CN111144027A - Approximation method based on BP neural network full characteristic curve function - Google Patents

Approximation method based on BP neural network full characteristic curve function Download PDF

Info

Publication number
CN111144027A
CN111144027A CN202010001277.1A CN202010001277A CN111144027A CN 111144027 A CN111144027 A CN 111144027A CN 202010001277 A CN202010001277 A CN 202010001277A CN 111144027 A CN111144027 A CN 111144027A
Authority
CN
China
Prior art keywords
network
function
torque
unit
neural network
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
CN202010001277.1A
Other languages
Chinese (zh)
Inventor
苏文涛
李洋
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.)
Liaoning Shihua University
Original Assignee
Liaoning Shihua 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 Liaoning Shihua University filed Critical Liaoning Shihua University
Priority to CN202010001277.1A priority Critical patent/CN111144027A/en
Publication of CN111144027A publication Critical patent/CN111144027A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Control Of Water Turbines (AREA)

Abstract

The invention provides an approximation method based on a BP neural network full characteristic curve function, which comprises the following steps of 1, collecting sample data of a full characteristic curve of a water pump-turbine model test; step 2, normalizing and extending the collected data; and 3, establishing a three-layer BP neural network to train the processed data, obtaining an ideal network, and completing approximation of the full-characteristic nonlinear relation of the water pump turbine, so that the data obtained after training is drawn into a space curved surface. The method utilizes the BP neural network for fitting, and has the characteristics of high precision, strong generalization capability, capability of meeting the requirement of continuous guidance of curved surfaces and the like.

Description

Approximation method based on BP neural network full characteristic curve function
Technical Field
The invention belongs to the technical field of water turbine full characteristic curve expression, and particularly relates to an approximation method based on a BP neural network full characteristic curve function.
Background
The full characteristic curve is obtained from model test data, can be used for calculating the transition process of the unit, and is also an important basis for establishing a water turbine model in a water pump turbine regulating system. However, due to the severe nonlinear characteristics of the full characteristic curve, there is great difficulty in the calculation of the transition process, and therefore, it is very important to find a suitable method for describing the full characteristic curve. The artificial neural network is a technology which is artificially constructed by human beings on the basis of understanding of the brain neural network and can realize certain functions. The extremely strong nonlinear mapping capability of the neural network has important application in various fields. The full characteristic curve of the pump turbine shows strong nonlinear characteristics due to the existence of the S-shaped area, and the neural network just becomes an effective means for processing the problem.
Disclosure of Invention
The invention aims to solve the problem of insufficient representation of a full characteristic curve of a water pump turbine in the prior art, and provides an approximation method based on a BP neural network full characteristic curve function. The invention utilizes the existing model test data to properly arrange and train the neural network after extension to obtain the network function.
The invention is realized by the following technical scheme, and provides an approximation method based on a BP neural network full characteristic curve function, which comprises the following steps:
step 1, collecting sample data of a full characteristic curve of a water pump and water turbine model test;
step 2, normalizing and extending the collected data;
and 3, establishing a three-layer BP neural network to train the processed data, obtaining an ideal network, and completing approximation of the full-characteristic nonlinear relation of the water pump turbine, so that the data obtained after training is drawn into a space curved surface.
Further, the normalization uses mapminmax () function, which limits all parameters in the sample data to between [0,1] using the principle of probability statistics.
Further, in the process of continuation, two principles need to be followed: (1) when the opening of the guide vane is 0, the flow of the water turbine is 0 no matter how large the rotating speed is, so that the boundary condition when the opening of the guide vane is 0 can be constructed, and the flow characteristic of the water turbine at the runaway rotating speed also needs to be considered; (2) the output torque of the water turbine during the runaway is 0, and the torque boundary condition of the water turbine under the opening degree of 0 is also considered.
Further, in step 3, the expression equation of the change of the pump turbine parameter is as follows:
Figure BDA0002353596220000021
wherein Q is11Is the unit flow rate of the pump turbine, M11Unit torque, α guide vane opening, n11Is unit rotating speed; the unit flow and the unit torque are both functions formed by the opening degree of the guide vane and the unit rotating speed, therefore, the network to be established is a three-layer BP neural network which has double inputs and single outputs and is composed of a plurality of neurons, the number of nodes of hidden layers of a flow characteristic curve and a torque characteristic curve is the same, and an ideal network is obtained.
Further, in step 3, the approximation of the full-characteristic nonlinear relation of the pump turbine is completed, so that the data obtained after training is drawn into a space curved surface, specifically:
(1) establishing a network; the feedforwardnet function is used to create the BP network, calling the format feedforwardnet (hiddenSizes, rainfcn), where hiddenSizes are the number of hidden layer nodes, rainfcn is the training function, default is rainlm, both the flow and torque characteristics use the BP network, the number of hidden layer nodes is set to 50, and the creation function is as follows:
net1＝feedforwardnet(50)；
net2＝feedforwardnet(50)；
(2) training a network; using a tranlmm function as a training function in a BP neural network, defaulting to the tranlmm function when the network is established in the step (1), wherein only a sample needs to be input for training, and processed sample data are respectively stored in matrixes N, Q and M, wherein the matrix N is a two-dimensional array containing normalized guide vane opening and unit rotating speed, Q is unit flow after normalization, and M is unit torque after normalization;
(3) simulating a network; after the network is built and trained, the built network can be simulated, the simulation calling function is y-sim (net, x), wherein input x is a two-dimensional array, namely the combination of the guide vane opening and the unit rotating speed, and output y is one-dimensional flow or torque.
The invention characterizes the function Q according to the full characteristics of the pump turbine11＝fQ(α,n11) And M11＝fM(α,n11) A network structure with full characteristics is established, the spatial curved surface representation of a full characteristic curve is completed on the basis, and the phenomena of crossing, aggregation, twisting and the like of equal-opening-degree lines are eliminated; the BP neural network is used for fitting, and the method has the characteristics of high precision, strong generalization capability, capability of meeting the requirement of continuous guidance of a curved surface and the like.
Drawings
FIG. 1 is a flow chart of an approximation method based on a BP neural network full characteristic curve function according to the present invention;
FIG. 2 is a schematic diagram of a three-layer BP neural network;
FIG. 3 is a schematic structural diagram of a flow or torque characteristic curve network of a pump turbine based on a three-layer BP neural network;
FIG. 4 is a schematic diagram of an untrained network for flow characteristics and torque characteristics;
FIG. 5 is a schematic diagram of a flow characteristic training process;
fig. 6 is a schematic diagram of a torque characteristic training process.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
With reference to fig. 1 to fig. 3, the present invention provides an approximation method based on a full characteristic curve function of a BP neural network, the method includes the following steps:
step 1, collecting sample data of a full characteristic curve of a water pump and water turbine model test;
step 2, normalizing and extending the collected data;
and 3, establishing a three-layer BP neural network to train the processed data, obtaining an ideal network, and completing approximation of the full-characteristic nonlinear relation of the water pump turbine, so that the data obtained after training is drawn into a space curved surface.
The mathematical expression of the characteristics of the pump turbine is that the runner flow Q and the runner moment M of the francis turbine are related to the water head H of the turbine, the guide vane opening α and the runner rotating speed n, and the function expression is as follows:
Figure BDA0002353596220000031
according to the related parameters of the water pump and water turbine model test with the rotating wheel model number A1131, the unit rotating speed n under the opening degree of 13 guide vanes is obtained11Unit flow rate Q11And unit torque M11Further, a flow rate characteristic curve (Q) in a two-dimensional plane is plotted11-n11) And torque characteristic curve (M)11-n11)；
Carrying out continuation and normalization processing on the sample data: the sample data only has unit flow and unit rotating speed under 13 guide vane opening degrees, the guide vane opening degree intervals are different, all the opening degrees are not covered, and the flow values and the torque value numbers under different opening degrees are also different. When the neural network is trained, a network with global property and relatively high precision cannot be obtained due to insufficient sample data and conditions, so that extending the test data is a necessary step for improving the network precision.
The normalization uses a mapminmax () function, and limits parameters in sample data to be between [0 and 1] by utilizing the principle of probability statistics, so that the normalization aims to solve the problems caused by different dimensions and orders of magnitude of input learning samples, greatly accelerate the speed of network training and prevent neurons from being saturated due to overlarge absolute value of net input.
In the process of continuation, two principles need to be followed: (1) when the opening of the guide vane is 0, the flow of the water turbine is 0 no matter how large the rotating speed is, so that the boundary condition when the opening of the guide vane is 0 can be constructed, and the flow characteristic of the water turbine at the runaway rotating speed also needs to be considered; (2) the output torque of the water turbine during the runaway is 0, and the torque boundary condition of the water turbine under the opening degree of 0 is also considered.
After the normalization processing and continuation of the sample data are completed, training of the BP neural network is started, and in step 3, the change expression equation of the water pump turbine parameters is as follows:
Figure BDA0002353596220000041
wherein Q is11Is the unit flow rate of the pump turbine, M11Unit torque, α guide vane opening, n11Is unit rotating speed; the unit flow and the unit torque are both functions formed by the opening degree of the guide vane and the unit rotating speed, therefore, the network to be established is a three-layer BP neural network with double inputs and single outputs and composed of a plurality of neurons, the hidden layer nodes of the flow characteristic curve and the torque characteristic curve are the same, and therefore an ideal network is obtained, as shown in fig. 3, the unit rotating speed n in fig. 3 is11The opening α of the guide vane is used as input parameter, and the unit flow rate Q11And unit torque M11In order to be the output parameter,
ω11、ω12、ω13、…、ω1nis the network input weight value omega corresponding to α21、ω22、ω23、…、ω2nIs n11A corresponding network weight; b1、b2、b3、…、bnInputting a threshold value for the first layer network; n is1、n2、n3、…、nnCalculating the output value of the first layer network after the weight value and the threshold value; (n) is the selected activation function; omega1、ω2、ω3、…、ωnInvisible weight value of the activation function; b0Is the output layer weight; n is0Outputting for the network; NN (neural network)outIs the processed output value.
In step 3, the approximation of the full-characteristic nonlinear relation of the pump turbine is completed, so that the data obtained after training is drawn into a space curved surface, specifically:
(1) establishing a network; the feedforwardnet function is used to create the BP network, calling the format feedforwardnet (hiddenSizes, rainfcn), where hiddenSizes are the number of hidden layer nodes, rainfcn is the training function, default is rainlm, both the flow and torque characteristics use the BP network, the number of hidden layer nodes is set to 50, and the creation function is as follows:
net1＝feedforwardnet(50)；
net2＝feedforwardnet(50)；
the created network is shown in fig. 4;
(2) training a network; using a tranlmm function as a training function in a BP neural network, defaulting to the tranlmm function when the network is established in the step (1), wherein only a sample needs to be input for training, and processed sample data are respectively stored in matrixes N, Q and M, wherein the matrix N is a two-dimensional array containing normalized guide vane opening and unit rotating speed, Q is unit flow after normalization, and M is unit torque after normalization;
the MATLAB code of the training function is as follows:
net1＝train(net1,N,Q)；
net2＝train(net2,N,M)；
the training process is shown in fig. 5 and fig. 6, and fig. 5 is a schematic diagram of the flow characteristic training process; fig. 6 is a schematic diagram of a torque characteristic training process. The training process of unit flow goes through 71 steps for 6 seconds, and the final mean square error is 3.32 multiplied by 10-4(ii) a The training of unit torque goes through 42 steps for 3 seconds, and the final mean square error is 6.93 multiplied by 10-5。
(3) Simulating a network; after the network is built and trained, the built network can be simulated, the simulation calling function is y-sim (net, x), wherein input x is a two-dimensional array, namely the combination of the guide vane opening and the unit rotating speed, and output y is one-dimensional flow or torque.
The approximation method based on the full characteristic curve function of the BP neural network, which is provided by the invention, is described in detail, a specific example is applied in the text to explain the principle and the implementation mode of the invention, and the description of the above embodiment is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.

Claims (5)

1. An approximation method based on a BP neural network full characteristic curve function is characterized in that: the method comprises the following steps:
step 1, collecting sample data of a full characteristic curve of a water pump and water turbine model test;
step 2, normalizing and extending the collected data;
and 3, establishing a three-layer BP neural network to train the processed data, obtaining an ideal network, and completing approximation of the full-characteristic nonlinear relation of the water pump turbine, so that the data obtained after training is drawn into a space curved surface.
2. The method of claim 1, wherein: the normalization uses the mapminmax () function to limit the parameters in the sample data to be between [0,1] using the principle of probability statistics.
3. The method of claim 2, wherein: in the process of continuation, two principles need to be followed: (1) when the opening of the guide vane is 0, the flow of the water turbine is 0 no matter how large the rotating speed is, so that the boundary condition when the opening of the guide vane is 0 can be constructed, and the flow characteristic of the water turbine at the runaway rotating speed also needs to be considered; (2) the output torque of the water turbine during the runaway is 0, and the torque boundary condition of the water turbine under the opening degree of 0 is also considered.
4. The method of claim 3, wherein: in step 3, the expression equation of the change of the pump turbine parameter is as follows:
Figure FDA0002353596210000011
wherein Q is11Is the unit of the pump turbineFlow rate, M11Unit torque, α guide vane opening, n11Is unit rotating speed; the unit flow and the unit torque are both functions formed by the opening degree of the guide vane and the unit rotating speed, therefore, the network to be established is a three-layer BP neural network which has double inputs and single outputs and is composed of a plurality of neurons, the number of nodes of hidden layers of a flow characteristic curve and a torque characteristic curve is the same, and an ideal network is obtained.
5. The method of claim 4, wherein: in step 3, the approximation of the full-characteristic nonlinear relation of the pump turbine is completed, so that the data obtained after training is drawn into a space curved surface, specifically:
(1) establishing a network; the feedforwardnet function is used to create the BP network, calling the format feedforwardnet (hiddenSizes, rainfcn), where hiddenSizes are the number of hidden layer nodes, rainfcn is the training function, default is rainlm, both the flow and torque characteristics use the BP network, the number of hidden layer nodes is set to 50, and the creation function is as follows:
net1＝feedforwardnet(50)；
net2＝feedforwardnet(50)；
(2) training a network; using a tranlmm function as a training function in a BP neural network, defaulting to the tranlmm function when the network is established in the step (1), wherein only a sample needs to be input for training, and processed sample data are respectively stored in matrixes N, Q and M, wherein the matrix N is a two-dimensional array containing normalized guide vane opening and unit rotating speed, Q is unit flow after normalization, and M is unit torque after normalization;
(3) simulating a network; after the network is built and trained, the built network can be simulated, the simulation calling function is y-sim (net, x), wherein input x is a two-dimensional array, namely the combination of the guide vane opening and the unit rotating speed, and output y is one-dimensional flow or torque.
CN202010001277.1A 2020-01-02 2020-01-02 Approximation method based on BP neural network full characteristic curve function Pending CN111144027A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010001277.1A CN111144027A (en) 2020-01-02 2020-01-02 Approximation method based on BP neural network full characteristic curve function

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010001277.1A CN111144027A (en) 2020-01-02 2020-01-02 Approximation method based on BP neural network full characteristic curve function

Publications (1)

Publication Number Publication Date
CN111144027A true CN111144027A (en) 2020-05-12

Family

ID=70523234

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010001277.1A Pending CN111144027A (en) 2020-01-02 2020-01-02 Approximation method based on BP neural network full characteristic curve function

Country Status (1)

Country Link
CN (1) CN111144027A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112464478A (en) * 2020-11-30 2021-03-09 中国长江电力股份有限公司 Control law optimization method and device for water turbine speed regulating system
CN112904721A (en) * 2021-01-18 2021-06-04 武汉大学 Coordinated control method for variable-speed pumped storage unit
CN115573926A (en) * 2022-11-21 2023-01-06 南京群顶科技股份有限公司 Machine room water pump energy-saving operation method combining BP neural network fitting characteristic curve

Non-Patent Citations (9)

* Cited by examiner, † Cited by third party
Title
刘志淼;张德虎;张醒;刘莹莹;: "水泵水轮机全特性的集成PSO_BP神经网络模型" *
刘志淼;张德虎;张醒;刘莹莹;: "水泵水轮机全特性的集成PSO_BP神经网络模型", 中国农村水利水电, no. 10, pages 1 - 4 *
张培;陈光大;张旭;: "BP和RBF神经网络在水轮机非线性特性拟合中的应用比较", no. 11 *
童星;把多铎;杨京广;: "轴流转桨式水轮机特性神经网络三维建模", no. 06 *
谭剑波;宋亮;冯建栋;: "水轮机综合调节特性模糊神经网络建模仿真", no. 04 *
谭剑波;马孝义;何自立;: "水轮机特性曲面型值点延拓神经网络仿真研究", no. 06 *
谭剑波等: "基于BP神经网络的水轮机综合特性建模仿真", 中国农村水利水电, no. 3, pages 1 - 4 *
谭建波等: "基于BP神经网络的水轮机综合特性建模仿真" *
黄文涛;常黎;黄正军;刘德祥;张登;: "基于BP网络的水泵水轮机全特性空间曲面描述", no. 12 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112464478A (en) * 2020-11-30 2021-03-09 中国长江电力股份有限公司 Control law optimization method and device for water turbine speed regulating system
CN112464478B (en) * 2020-11-30 2023-06-30 中国长江电力股份有限公司 Control law optimization method and device for water turbine speed regulation system
CN112904721A (en) * 2021-01-18 2021-06-04 武汉大学 Coordinated control method for variable-speed pumped storage unit
CN115573926A (en) * 2022-11-21 2023-01-06 南京群顶科技股份有限公司 Machine room water pump energy-saving operation method combining BP neural network fitting characteristic curve

Similar Documents

Publication Publication Date Title
CN112131760B (en) CBAM model-based prediction method for residual life of aircraft engine
CN111144027A (en) Approximation method based on BP neural network full characteristic curve function
WO2019127363A1 (en) Weight coding method for neural network, computing apparatus, and hardware system
CN108062572A (en) A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on DdAE deep learning models
CN104751227B (en) Construction method and system for the deep neural network of speech recognition
WO2016145676A1 (en) Big data processing method based on deep learning model satisfying k-degree sparse constraint
CN104700828A (en) Deep long-term and short-term memory recurrent neural network acoustic model establishing method based on selective attention principles
CN107393542A (en) A kind of birds species identification method based on binary channels neutral net
CN108231086A (en) A kind of deep learning voice enhancer and method based on FPGA
CN104952448A (en) Method and system for enhancing features by aid of bidirectional long-term and short-term memory recurrent neural networks
CN105512289A (en) Image retrieval method based on deep learning and Hash
CN104751228A (en) Method and system for constructing deep neural network
CN106897254A (en) A kind of network representation learning method
CN111753207B (en) Collaborative filtering method for neural map based on comments
CN114722690B (en) Acoustic super-surface sound field rapid prediction method based on variable reliability neural network
CN113157919B (en) Sentence text aspect-level emotion classification method and sentence text aspect-level emotion classification system
CN106778910A (en) Deep learning system and method based on local training
CN112578089B (en) Air pollutant concentration prediction method based on improved TCN
CN108470061A (en) A kind of emotional semantic classification system for visual angle grade text
CN106297778A (en) The neutral net acoustic model method of cutting out based on singular value decomposition of data-driven
CN110728186B (en) Fire detection method based on multi-network fusion
CN106021880B (en) Jacket platform structural response calculation method based on BP neural network
CN109359190B (en) Method for constructing vertical analysis model based on evaluation object formation
CN109800517A (en) Improved reverse modeling method for magnetorheological damper
CN116644865A (en) Commercial vehicle fuel consumption prediction method, electronic equipment and storage medium

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