CN103258234A - Method for predicting mechanical properties of inertial container based on particle swarm optimization BP neural network - Google Patents
Method for predicting mechanical properties of inertial container based on particle swarm optimization BP neural network Download PDFInfo
- Publication number
- CN103258234A CN103258234A CN2013101580438A CN201310158043A CN103258234A CN 103258234 A CN103258234 A CN 103258234A CN 2013101580438 A CN2013101580438 A CN 2013101580438A CN 201310158043 A CN201310158043 A CN 201310158043A CN 103258234 A CN103258234 A CN 103258234A
- Authority
- CN
- China
- Prior art keywords
- container
- network
- neural network
- particle
- mechanical property
- 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.)
- Granted
Links
Images
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本发明公开了一种基于粒子群优化BP神经网络的惯容器力学性能预测方法，包括如下步骤：（1）进行惯容器的力学性能试验，获取惯容器在不同惯容系数和不同激振输入下的力学响应；（2）根据试验结果得到与惯容器力学性能相关的试验数据，并建立基于BP神经网络的惯容器力学性能预测模型；（3）将网络的权值和阈值定义为粒子群搜索空间中的一个粒子，并对算法的相关参数进行初始化；（4）基于粒子群优化算法对粒子进行迭代寻优，根据结束条件确定最优个体，并将其作为BP网络的权值和阈值；（5）利用经过优化的网络对惯容器的力学性能进行预测。本发明可以对类似机械元件的力学性能进行有效地预测。
The invention discloses a method for predicting the mechanical properties of an inerter based on particle swarm optimization BP neural network, which includes the following steps: (1) conducting a mechanical performance test of the inerter to obtain the inerter under different inertia coefficients and different excitation inputs (2) According to the test results, the experimental data related to the mechanical properties of the container is obtained, and the prediction model of the mechanical properties of the container based on the BP neural network is established; (3) The weight and threshold of the network are defined as particle swarm search A particle in the space, and initialize the relevant parameters of the algorithm; (4) Iteratively optimize the particles based on the particle swarm optimization algorithm, determine the optimal individual according to the end condition, and use it as the weight and threshold of the BP network; (5) Use the optimized network to predict the mechanical properties of the inerter. The invention can effectively predict the mechanical properties of similar mechanical elements.
Description
Technical field
The present invention relates to a kind of container mechanical property prediction method of being used to, refer in particular to be based on particle group optimizing BP neural network be used to the container mechanical property prediction method.
Background technology
Be used to container since being suggested, be one of focus of mechanical vibration isolation area research always, and The experimental results shows is used to the anti-vibration performance that container can improve vibrating isolation system.Yet these research major parts are that to be used to container ideal linearity mathematical model be prerequisite, have ignored and have been used to the non-linear factor that exists in the container practical structures.Find that by actual tests non-linear factor has remarkable influence to the mechanical property of being used to container, and influence mechanism is complicated and changeable, often can only describe qualitatively it, is difficult to fundamentally grasp and is used to the container dynamic behaviour under the actual condition.Artificial neural network is based on the network model that biological neural network grows up, and has the advantage of handling strong nonlinearity, and therefore, the present invention considers to set up and is used to container adaptive neural network model, is used to the mechanical properties prediction of container.
The BP neural network has self-organization, self study and adaptive ability, principle simply is easy to realize, be used widely in various fields, but it adopts the gradient descent method that weights and the threshold value of network are revised in learning process, shortcomings such as existing and be absorbed in local optimum easily, speed of convergence is slow and generalization ability is weak.
Particle cluster algorithm is a kind of random optimization algorithm based on colony intelligence, its fast convergence rate, ability of searching optimum are strong, and need be by the characteristic information of problem itself, avoided requiring in the gradient descent method process of gradient, also avoid the selection in the genetic algorithm, intersection, variation etc. to evolve and operated, shortened the training time of neural network.There are some researches show, utilize particle cluster algorithm that the training process of BP networks is optimized, can improve the extensive performance of neural network, improve the precision of prediction of network.
Summary of the invention
The objective of the invention is to propose a kind of based on particle group optimizing BP neural network be used to the container mechanical property prediction method, realize being used to the good predict of container mechanical property.
The technical solution adopted in the present invention is:
Based on particle group optimizing BP neural network be used to the container mechanical property prediction method, comprise the steps:
(1) is used to the mechanical property test of container, by the flywheel of the different quality of packing into, obtains and be used to container and be used to hold mechanical response under coefficient and the input of different excitings in difference;
(2) obtain the test figure relevant with being used to the container mechanical property according to test findings, and set up and to be used to container mechanical properties prediction model based on the BP neural network;
(3) weights of BP network and threshold value are defined as a particle in the population search volume, and the correlation parameter of algorithm is carried out initialization;
(4) based on particle swarm optimization algorithm particle is carried out the iteration optimizing, determine optimum individual according to termination condition, and with its weights and threshold value as the BP network;
(5) the BP network that utilizes process to optimize is predicted the mechanical property of being used to container.
Described exciting is input as the sinusoidal period input, and different frequencies and corresponding amplitude are adopted in each input.
The described container mechanical properties prediction model of being used to is three layers of BP network, wherein the neuron number of input layer, hidden layer, output layer is respectively 4,5,1, the transport function of described hidden layer neuron is tanh S type function, and the neuronic transport function of output layer is linear transfer function.
Being used to being input as of container mechanical properties prediction model is used to hold coefficient and is used to displacement, speed and the acceleration of container under a plurality of transient time points, be output as the power output of being used to container, described displacement of being used to container is to preserve directly in real time by actuating vibration table to obtain, on this basis the displacement input is carried out curve fitting, then curvilinear equation is carried out the difference differentiate, just can obtain described speed and the acceleration of being used to container.
Described a plurality of transient time point refers to the sampled point of test figure, because exciting is input as the cycle input, in order to raise the efficiency, the duration of will sampling is decided to be two cycles, sample frequency under each cycle is 12Hz, and therefore, it is 24 that every kind of transient time of testing under the operating mode counts.
Termination condition in the described step (4) satisfies the least error requirement for reaching iterations or network error; Described network error computation process comprises the steps: that (1) give BP network with network weight and the threshold value of particle correspondence; (2) calculate each training sample according to the output valve of BP network forward direction, by comparing with expectation value, draw the error of each training sample; (3) calculate the error mean square root of all training samples, namely get network error.
The present invention can predict that its beneficial effect is effectively to the mechanical property of similar mechanical organ:
1. on the basis of being used to the container mechanical property test, obtain being used to the influential reference data of container power output, the gained data are reliably reasonable, guaranteed the training quality of network.
At reference data set up based on the BP neural network be used to container mechanical properties prediction model, and be optimized with the training process of particle swarm optimization algorithm to the BP network, improved the precision of prediction of network, realized being used to the good predict of container mechanical property.
Description of drawings
Fig. 1 is the overview flow chart of being used to the container mechanical property prediction method based on particle group optimizing BP neural network;
Fig. 2 is for being used to the chamber test arrangenent diagram in the embodiment;
Fig. 3 is for being used to container mechanical properties prediction model structure figure;
Fig. 4 is predicated error comparison diagram before and after the network optimization.
Embodiment
The invention will be further described below in conjunction with the drawings and specific embodiments.
Of the present invention based on particle group optimizing BP neural network be used to the container mechanical property prediction method, overall procedure specifically comprises the steps: as shown in Figure 1
(1) is used to the mechanical property test of container.
The subjects that adopts among the embodiment is that ball screw type is used to container, be used to container and adopt three kinds of flywheels that vary in size, be used to hold coefficient accordingly and be respectively 30kg, 130kg and 332kg, testing equipment is numerical control hydraulic servo vibration exciting testing table, test is arranged as shown in Figure 2, this testing table can support exciting head 1 according to certain displacement request motion, and displacement and the load signal of real-time monitored and preservation exciting head.To be used to the upper end of container 3 fixes by latch 4, the lower end of being used to container 3 links to each other with exciting head 1 by latch 2, sinusoidal input is adopted in test, the correlation parameter of exciting input is as shown in table 1, in order to prevent that test force from exceeding the load of actuating vibration table, when dither, adopted less amplitude, obtained 51 kinds of container 3 mechanical properties of being used to of testing under the operating mode by test and responded.
The correlation parameter of table 1 exciting input
(2) obtain the test figure relevant with being used to container 3 mechanical properties.
The parameter that container 3 mechanical properties are used in influence also comprises displacement, speed and the acceleration of being used to container 3 except being used to hold the coefficient, wherein be used to the displacement of container 3 and can preserve directly acquisition in real time by actuating vibration table, in order to obtain to be used to speed and the acceleration of container 3, displacement input carries out curve fitting by Matlab curve match tool box CFTOOL, then curvilinear equation is carried out the difference differentiate, just can obtain to be used to speed and the acceleration of container 3.
(3) set up and to be used to container 3 mechanical properties prediction models based on the BP neural network.
Being input as of model is used to hold coefficient and is used to displacement, speed and the acceleration of container 3 under a plurality of transient time points, be output as and be used to the container power output under corresponding transient time point, obtaining of transient time point is by test figure is sampled, sampling period is two cycles, sample frequency under each cycle is 12Hz, therefore to count be 24 the transient time under every kind of test operating mode, and the final data that obtain to be used for network training and test have 51 * 24=1224 group.Because model has 4 inputs, 1 output, therefore the structure of determining network is 4-5-1, the neuron number that is input layer, hidden layer, output layer is respectively 4,5,1, as Fig. 3, the transport function of hidden layer neuron is tanh S type function, and the neuronic transport function of output layer is linear transfer function.
(4) be optimized based on the training process of particle swarm optimization algorithm to the BP network.
According to topology of networks, the number summation that obtains network weight and threshold value is 4 * 5+5 * 1+5 * 1+1=31, simultaneously the weights of network and threshold value is defined as a particle in the population search volume, and the dimension of particle is 31.Population number, velocity range, inertia weight w, acceleration parameter r to particle
1And r
2Carry out initialization, maximum iteration time is decided to be 300, least error requires to be decided to be 0.1.
Based on particle swarm optimization algorithm particle is carried out the iteration optimizing, idiographic flow is as follows:
1. produce all particle position x and speed v at random;
2. import a particle, calculating its fitness is above-mentioned network error;
3. continue other particles of input, calculate the fitness of all particles;
4. determine self desired positions P of particle
tAnd the fitness pbest under this position and overall desired positions G
tAnd the fitness gbest under this position;
5. upgrade all particle's velocity and position according to following formula;
In the formula, be that interval [0,1] goes up equally distributed random number rand();
6. judge whether particle's velocity surpasses maximal rate, if surpass, speed is updated to maximal rate, whether judges particle's velocity less than minimum speed, if less than, speed is updated to minimum speed, other situation speed are normally upgraded;
7. recomputate the particle fitness after the renewal, and upgrade P
tWith pbest and G
tAnd gbest, if the fitness＜pbest of current particle, upgrading pbest is the fitness of current particle, corresponding particle position is self desired positions P
t, otherwise P
tConstant with pbest; If the fitness＜gbest of current particle, upgrading gbest is the fitness of current particle, and corresponding particle position is overall desired positions G
t, otherwise G
tConstant with gbest;
8. continue the next round iteration, until satisfying termination condition.
(5) utilize the mechanical property of being used to container through the network prediction of optimizing.
By obtaining final optimum individual after the above-mentioned iteration optimizing, and give BP network with individual corresponding network weight and threshold value, thereby obtain through being used to container mechanical properties prediction model after optimizing.Network 1100 groups of data of picked at random in 1224 groups of data are used for the optimizing process of step (4), remaining 124 groups of model prediction precision that are used for after check is optimized.The predicated error of network is to such as Fig. 4 before and after optimizing, can obviously find out through the network estimated performance after optimizing from Fig. 4 and to be better than without the network of optimizing, be 10.4 by the network predicated error root-mean-square value that calculates after the optimization, network predicated error root-mean-square value without optimization is 19.1, the result shows, has reduced by 45.5% through optimizing back network predicated error.
Above result shows, that adopts that the inventive method sets up is used to container mechanical properties prediction model and can predicts well the mechanical property of being used to container, and precision of prediction had increased significantly before optimizing, the estimated performance of network is improved, for the mechanical properties prediction of being used to container and similar mechanical organ provides effective method.
Claims (6)
- Based on particle group optimizing BP neural network be used to the container mechanical property prediction method, it is characterized in that, comprise the steps:(1) is used to the mechanical property test of container, by the flywheel of the different quality of packing into, obtains and be used to container and be used to hold mechanical response under coefficient and the input of different excitings in difference;(2) obtain the test figure relevant with being used to the container mechanical property according to test findings, and set up and to be used to container mechanical properties prediction model based on the BP neural network;(3) weights of BP network and threshold value are defined as a particle in the population search volume, and the correlation parameter of algorithm is carried out initialization;(4) based on particle swarm optimization algorithm particle is carried out the iteration optimizing, determine optimum individual according to termination condition, and with its weights and threshold value as the BP network;(5) the BP network that utilizes process to optimize is predicted the mechanical property of being used to container.
- According to claim 1 based on particle group optimizing BP neural network be used to the container mechanical property prediction method, it is characterized in that the exciting in the described step (1) is input as the sinusoidal period input, different frequencies and corresponding amplitude are adopted in each input.
- 3. according to claim 1ly be used to the container mechanical property prediction method based on particle group optimizing BP neural network, it is characterized in that, the container mechanical properties prediction model of being used in the described step (2) is three layers of BP network, and wherein the neuron number of input layer, hidden layer, output layer is respectively 4,5,1; The transport function of described hidden layer neuron is tanh S type function, and the neuronic transport function of output layer is linear transfer function; Being input as of described model is used to hold coefficient and is used to displacement, speed and the acceleration of container under a plurality of transient time points, is output as and is used to the container power output.
- 4. according to claim 3ly be used to the container mechanical property prediction method based on particle group optimizing BP neural network, it is characterized in that, described displacement of being used to container is to preserve directly in real time by actuating vibration table to obtain, on this basis the displacement input is carried out curve fitting, then curvilinear equation is carried out the difference differentiate, just can obtain described speed and the acceleration of being used to container; Described a plurality of transient time point refers to test figure is carried out uniform sampling, and the sampling period is two cycles, and the sample frequency under each cycle is 12Hz.
- According to claim 1 based on particle group optimizing BP neural network be used to the container mechanical property prediction method, it is characterized in that the termination condition in the described step (4) satisfies the least error requirement for reaching iterations or network error.
- According to claim 5 based on particle group optimizing BP neural network be used to the container mechanical property prediction method, it is characterized in that the computation process of described network error comprises the steps:(1) gives BP network with network weight and the threshold value of particle correspondence;(2) calculate each training sample according to the output valve of BP network forward direction, by comparing with expectation value, draw the error of each training sample;(3) calculate the error mean square root of all training samples, namely get network error.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310158043.8A CN103258234B (en) | 2013-05-02 | 2013-05-02 | Based on the used container mechanical property prediction method of particle group optimizing BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310158043.8A CN103258234B (en) | 2013-05-02 | 2013-05-02 | Based on the used container mechanical property prediction method of particle group optimizing BP neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103258234A true CN103258234A (en) | 2013-08-21 |
| CN103258234B CN103258234B (en) | 2015-10-28 |
Family
ID=48962136
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310158043.8A Expired - Fee Related CN103258234B (en) | 2013-05-02 | 2013-05-02 | Based on the used container mechanical property prediction method of particle group optimizing BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103258234B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103810524A (en) * | 2014-03-08 | 2014-05-21 | 辽宁工程技术大学 | Method for predicting ground subsidence in underground metro construction process |
| CN104360035A (en) * | 2014-11-02 | 2015-02-18 | 北京工业大学 | Sewage TP soft measurement method based on self-organized particle swarm and radial basis function neural network |
| CN105136469A (en) * | 2015-07-23 | 2015-12-09 | 江苏大学 | Unmanned vehicle speed control method based on PSO and RBF neutral network |
| CN106156859A (en) * | 2016-06-17 | 2016-11-23 | 江苏大学 | A kind of used container mechanics output Forecasting Methodology |
| CN106503288A (en) * | 2016-09-18 | 2017-03-15 | 江苏大学 | Interconnection hydraulic cylinder mechanical properties prediction algorithm based on Support Vector Machines for Regression |
| CN109146055A (en) * | 2018-09-03 | 2019-01-04 | 北京珈信科技有限公司 | Modified particle swarm optimization method based on orthogonalizing experiments and artificial neural network |
| CN111025899A (en) * | 2019-11-21 | 2020-04-17 | 复旦大学 | A Nonlinear Dynamic Mass System Prediction Method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070010956A1 (en) * | 2003-09-26 | 2007-01-11 | Chahe Nerguizian | Method and system for indoor geolocation using an impulse response fingerprinting technique |
| CN101872433A (en) * | 2010-05-21 | 2010-10-27 | 杭州电子科技大学 | A Beer Taste Prediction Method Based on Neural Network Technology |
| CN102072922A (en) * | 2009-11-25 | 2011-05-25 | 东北林业大学 | Particle swarm optimization neural network model-based method for detecting moisture content of wood |
-
2013
- 2013-05-02 CN CN201310158043.8A patent/CN103258234B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070010956A1 (en) * | 2003-09-26 | 2007-01-11 | Chahe Nerguizian | Method and system for indoor geolocation using an impulse response fingerprinting technique |
| CN102072922A (en) * | 2009-11-25 | 2011-05-25 | 东北林业大学 | Particle swarm optimization neural network model-based method for detecting moisture content of wood |
| CN101872433A (en) * | 2010-05-21 | 2010-10-27 | 杭州电子科技大学 | A Beer Taste Prediction Method Based on Neural Network Technology |
Non-Patent Citations (1)
| Title |
|---|
| 于颖 等: "粒子群算法在工程优化设计中的应用", 《机械工程学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103810524A (en) * | 2014-03-08 | 2014-05-21 | 辽宁工程技术大学 | Method for predicting ground subsidence in underground metro construction process |
| CN104360035A (en) * | 2014-11-02 | 2015-02-18 | 北京工业大学 | Sewage TP soft measurement method based on self-organized particle swarm and radial basis function neural network |
| CN104360035B (en) * | 2014-11-02 | 2016-03-30 | 北京工业大学 | A TP soft-sensing method for sewage total phosphorus based on self-organized particle swarm-radial basis neural network |
| CN105136469A (en) * | 2015-07-23 | 2015-12-09 | 江苏大学 | Unmanned vehicle speed control method based on PSO and RBF neutral network |
| CN106156859A (en) * | 2016-06-17 | 2016-11-23 | 江苏大学 | A kind of used container mechanics output Forecasting Methodology |
| CN106156859B (en) * | 2016-06-17 | 2019-05-31 | 江苏大学 | A kind of used container mechanics output prediction technique |
| CN106503288A (en) * | 2016-09-18 | 2017-03-15 | 江苏大学 | Interconnection hydraulic cylinder mechanical properties prediction algorithm based on Support Vector Machines for Regression |
| CN106503288B (en) * | 2016-09-18 | 2019-06-28 | 江苏大学 | Interconnection hydraulic cylinder mechanical property prediction method based on Support Vector Machines for Regression |
| CN109146055A (en) * | 2018-09-03 | 2019-01-04 | 北京珈信科技有限公司 | Modified particle swarm optimization method based on orthogonalizing experiments and artificial neural network |
| CN111025899A (en) * | 2019-11-21 | 2020-04-17 | 复旦大学 | A Nonlinear Dynamic Mass System Prediction Method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103258234B (en) | 2015-10-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103258234A (en) | Method for predicting mechanical properties of inertial container based on particle swarm optimization BP neural network | |
| CN112926117A (en) | BAS-BP-based pile foundation horizontal bearing capacity prediction method | |
| CN105260786B (en) | A kind of simulation credibility of electric propulsion system assessment models comprehensive optimization method | |
| CN104899431B (en) | Based on ant colony and swarm of particles into LSSVM fluctuating wind speed Forecasting Methodologies | |
| CN104991997B (en) | The broad sense rate correlation P-I hysteresis modeling methods of adaptive differential evolution algorithm optimization | |
| CN105138717A (en) | Transformer state evaluation method by optimizing neural network with dynamic mutation particle swarm | |
| CN102562239A (en) | Method for monitoring exhaust temperature of aircraft engine | |
| CN109871622A (en) | A method and system for calculating line loss in low-voltage station area based on deep learning | |
| CN110147611A (en) | A kind of Structural Damage Identification based on IWOA | |
| CN113884936A (en) | A method for predicting the state of health of lithium-ion batteries based on ISSA coupled DELM | |
| CN105335619A (en) | Collaborative optimization method applicable to parameter back analysis of high calculation cost numerical calculation model | |
| KR20220111215A (en) | Apparatus and method for predicting drug-target interaction using deep neural network model based on self-attention | |
| CN109508498A (en) | Rubber shock absorber formula designing system and method based on BP artificial neural network | |
| CN106710222A (en) | Method and device for predicting traffic flow | |
| CN112966399B (en) | A method and system for predicting working conditions of a pulse tube refrigerator based on machine learning | |
| CN117421991A (en) | Numerical wind tunnel inlet wind speed intelligent control method based on deep reinforcement learning | |
| CN110110380A (en) | A kind of piezo actuator Hysteresis Nonlinear modeling method and application | |
| Ru et al. | A GA and particle swarm optimization based hybrid algorithm | |
| CN105184112A (en) | Protein structure prediction method based on improved niche genetic algorithm | |
| CN118818974A (en) | A piezoelectric ceramic hysteresis modeling parameter identification method, system, device and medium | |
| CN118962457A (en) | Lithium battery SOH prediction method based on TimesNet-BiGRU neural network | |
| CN110489800B (en) | Structural dynamic load sparse identification method based on matrix regularization | |
| Zainuddin et al. | Comparison of crossover in genetic algorithm for discrete-time system identification | |
| Li et al. | Parameter identification based on PSO algorithm for piezoelectric actuating system with rate-dependent Prandtl-Ishlinskii hysteresis modeling method | |
| Liu et al. | Parameter estimation of piezoelectric transducers circuit model using GA-PSO algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C53 | Correction of patent of invention or patent application | ||
| CB03 | Change of inventor or designer information |
Inventor after: Sun Xiaoqiang Inventor after: Chen Long Inventor after: Wang Shaohua Inventor after: Zhang Xiaoliang Inventor before: Chen Long Inventor before: Sun Xiaoqiang Inventor before: Zhang Xiaoliang Inventor before: Wang Ruochen |
|
| COR | Change of bibliographic data |
Free format text: CORRECT: INVENTOR; FROM: CHEN LONG SUN XIAOQIANG ZHANG XIAOLIANG WANG RUOCHEN TO: SUN XIAOQIANG CHEN LONG WANG SHAOHUA ZHANG XIAOLIANG |
|
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20151028 |