optimization problems. The basic principle of the slime mould algorithm is to simulate the process of slime moulds moving towards food sources. Each "slime mould particle" in the algorithm represents a potential solution and moves in the search space, attempting to find the optimal solution. The slime mould particles communicate and interact with each other through pheromones (similar to pheromones in the ant colony algorithm) to coordinate the movement direction. During the movement process, the slime mould particles adjust their movement directions according to the surrounding environmental information (such as distance, obstacles, etc.) and continuously update their own states. Finally, when the slime mould particles reach certain conditions (such as reaching the food source or reaching the upper limit of the iteration number), the algorithm stops and outputs the found optimal solution or approximate solution. By integrating this natural phenomenon into its optimization process, SMA has great potential in efficiently solving complex optimization problems.

[0029] The position update formula for simulating the predation behavior of slime moulds is as follows: ; (6) where l is the current iteration number, is the position of the individual with the highest fitness at the l -th iteration, and are the positions of any two slime mould individuals; and are the oscillation parameters, , a as shown in Equation (7), vc the value of oscillates between r and finally tends to 0; is a random number between p ; W is the slime mould mass, representing the fitness weight, as shown in Equation (8); the control variable

[0030] ; ; ; where is the maximum number of iterations; , is the population size; is the fitness value of the i -th slime mould; F b is the best fitness value in all iterations; is the fitness value sequence of the slime mold; F w represents the worst fitness value in the current iteration; , is the sorting function.

[0031] The mathematical model of the slime mold wrapping food is shown in Equation (10): ; where, and represent the upper and lower boundaries of the search range; represents a random value between; z is a custom parameter, set to 0.03.

Claims

1. A turbine modeling method combining slime mold algorithm and BP neural network, characterized in that: The following steps are involved: S1, obtain prototype turbine test data; S2, improve the opening characteristics and efficiency characteristics data; S3, mixing and reorganizing the improved opening characteristic and efficiency characteristic data to obtain flow characteristic and torque characteristic sample data; S4, combining the complete torque characteristic data and BP neural network to construct an opening characteristic neural network model with unit speed and unit torque as input and guide vane opening as output; S5, correcting the guide vane opening in the turbine test data based on the slime mold algorithm; S6, combining the corrected turbine test data and BP neural network to construct the neural network model of turbine flow characteristic and torque characteristic.

2. A turbine modeling method combining slime mold algorithm and BP neural network according to claim 1, characterized in that: In step S1, the prototype turbine test data is obtained using the image reading software, and the specific method is as follows: Convert the prototype turbine comprehensive characteristic curve diagram containing the prototype turbine test data into fig format; Import the comprehensive characteristic curve of the prototype turbine in fig format into the "GetData Graph Digitizer2.24" software; Select the xy coordinate system of the prototype turbine comprehensive characteristic curve in "GetData Graph Digitizer 2.24"; Sequentially read the efficiency characteristic data on the equal efficiency line and the opening characteristic data on the equal opening line; Export the read data in Excel format.

3. The method for modeling a hydraulic turbine combining a slime mold algorithm and a BP neural network according to claim 1, characterized in that: The specific method of step S2 is: Determine the number of neurons in the hidden layer of the BP neural network; Unit speed n 11 and unit moment M 11 is the input, guide vane opening Y and efficiency η They are the output opening characteristic neural network model OCNN, Y = Y ( n 11 , M 11 ) and the efficiency characteristic neural network model ECNN, η = η ( n 11 , M 11 );where n 11 Satisfies formula (1): ；（1） Substitute the efficiency characteristic data into the opening characteristic neural network model to obtain the corresponding guide vane opening Y， That is, complete efficiency characteristic data is obtained; Substitute the opening characteristic data into the efficiency characteristic neural network model to obtain the corresponding efficiency, and calculate the unit flow rate according to formula (2): Q 11 , that is, to obtain the complete opening characteristic data: ；（2） In the formula, γ = ρg , ρ is the density of water, g is the acceleration due to gravity; D is the runner diameter; H is the working head; X It is the rotation speed.

4. The method for modeling a water turbine combining a slime mold algorithm and a BP neural network according to claim 3, characterized in that: The number of neurons in the hidden layer of the BP neural network can be determined according to the empirical formula: ；（3） In the formula, N i is the number of neurons in the input layer, N o is the number of neurons in the output layer, N h is the number of neurons in the hidden layer, Z A constant between 0 and 10.

5. A turbine modeling method combining slime mold algorithm and BP neural network according to claim 4, characterized in that: In the modeling process based on BP neural network, the mean square error given by formula (4) is selected MSE As the fitness function, the weights and thresholds of the BP neural network are continuously adjusted through the gradient descent method to make the model output close to the expected output: ；（4） In the formula, k represents the number of sample data, Represents the forecast data, D i Represents sample data.

6. The method for modeling a hydraulic turbine combining a slime mold algorithm and a BP neural network according to claim 1, characterized in that: The specific method of step S3 is: The complete opening characteristic and efficiency characteristic data are mixed and reorganized to obtain flow characteristic and torque characteristic sample data; The opening characteristic and efficiency characteristic data including unit speed, guide vane opening, unit flow rate and unit torque are mixed and reorganized to obtain flow characteristic and torque characteristic sample data with a large sample space.

7. The method for modeling a water turbine combining a slime mold algorithm and a BP neural network according to claim 1, characterized in that: The specific method of step S4 is: Determine the number of neurons in the hidden layer of the BP neural network; The complete torque characteristic data and BP neural network are combined to construct an opening characteristic neural network model with unit speed and unit torque as input and guide vane opening as output.

8. The method for modeling a water turbine combining a slime mold algorithm and a BP neural network according to claim 1, characterized in that: The specific method of step S5 is: Select a suitable correction function, combine the slime mold algorithm SMA and the actual operation data of the turbine to obtain the coefficients in the correction function, that is, to correct the guide vane opening in the turbine test data.

9. A turbine modeling method combining slime mold algorithm and BP neural network according to claim 8, characterized in that: Based on the modeling samples from prototype turbine test data and actual operation data Y Instead of mapping values ​​in the opening characteristic neural network model There is a linear relationship, and the correction function uses a first-order function: ； （5） In the formula, p 0 is the correction factor.

10. The method for modeling a water turbine combining a slime mold algorithm and a BP neural network according to claim 1, characterized in that: The specific method of step S6 is: Determine the number of neurons in the hidden layer of the BP neural network; combine the modified prototype turbine test data and the BP neural network to construct a turbine flow characteristic neural network model DCNN with unit speed and guide vane opening as input and unit flow as output, and a turbine torque characteristic neural network model TCNN with unit speed and guide vane opening as input and unit torque as output.

Images