CN115455821A - Centrifugal pump turbine performance prediction method based on improved PSO-GA algorithm - Google Patents

Centrifugal pump turbine performance prediction method based on improved PSO-GA algorithm Download PDF

Info

Publication number
CN115455821A
CN115455821A CN202211109337.7A CN202211109337A CN115455821A CN 115455821 A CN115455821 A CN 115455821A CN 202211109337 A CN202211109337 A CN 202211109337A CN 115455821 A CN115455821 A CN 115455821A
Authority
CN
China
Prior art keywords
population
individual
algorithm
individuals
adopting
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202211109337.7A
Other languages
Chinese (zh)
Inventor
周佩剑
余文进
牟介刚
曾卫涛
李健
吴宇涵
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Jiliang University
Original Assignee
China Jiliang University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China Jiliang University filed Critical China Jiliang University
Priority to CN202211109337.7A priority Critical patent/CN115455821A/en
Publication of CN115455821A publication Critical patent/CN115455821A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Abstract

The invention discloses a method for predicting the turbine performance of a centrifugal pump by using an improved PSO-GA algorithm, belonging to the field of turbine performance prediction of centrifugal pumps. The method comprises the steps of firstly constructing a BP neural network, and adopting the reciprocal of the mean square error of the neural network as a fitness function of an optimization algorithm. Secondly, initializing an initial population of the C-PSO-GA hybrid optimization algorithm, and performing normalization operation on the data set. And then optimizing the initial weight and the threshold value of the BP neural network by adopting a C-PSO-GA hybrid optimization algorithm. And finally, training the optimized BP neural network by adopting a training set, and taking one random pump in a verification set for verification. The method further improves the operation efficiency of the algorithm, is simpler and more convenient to use compared with the existing simulation method, has a short calculation period, and can accurately predict the turbine performance of the centrifugal pump.

Description

Centrifugal pump turbine performance prediction method based on improved PSO-GA algorithm
Technical Field
The invention belongs to the field of turbine performance prediction of centrifugal pumps, and particularly relates to a centrifugal pump turbine performance prediction method based on an improved PSO-GA algorithm.
Background
In the field of predicting the performance of the centrifugal pump as a turbine, for the traditional method of simulating by adopting CFD, the accuracy and the prediction speed can really reach a high degree, but the modeling process is complicated, the prediction precision is greatly influenced by the grid quality, and a large amount of resources are required to be occupied for operation, so that the method cannot be applied to engineering problems that the performance parameters of the centrifugal pump as the turbine and the like need to be quickly obtained. For the method of predicting by using the loss function model, the model has the problems of high difficulty in establishing, inconvenience in obtaining input parameters, poor generalization capability of the model and the like, and the engineering problem that the centrifugal pump is required to be rapidly obtained as the turbine performance parameters is not well solved.
In recent years, machine learning methods mainly based on neural networks have been rapidly developed, and neural networks are learning models which achieve fitting effects through continuous iteration and have good applicability and generalization performance. At present, an artificial neural network is utilized to predict the optimal working condition point of the centrifugal pump under the turbine state, certain precision can be achieved, but the hydraulic characteristic prediction method under the whole working condition under the turbine condition is still less, and the precision is lower.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides the centrifugal pump turbine performance prediction method of the improved PSO-GA algorithm, which can more accurately and rapidly predict the hydraulic characteristics of the pump in the turbine state under the condition that partial parameters in the pumping state are known.
The purpose of the invention is realized by the following technical scheme:
a centrifugal pump turbine performance prediction method of an improved PSO-GA algorithm (C-PSO-GA) comprises the following steps:
the method comprises the following steps: a data set is constructed and divided into a training set and a validation set.
The training set comprises operation data of a plurality of centrifugal pumps in a turbine state, and each training sample comprises geometric parameters, specific rotating speed and flow of the centrifugal pump and a lift corresponding to each flow in the turbine state.
Step two: and constructing a BP neural network, and adopting the reciprocal of the mean square error of the neural network as a fitness function of the optimization algorithm.
Step three: initializing a C-PSO-GA mixed optimization algorithm to initialize a population, and carrying out normalization operation on a data set.
Step four: and optimizing the initial weight and the threshold value of the BP neural network by adopting a C-PSO-GA hybrid optimization algorithm.
The C-PSO-GA optimization algorithm selects n individuals, each individual comprises all weights and thresholds of a BP neural network, population initialization is carried out by adopting a reverse learning initialization population strategy, inertial weight w is a self-adaptive weight coefficient, and the calculation method comprises the following steps:
Figure BDA0003842540650000021
wherein w k Is the kth weight, n 1 The value is 0.6,n 2 The value is 0.3, ax iter is the maximum number of iterations, and k is the number of iterations.
In each iteration, the individual position information of the best fitness value calculated in the population is stored in the global best position information (gbest), and the position information of the best fitness value calculated by each individual is stored in the individual best position information (pbest).
Step five: and training the optimized BP neural network by adopting a training set, and taking one random pump in a verification set for verification.
Further, the BP neural network comprises an input layer, two hidden layers and an output layer, wherein the input layer comprises 9 neurons which respectively correspond to the specific rotation speed of the centrifugal pump in the pumping state, the flow rate of each working condition, the number of blades, the diameter of an impeller inlet, the width of a volute outlet, the diameter of the volute inlet, the width of the impeller outlet, the placement angle of the impeller inlet and the placement angle of the impeller outlet; the output layer comprises 1 neuron and is the lift corresponding to the flow input by the input layer; and acquiring the number of the neurons of the first layer and the second layer of the hidden layer by adopting a trial and error method.
Further, the BP neural network activation function is a LeakyReLU function.
Further, the improved C-PSO-GA hybrid optimization algorithm performs population initialization by using a reverse learning initialization population strategy, and the calculation method is as follows:
firstly, initializing a random population by adopting uniform distribution, wherein the upper limit of individual population is b, and the lower limit is a.
Secondly, generating an individual reverse digital matrix corresponding to each population individual matrix, wherein the reverse digital generation method is as follows:
P＝a+b+p
wherein, P is the reverse number of the initial population individual, and P is the number of the population individual.
And finally, taking out the individuals from the initialized random population and the reverse digital matrix according to the sequence, and calculating the fitness of the individuals by using a fitness function. And selecting individuals with higher fitness and putting the individuals into corresponding positions of the final initial population.
Further, the improved PSO-GA hybrid optimization algorithm (C-PSO-GA) performs normalization on the data set during optimization, and performs normalization by combining a sin function and a min-max normalization method, where a calculation formula is as follows:
Figure BDA0003842540650000031
wherein y is the value after normalization, x is the original data value before normalization, max is the maximum value of the original data set, and min is the minimum value of the original data set.
Further, the C-PSO-GA hybrid optimization algorithm updates the individual information by using a position and velocity information update formula, which specifically includes:
Figure BDA0003842540650000032
Figure BDA0003842540650000033
wherein i is the ith individual,
Figure BDA0003842540650000034
is the current speed information of the ith individual of the kth generation, w is the inertia weight, c 1 And c 2 Are an individual learning factor and a group learning factor, r 1 And r 2 Is [0,1 ]]The random number in the random number table (R),
Figure BDA0003842540650000035
for the kth generation ith individual optimal location information, g k For the optimal location information of the kth generation population,
Figure BDA0003842540650000036
current position information of the ith individual in the kth generation.
Further, the improved PSO-GA hybrid optimization algorithm adopts an elite-preserving crossover strategy to carry out crossover operation, and the calculation method is as follows:
after one iteration is finished, randomly selecting a plurality of groups of individuals from the group of individuals and carrying out cross operation on the optimal position of the population to obtain offspring individuals, mixing parent individuals and the offspring individuals, and then adding the mixture into a fitness function for calculation, and calculating R of the mixture 2 Evaluating the value, selecting the individual with larger fitness function value as elite individual to be put into the next generation population, and selecting R when the fitness function values are similar 2 Individuals with larger values were designated as elite individuals. The crossover formula is as follows:
Figure BDA0003842540650000037
Figure BDA0003842540650000038
wherein, L is a cross factor, and m is an adaptive cross coefficient.
The invention has the following beneficial effects:
(1) The cross strategy of preserving elite adopted by the invention can improve the convergence rate of the prediction method to a certain extent and further improve the operation efficiency. Compared with the traditional PSO-BP or GA-BP algorithm, the method has the advantages of higher convergence speed, higher optimization precision and higher effectiveness and operation efficiency.
(2) Compared with the existing simulation method, the method is simpler and more convenient to use, has a short calculation period, and can accurately predict the turbine performance of the centrifugal pump.
Drawings
FIG. 1 is a flow chart of a method for predicting turbine performance of a centrifugal pump using an improved PSO-GA-BP algorithm;
FIG. 2 (a) is a structural parameter diagram of a pumping chamber;
FIG. 2 (b) is a structural parameter diagram of an impeller;
FIG. 3 is an iteration R of the number of neurons in different hidden layers 2 A change in value condition;
FIG. 4 is a comparison graph of the change in fitness value of four algorithms;
fig. 5 (a) is a comparison graph of head prediction errors predicted by four methods;
fig. 5 (b) is a diagram of head prediction predicted by four methods.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments, and the objects and effects of the present invention will be more apparent, it being understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.
As shown in FIG. 1, the method for predicting the turbine performance of the centrifugal pump based on the C-PSO-GA-BP algorithm comprises the following steps:
the method comprises the following steps: a training set is constructed and divided into a training set and a validation set according to a ratio of 8 to 2.
The training set comprises operation data of a plurality of centrifugal pumps in a turbine state, and each training sample comprises geometric parameters, specific rotating speed and flow of the centrifugal pump and a lift corresponding to each flow in the turbine state.
In the embodiment, each training sample has 9 input layers of data, including the specific speed Ns in the pumping state of the centrifugal pump, the flow Q in each working condition, the blade number Z and the inlet diameter D of the impeller 0 Volute outlet width D 2 Volute inlet diameter D 1 Impeller outlet width b and impeller inlet placement angle beta 1 And impeller exit setting angle beta 2 . The output of the output layer is one, which is the head H corresponding to the flow input by the input layer. The structural parameter diagrams of the impeller and the pumping chamber of the centrifugal pump are shown in FIG. 2 (a) and FIG. 2 (b), in which D 0 Is the diameter of the impeller inlet D 2 Is the volute outlet width, D 1 Is the volute inlet diameter, b is the impeller outlet width, beta 1 Setting angle and beta for impeller inlet 2 And setting an angle for an impeller outlet.
The training set partial data is shown in table 1 below:
TABLE 1 data of training set for centrifugal pump as turbine part
Figure BDA0003842540650000051
Step two: constructing a BP neural network, and adopting the reciprocal of the mean square error of the neural network as a fitness function of an optimization algorithm:
in this embodiment, the artificial neural network includes one input layer, two hidden layers, and one output layer. Determining the number of neurons in two hidden layers by adopting a trial and error method, and after iterating the number of neurons in different hidden layers for 500 times, R 2 The change of the values is shown in FIG. 3, and it can be seen that when the hidden layer of the first layer is 16 neurons and the hidden layer of the second layer is 10 neurons, R is 2 The value is the largest, so the first layer of hidden layer is selected to be 16 neurons, and the second layer of hidden layer is selected to be 10 neurons. The specific structure of the BP neural network is shown in table 2 below.
TABLE 2 BP neural network concrete structure
Figure BDA0003842540650000052
Figure BDA0003842540650000061
The loss function adopts the mean square error as the loss function of the BP neural network, compared with other loss functions, the mean square error is simple and convenient to calculate, and the gradient is reduced along with the reduction of the error, thereby being beneficial to convergence. Compared with other activation functions, the LeakyReLU function is adopted as the activation function, the ReLU activation function can better mine relevant features and fit training data, and the problem of gradient disappearance can be effectively solved by adopting the ReLU function. The LeakyReLU activation function adds a small slope to a negative value part on the basis of the ReLU function, so that the problem of neuron death is solved. The BP neural network activation function is a LeakyReLU function, and the expression of the BP neural network activation function is as follows:
Figure BDA0003842540650000062
wherein, a i Is a fixed parameter within (1, + ∞), 1/a i A negative slope coefficient, which takes a value of 0.01; x is the number of i Representing the input quantity, y, of neurons of the current layer i Representing the output of the current layer neurons.
In addition, the reciprocal of the mean square error of the neural network is used as the fitness function of the optimization algorithm, and compared with the situation that the mean square error is directly used as the fitness function of the optimization algorithm, the reciprocal of the mean square error can be used for amplifying the fitness function value, so that the optimization effect can be intuitively felt.
Step three: initializing a C-PSO-GA mixed optimization algorithm to initialize a population, and carrying out normalization operation on a data set.
In this embodiment, a reverse learning initialization population strategy is adopted for population initialization, and compared with a conventional uniform random initialization or 0-1 initialization method, the method retains more elegance population individuals, and can improve the convergence rate of the optimization algorithm to a certain extent, and the calculation method is as follows:
firstly, initializing a random population by adopting uniform distribution, wherein the upper limit of individual population is b, and the lower limit is a.
Secondly, generating an individual reverse digital matrix corresponding to each population individual matrix, wherein the reverse digital generation method is as follows:
P＝a+b+p
wherein, P is the reverse number of the initial population individual, and P is the number of the population individual.
And finally, taking out the individuals from the initialized random population and the reverse digital matrix according to the sequence, and calculating the fitness of the individuals by using a fitness function. And selecting individuals with higher fitness and putting the individuals into corresponding positions of the final initial population.
In the embodiment, the normalization operation is performed by combining a sin function and a min-max normalization method, the centrifugal pump is used for turbine data concentration, and when the flow is in a small value or a large value, the change rate of the lift is high, so that when the min-max normalization method is simply adopted, the lift data are all accumulated in an interval of 0.3-0.7, and the prediction error is large. And by adopting a mode of combining the sin function and the min-max normalization method, the lift data can be uniformly distributed in the interval of 0-1, so that the error is reduced to a certain extent. The specific calculation formula is as follows:
Figure BDA0003842540650000071
wherein y is the value after normalization, x is the original data value before normalization, max is the maximum value of the original data set, and min is the minimum value of the original data set.
Step four: and optimizing the initial weight and the threshold value of the BP neural network by adopting a C-PSO-GA hybrid optimization algorithm.
In this embodiment, the C-PSO-GA optimization algorithm selects 20 individuals, each of which includes all weights and thresholds of the BP neural network, performs population initialization using a back learning initialization population strategy, where the iteration number is selected to be 50, and the inertial weight w is an adaptive weight coefficient, and the calculation method is as follows:
Figure BDA0003842540650000072
wherein, w k Is the kth weight, n 1 The value is 0.6,n 2 The value is 0.3, ax _iteris the maximum number of iterations, and k is the iteration algebra.
The cross probability is 0.6,c 1 And c 2 The value is 2, the number of BP neural network iterations is selected as 500, and the iteration is substituted into a training set for operation. In each generation, the individual position information of the best fitness value calculated in the population is stored in the global best position information (gbest), the position information of the best fitness value calculated by each individual is stored in the individual best position information (pbest), and the position and speed information updating formula is as follows:
Figure BDA0003842540650000073
Figure BDA0003842540650000074
wherein, i is the ith individual,
Figure BDA0003842540650000075
is the current speed information of the ith individual of the kth generation, w is the inertia weight, c 1 And c 2 Are an individual learning factor and a group learning factor, r 1 And r 2 Is [0,1 ]]The random number of the inner part of the random number,
Figure BDA0003842540650000076
for the ith generation of the ith individual optimum position information, g k For the optimal location information of the kth generation population,
Figure BDA0003842540650000081
is the current position information of the ith individual in the kth generation.
In addition, after one iteration is finished, several groups of individuals are randomly selected to be crossed with the best position of the population (the probability is 0.6), offspring individuals are obtained, the parent individuals and the offspring individuals are mixed, and then the mixture enters a fitness function for calculation, and R of the mixture is calculated 2 Evaluating the value, dividing the fitness value into intervals, taking 100 numbers as one interval, selecting the individual with larger fitness value as the elite individual to be put into the next generation of population in different intervals, and selecting R in the same interval 2 Individuals with larger values were designated as elite individuals. The crossover formula is as follows:
Figure BDA0003842540650000082
Figure BDA0003842540650000083
wherein, L is a crossover factor, m is an adaptive crossover coefficient, and the value in this embodiment is 0.6.
And after the iteration is finished, taking the global optimal position information (gbest) as the initialization weight and the threshold value of the BP neural network to obtain the optimized BP neural network. As shown in fig. 4 below, it can be seen that, compared with the conventional PSO optimization algorithm, the PSO optimization algorithm using the adaptive crossover operator has a faster optimization speed and a better optimization effect. In addition, the population is initialized by adopting a reverse learning strategy, so that the optimization algorithm can find the position information with a higher fitness value at the initial stage more quickly, the function convergence speed is increased, and the precision is improved to a certain extent.
Step five: training the optimized BP neural network by adopting a training set, and verifying by taking a random pump in a verification set:
fig. 5 (a) and 5 (b) are a diagram for comparing the performance prediction and the prediction error of a centrifugal pump with a specific speed of 75, and it can be seen that, by comparing the prediction errors of the three, the errors are basically within 10% by using PSO-BP and GA-BP neural networks for prediction, wherein the maximum error occurs at the prediction point 2, the maximum value is 7.01%, and the prediction error is larger, while by using the PSO-GA-BP neural networks for prediction, the prediction error is basically within 4%, the maximum error is 5.61%, and the average error value is 1.44%. The neural network optimized by the C-PSO-GA algorithm is adopted for prediction, the maximum prediction error is 2.14%, the average error is 0.97%, and compared with other algorithm optimization, the average prediction error of the C-PSO-GA algorithm by adopting a reverse learning method and a self-adaptive crossover operator is improved by about 2.5%, and the average errors are similar. The comparison test results show that the C-PSO-GA algorithm provided by the invention can effectively establish a multi-working-condition prediction model, and compared with other algorithms, the algorithm has smaller maximum error, can provide higher precision and can better meet the requirements of engineering practice.
The method of the invention is different from the existing method for predicting the turbine performance of the centrifugal pump, in that the conditions of poor local search capability and low search precision of the PSO-BP algorithm are considered, the thought of genetic algorithm crossing is added, individuals are subjected to crossing operation, the search precision of the PSO algorithm is improved to a certain extent, a reverse learning strategy is added, the convergence rate and the search precision of the algorithm are further improved, and a better prediction effect is obtained.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and although the invention has been described in detail with reference to the foregoing examples, it will be apparent to those skilled in the art that various changes in the form and details of the embodiments may be made and equivalents may be substituted for elements thereof. All modifications, equivalents and the like which come within the spirit and principle of the invention are intended to be included within the scope of the invention.

Claims (6)

1. A centrifugal pump turbine performance prediction method of an improved PSO-GA algorithm is characterized by comprising the following steps:
the method comprises the following steps: constructing a data set, and dividing the data set into a training set and a verification set;
the training set comprises operation data of a plurality of centrifugal pumps in a turbine state;
step two: constructing a BP neural network, and adopting the reciprocal of the mean square error of the neural network as a fitness function of an optimization algorithm;
step three: initializing a C-PSO-GA hybrid optimization algorithm to initialize a population, and performing normalization operation on a data set;
the population initialization is carried out by adopting an initialization population strategy for reverse learning, and the process is as follows:
firstly, initializing a random population by adopting uniform distribution, wherein the upper limit of individual population is b, and the lower limit is a;
secondly, generating an individual reverse digital matrix corresponding to each population individual matrix;
finally, according to the sequence, taking out the individuals from the initialized random population and the reverse digital matrix, calculating the fitness of the individuals by using a fitness function, selecting the individuals with high fitness, and putting the individuals into the corresponding positions of the final initial population;
the data set is normalized by adopting a mode of combining a sin function and a min-max normalization method, and the calculation formula is as follows:
Figure FDA0003842540640000011
wherein y is the value after normalization, x is the original data value before normalization, max is the maximum value of the original data set, and min is the minimum value of the original data set;
step four: optimizing the initial weight and the threshold of the BP neural network by adopting a C-PSO-GA hybrid optimization algorithm;
selecting n individuals by using a C-PSO-GA hybrid optimization algorithm, wherein each individual comprises all initial weights and thresholds of a BP neural network, and performing population initialization by adopting a reverse learning initialization population strategy; the inertial weight w is an adaptive weight coefficient and is calculated as follows:
Figure FDA0003842540640000012
wherein w k Is the kth weight, n 1 The value is 0.6,n 2 The value is 0.3, ax \\iteris the maximum iteration number, and k is the iteration algebra;
in each iteration, the individual position information of the optimal fitness value calculated in the population is stored in the global optimal position information, the position information of the optimal fitness value calculated by each individual is stored in the individual optimal position information, and the position and speed information updating formula updates the individual information, wherein the specific updating formula is as follows:
Figure FDA0003842540640000021
Figure FDA0003842540640000022
wherein i is the ith individual,
Figure FDA0003842540640000023
is the current speed information of the ith individual of the kth generation, w is the inertia weight, c 1 And c 2 Are an individual learning factor and a group learning factor, r 1 And r 2 Is [0,1 ]]The random number of the inner part of the random number,
Figure FDA0003842540640000024
for the ith generation of the ith individual optimum position information, g k For the optimal location information of the kth generation population,
Figure FDA0003842540640000025
current position information of the ith individual in the kth generation;
adopting a cross strategy of preserving elite to carry out cross operation, wherein the process is as follows:
after one iteration is finished, randomly selecting a plurality of groups of individuals and the optimal position of the population to carry out cross operation to obtain offspring individuals, mixing parent individuals and offspring individuals to obtain a mixture, adding the mixture into a fitness function to calculate, and calculating R of the mixture 2 Evaluating the value, and selecting an individual with a larger fitness function value as an elite individual to be put into the next generation of population; the crossover formula is as follows:
Figure FDA0003842540640000026
Figure FDA0003842540640000027
wherein, L is a cross factor, and m is a self-adaptive cross coefficient;
step five: and training the optimized BP neural network by adopting a training set, and taking a sample of a random pump in a verification set for verification.
2. The method for predicting the turbine performance of the centrifugal pump based on the improved PSO-GA algorithm, as claimed in claim 1, is characterized in that: the operation data in the step one comprise geometric parameters, specific rotating speed, flow and lift corresponding to each flow in a turbine state of the centrifugal pump.
3. The method for predicting the turbine performance of the centrifugal pump based on the improved PSO-GA algorithm, as claimed in claim 2, is characterized in that: the geometric parameters comprise the number of blades, the diameter of an impeller inlet, the width of a volute outlet, the diameter of the volute inlet, the width of an impeller outlet, an impeller inlet setting angle and an impeller outlet setting angle.
4. The method for predicting the turbine performance of the centrifugal pump of the improved PSO-GA algorithm in claim 1 is characterized in that: the BP neural network comprises an input layer, two hidden layers and an output layer, wherein the input layer comprises nine neurons, the output layer comprises one neuron, and the number of the neurons of the first layer and the second layer hidden layers is obtained by adopting a trial and error method;
the BP neural network activation function is a LeakyReLU function.
5. The method for predicting the turbine performance of the centrifugal pump based on the improved PSO-GA algorithm, as claimed in claim 1, is characterized in that: step three, the individual reverse number matrix, wherein the reverse number generation algorithm is as follows:
P＝a+b+p
wherein, P is the reverse number of the initial population individuals, and P is the number of the population individuals.
6. The method for predicting the turbine performance of the centrifugal pump of the improved PSO-GA algorithm in claim 1 is characterized in that: the cross operation in the fourth step selects R when the fitness function values are similar 2 Individuals with larger values were designated as elite individuals.
CN202211109337.7A 2022-09-13 2022-09-13 Centrifugal pump turbine performance prediction method based on improved PSO-GA algorithm Pending CN115455821A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211109337.7A CN115455821A (en) 2022-09-13 2022-09-13 Centrifugal pump turbine performance prediction method based on improved PSO-GA algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211109337.7A CN115455821A (en) 2022-09-13 2022-09-13 Centrifugal pump turbine performance prediction method based on improved PSO-GA algorithm

Publications (1)

Publication Number Publication Date
CN115455821A true CN115455821A (en) 2022-12-09

Family

ID=84303458

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211109337.7A Pending CN115455821A (en) 2022-09-13 2022-09-13 Centrifugal pump turbine performance prediction method based on improved PSO-GA algorithm

Country Status (1)

Country Link
CN (1) CN115455821A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117314785A (en) * 2023-10-30 2023-12-29 阿尔麦德智慧医疗(湖州)有限公司 AI-based ultrasound contrast diagnosis auxiliary system

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117314785A (en) * 2023-10-30 2023-12-29 阿尔麦德智慧医疗(湖州)有限公司 AI-based ultrasound contrast diagnosis auxiliary system

Similar Documents

Publication Publication Date Title
CN111353582B (en) Particle swarm algorithm-based distributed deep learning parameter updating method
CN107316099A (en) Ammunition Storage Reliability Forecasting Methodology based on particle group optimizing BP neural network
CN107992648A (en) A kind of Adaptive radial basis function neural network carries out aeroengine thrust algorithm for estimating
CN106991212B (en) Root strength prediction method based on GA _ PSO (genetic Algorithm-particle swarm optimization) GRNN (generalized regression neural network) algorithm
CN109086531A (en) Antenna design method neural network based
CN110008653A (en) A kind of aviation centrifugal pump blade profile optimum design method
Li et al. Numerical optimization for stator vane settings of multi-stage compressors based on neural networks and genetic algorithms
CN111898212B (en) Impeller mechanical profile design optimization method based on BezierGAN and Bayesian optimization
CN115455821A (en) Centrifugal pump turbine performance prediction method based on improved PSO-GA algorithm
CN112270139B (en) Pneumatic optimization design method for centrifugal compressor of fuel cell based on mother type library
CN111625998A (en) Method for optimizing structure of laminated solar cell
CN110222830A (en) A kind of depth feedforward network method for diagnosing faults based on self-adapted genetic algorithm optimization
Miao et al. Swarm intelligence based on modified PSO algorithm for the optimization of axial-flow pump impeller
CN113361047B (en) Impeller high-dimensional optimization and model selection method for multilayer sensing structure
CN113095477B (en) Wind power prediction method based on DE-BP neural network
CN106786499A (en) Based on the short-term wind power forecast method for improving AFSA optimizations ELM
CN106650028A (en) Optimization method and system based on agile satellite design parameters
CN117272868A (en) Nozzle optimization method and device based on RBF-ANN and genetic algorithm and computer storage medium
CN112070418A (en) Weapon target allocation method of multi-target whale optimization algorithm
CN114547987B (en) Centrifugal pump turbine performance prediction method based on improved artificial neural network
CN107066686A (en) A kind of axial-flow pump impeller Hydraulic Optimizing Design method based on genetic algorithm
CN114511158B (en) Wind turbine power prediction method based on wake deflection effect and 2DJensen model
CN113032934B (en) Intelligent checking method for dynamic parameters of wind power plant based on ME-TD3 algorithm
CN110766072A (en) Automatic generation method of computational graph evolution AI model based on structural similarity
CN105069192B (en) A kind of improved method that power of fan parameter of curve model is solved based on genetic algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination