CN108564172A - Organic Rankine Cycle output power predicting method based on BP neural network - Google Patents
Organic Rankine Cycle output power predicting method based on BP neural network Download PDFInfo
- Publication number
- CN108564172A CN108564172A CN201810361923.8A CN201810361923A CN108564172A CN 108564172 A CN108564172 A CN 108564172A CN 201810361923 A CN201810361923 A CN 201810361923A CN 108564172 A CN108564172 A CN 108564172A
- Authority
- CN
- China
- Prior art keywords
- neural network
- data
- output power
- rankine cycle
- organic rankine
- 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
Links
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
Abstract
Organic Rankine Cycle output power predicting method based on BP neural network, belongs to Thermal Power Engineering Field.Include the following steps：The acquisition of multigroup Organic Rankine Cycle operation data, every group includes following parameter：Volume flow, expanding machine torque, expander inlet pressure, outlet pressure of expansion machine, expander inlet temperature, condensator outlet temperature, working medium pump discharge pressure and the expanding machine output power of organic working medium；The normalized of gathered data；The foundation of BP neural network；The training and test of neural network；The anti-normalization processing of neural network output data, that is, expanding machine output power in test data.Compared with traditional thermodynamics analysis methods, the neural network model that the present invention is built can fast and accurately predict Organic Rankine Cycle output power, avoid the operation mechanism research to various parts, and it can be obviously improved working efficiency and precision of prediction, reliable guide is provided for the optimization of Organic Rankine Cycle.
Description
Technical field
The present invention relates to a kind of Organic Rankine Cycle output power predicting method based on BP neural network, belongs to thermal energy work
Journey field.
Background technology
During China industrializes fast-developing, there is a large amount of waste heat energies, such as：Fume afterheat, cooling medium
Waste heat, waste vapour waste water residual heat, high-temperature product and afterheat of slags, chemical reaction heat and combustible exhaust gas and waste material waste heat etc., in addition to
Can by technological transformation upgrade realize to these waste heat energies it is efficient utilization except, heat recovery be raising economy,
An energy saving important channel.
In numerous residual-heat utilization technologies, Organic Rankine Cycle is since its efficiency is higher, simple in structure, single-machine capacity model
70 DEG C or more of residual heat resources can be widely used in the advantages that standard modular production by enclosing wide, various parts.In recent years, state
Inside and outside scholar is to improve Organic Rankine Cycle output power to have carried out a large amount of research work.For the research of Organic Rankine Cycle
Mainly by establishing the theoretical analysis method after each component thermodynamical model.But Organic Rankine Cycle actual moving process
In influence factor it is more, be typical non-linear relation between output power and each thermodynamic parameter.At present, it tests
System output power differs larger with theoretical analysis result, i.e., theory analysis can not test output power to Organic Rankine Cycle
Accurately predicted.
BP neural network is also known as error backward propagation method, is a kind of forward direction type network of multilayer.In BP nerves
In network, signal is propagated forward, and error is backpropagation.BP neural network has very strong non-linear mapping capability
With network structure flexible, it is widely used in the prediction and optimization of energy power system performance.Therefore, foundation has
The BP neural network model of machine Rankine cycle, accurately predicts its output power.
Invention content
The purpose of the present invention is to solve existing thermodynamical models to be difficult to Accurate Prediction Organic Rankine Cycle output power
Deficiency, propose a kind of Organic Rankine Cycle output power predicting method based on BP neural network.The specific technology of the present invention
Scheme includes the following steps：
(1) acquisition of Organic Rankine Cycle operation data
Multigroup test data in organic rankine cycle system operational process is acquired, every group of test data includes following ginseng
Number：The volume flow of organic working mediumExpanding machine torque (Torexp), expander inlet pressure (pexp,in), expander outlet
Pressure (pexp,out), expander inlet temperature (Texp,in), condensator outlet temperature (Tcon,out), working medium pump discharge pressure
(pp,out) and expanding machine output powerAbove-mentioned multigroup test data is grouped at random, be divided into acclimation group data and
Test group data；
(2) normalized of gathered data
Due to each parameter the order of magnitude difference it is larger, in order to accelerate the convergence rate of neural network, need to each parameter into
Row normalized makes it be converted into the respective value in [0,1] section
In formula, XminFor the minimum value of each parameter, XmaxFor the maximum value of each parameter, X is the gathered data of each parameter,For
Each parameter normalization treated numerical value；
(3) foundation of BP neural network
Establish the prediction model of the Organic Rankine Cycle output power based on BP neural network, the neural network prediction model
Including an input layer, an intermediate hidden layer and an output layer；Wherein, input layer includes 7 neurons：Organic working medium
Volume flow, expanding machine torque, expander inlet pressure, outlet pressure of expansion machine, expander inlet temperature, condensator outlet temperature
Degree, working medium pump discharge pressure；Output layer includes a neuron, the i.e. output power of Organic Rankine Cycle；Hidden layer transmits letter
Number is tansig, and output layer transmission function is purelin；The neural network prediction model can pass through Calling MATLAB nerve net
It realizes in network tool box；The neuron of input layer and output layer is all made of the data after step (2) normalized；
(4) training and test of neural network
Neural network maximum frequency of training and training objective error are set, by the training after normalized in step (2)
Data are input in the neural network that step (3) is established and are trained, after the completion of waiting for neural metwork training, test data is defeated
Enter to neural network and tests；Wherein, training objective error is evaluated using mean square error：
In formula, Q is input layer set, and Y (k) is neural network prediction value, and t (k) is actual test value, is made
It obtains mean square error and is less than training objective error；
(5) in test data neural network output data, that is, expanding machine output power anti-normalization processing
After the completion of being tested neural network according to step (4), neural network output data is subjected to anti-normalization processing：
In formula, YminFor the minimum value of expanding machine output power in all acclimation group data, YmaxAt all domestications
The maximum value of expanding machine output power in reason group data,Output valve is tested for neural network, Y is the survey after anti-normalization processing
Try output valve；
(6) output power of Organic Rankine Cycle is predicted using the neural network of foundation
The data of previous cycle are re-incorporated INTO step (1) by circulation step (1)-(5), will be passed through at step (2) normalization
7 operating parameters after reason are input in the neural network that step (4) is established, and are obtained after corresponding output valve according still further to step
(5) anti-normalization processing is carried out, and then obtains the output power of subsequent cycle prediction；
The beneficial effects of the present invention are：Compared with traditional thermodynamics analysis methods, the nerve net of the invention built
Network model can fast and accurately predict Organic Rankine Cycle output power, avoid and ground to the operation mechanism of various parts
Study carefully, and working efficiency and precision of prediction can be obviously improved, reliable guide is provided for the optimization of Organic Rankine Cycle.
Description of the drawings
Fig. 1 is the structure chart of BP neural network
Fig. 2 is the prediction result and comparison of test results figure of BP neural network model；
Specific implementation mode
The present invention provides a kind of Organic Rankine Cycle output power predicting method based on BP neural network, with reference to
The embodiment of Vehicular internal combustion engine Organic Rankine Cycle residual neat recovering system is next, and the present invention is described in detail.
It is absorbed in evaporator using the exhaust heat of certain diesel engine for automobile as the heat source of Organic Rankine Cycle, organic working medium
Become saturation or superheated steam after diesel engine exhaust waste heat, subsequently enter expanding machine and push its acting, the steam exhaust after acting enters
Condenser returns to after being condensed into saturated liquid in fluid reservoir, and working medium extraction is sent in evaporator by working medium pump again, hereafter, organic
Rankine cycle carries out cycle operation.
(1) acquisition of Organic Rankine Cycle operation data
During being acquired to Organic Rankine Cycle operation data, by the stable conditions of diesel engine in rotating speed
1900rpm, torque 900N.m, output power 180kW.By adjusting organic working medium circular flow and expanding machine torque, acquisition
The test data of 2100 groups of organic rankine cycle systems operation, including：The volume flow of organic working mediumExpanding machine torque
(Torexp), expander inlet pressure (pexp,in), outlet pressure of expansion machine (pexp,out), expander inlet temperature (Texp,in), it is cold
Condenser outlet temperature (Tcon,out), working medium pump discharge pressure (pp,out) and expanding machine output powerCalling MATLAB
The randperm functions of R2010a randomly select 2000 groups from 2100 groups of test datas and are used as training sample, remaining 100 groups of examination
Data are tested as test sample.
(2) normalized of gathered data
Since the order of magnitude difference of each parameter is larger, in order to accelerate the convergence rate of neural network, by 2100 groups of experiment numbers
According to being normalized, it is made to be converted into the respective value in [0,1] section
In formula, XminFor the minimum value of each parameter, XmaxFor the maximum value of each parameter, X is the gathered data of each parameter,For
Each parameter normalization treated numerical value.
(3) foundation of BP neural network
The Organic Rankine Cycle output based on BP neural network is established by the newff functions of Calling MATLAB R2010a
Power prediction model, the neural network prediction model include an input layer, an intermediate hidden layer and an output layer.Its
In, input layer includes 7 neurons：The volume flow of organic working medium, expanding machine torque, expander inlet pressure, expanding machine go out
Mouth pressure, expander inlet temperature, condensator outlet temperature, working medium pump discharge pressure.Output layer includes a neuron, that is, is had
The output power of machine Rankine cycle；Hidden layer transmission function is tansig, and output layer transmission function is purelin.The nerve net
Network prediction model can be realized by Calling MATLAB R2010a Neural Network Toolbox.
(4) training and test of neural network
It is 2000 that neural network maximum frequency of training, which is arranged, and training objective error is 0.001, learning rate 0.3, study
Function selects the trainlm of L-M optimization algorithms.Wherein, training objective error is evaluated using mean square error：
In formula, Q is input layer set, and Y (k) is neural network prediction value, and t (k) is real output value.
Training data after normalized in step (2) is input in the neural network that step (3) is established and is carried out
Training thinks that neural metwork training is completed when mean square error is less than training objective error 0.001.The results show that by 3
After second, the training of 160 steps, the mean square error of network reaches 0.000486, and training is completed.Then, 100 groups of test datas are input to
Neural network is tested.
(5) anti-normalization processing of test data
After the completion of being tested neural network according to step (4), neural network output data is subjected to anti-normalization processing：
In formula, YminFor expanding machine output power in 2000 groups of test datasMinimum value, YmaxFor 2000 groups of examinations
Test expanding machine output power in dataMaximum value,Output valve is tested for neural network, Y is after anti-normalization processing
Test output valve.
Test output valve after anti-normalization processing is compared with test data, the comparing result meter provided by Fig. 2
It calculates it is found that the maximum relative error of neural network model is 7.7%, related coefficient 0.9934, and most test data
Relative error be less than 5%.It follows that neural network model has higher precision of prediction, this neural network can be used
In the prediction of Organic Rankine Cycle output power.
(6) output power of Organic Rankine Cycle is predicted using the neural network of foundation
It will be input to the neural network that step (4) is established by 7 operating parameters after step (2) normalized
In, it obtains carrying out anti-normalization processing, and then the output power predicted according still further to step (5) after corresponding output valve.
Claims (2)
1. a kind of Organic Rankine Cycle output power predicting method based on BP neural network, which is characterized in that including walking as follows
Suddenly：
(1) acquisition of Organic Rankine Cycle operation data
Multigroup test data in organic rankine cycle system operational process is acquired, every group of test data includes following parameter：Have
The volume flow of machine working mediumExpanding machine torque (Torexp), expander inlet pressure (pexp,in), outlet pressure of expansion machine
(pexp,out), expander inlet temperature (Texp,in), condensator outlet temperature (Tcon,out), working medium pump discharge pressure (pp,out) and
Expanding machine output powerAbove-mentioned multigroup test data is grouped at random, is divided into acclimation group data and test group number
According to；
(2) normalized of gathered data
Since the order of magnitude difference of each parameter is larger, in order to accelerate the convergence rate of neural network, need to return each parameter
One change is handled, it is made to be converted into the respective value in [0,1] section
In formula, XminFor the minimum value of each parameter, XmaxFor the maximum value of each parameter, X is the gathered data of each parameter,For each ginseng
Numerical value after number normalized；
(3) foundation of BP neural network
The prediction model of the Organic Rankine Cycle output power based on BP neural network is established, which includes
One input layer, an intermediate hidden layer and an output layer；Wherein, input layer includes 7 neurons：The volume of organic working medium
Flow, expanding machine torque, expander inlet pressure, outlet pressure of expansion machine, expander inlet temperature, condensator outlet temperature,
Working medium pump discharge pressure；Output layer includes a neuron, the i.e. output power of Organic Rankine Cycle；Hidden layer transmission function is
Tansig, output layer transmission function are purelin；The neural network prediction model can pass through Calling MATLAB neural network work
Have case to realize；The neuron of input layer and output layer is all made of the data after step (2) normalized；
(4) training and test of neural network
Neural network maximum frequency of training and training objective error are set, by the training data after normalized in step (2)
It is input in the neural network that step (3) is established and is trained, after the completion of waiting for neural metwork training, test data is input to
Neural network is tested；Wherein, training objective error is evaluated using mean square error：
In formula, Q is input layer set, and Y (k) is neural network prediction value, and t (k) is real output value so that
Square error is less than training objective error；
(5) in test data neural network output data, that is, expanding machine output power anti-normalization processing
After the completion of being tested neural network according to step (4), neural network output data is subjected to anti-normalization processing：
In formula, YminFor the minimum value of expanding machine output power in all acclimation group data, YmaxFor all acclimation groups
The maximum value of expanding machine output power in data,Output valve is tested for neural network, Y is that the test after anti-normalization processing is defeated
Go out value.
2. a kind of Organic Rankine Cycle output power predicting method based on BP neural network described in accordance with the claim 1,
It is characterized in that, circulation step (1)-(5), the data of previous cycle is re-incorporated INTO step (1), step (2) will be passed through and normalized
Treated, and 7 operating parameters are input in the neural network that step (4) is established, and are obtained after corresponding output valve according still further to step
(5) anti-normalization processing is carried out, and then obtains the output power of subsequent cycle prediction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810361923.8A CN108564172A (en) | 2018-04-20 | 2018-04-20 | Organic Rankine Cycle output power predicting method based on BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810361923.8A CN108564172A (en) | 2018-04-20 | 2018-04-20 | Organic Rankine Cycle output power predicting method based on BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108564172A true CN108564172A (en) | 2018-09-21 |
Family
ID=63536168
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810361923.8A Pending CN108564172A (en) | 2018-04-20 | 2018-04-20 | Organic Rankine Cycle output power predicting method based on BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108564172A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109190327A (en) * | 2018-11-23 | 2019-01-11 | 华北电力大学（保定） | Organic Rankine Cycle system analysis optimization method, device and equipment |
| CN110263990A (en) * | 2019-06-10 | 2019-09-20 | 山东大学 | Vortex compounding machine flow torque prediction method and system neural network based |
| CN113255211A (en) * | 2021-05-14 | 2021-08-13 | 湘潭大学 | BP neural network and multi-objective optimization based organic Rankine cycle working medium screening method |
| CN114543256A (en) * | 2022-02-09 | 2022-05-27 | 青岛海尔空调电子有限公司 | Household charging method and device for multi-split air conditioner and multi-split air conditioner |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101487466A (en) * | 2009-02-25 | 2009-07-22 | 华东理工大学 | On-line soft measuring method for compression ratio and polytropic efficiency of centrifugal compressor |
| CN102135021A (en) * | 2011-02-25 | 2011-07-27 | 华东理工大学 | Method for predicting shaft power of industrial extraction condensing steam turbine |
| CN106980897A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of injector performance parameter prediction method of the BP artificial neural networks based on learning rate changing |
-
2018
- 2018-04-20 CN CN201810361923.8A patent/CN108564172A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101487466A (en) * | 2009-02-25 | 2009-07-22 | 华东理工大学 | On-line soft measuring method for compression ratio and polytropic efficiency of centrifugal compressor |
| CN102135021A (en) * | 2011-02-25 | 2011-07-27 | 华东理工大学 | Method for predicting shaft power of industrial extraction condensing steam turbine |
| CN106980897A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of injector performance parameter prediction method of the BP artificial neural networks based on learning rate changing |
Non-Patent Citations (2)
| Title |
|---|
| FUBIN YANG 等: ""Artificial neural network (ANN) based prediction and optimization of an organic Rankine cycle(ORC) for diesel engine waste heat recovery"", 《ENERGY CONVERSION AND MANAGEMENT 164 (2018)》 * |
| SHENGMING DONG 等: ""Investigation of Support Vector Machine and Back Propagation Artificial Neural Network for performance prediction of the organic Rankine cycle system"", 《ENERGY 144（2018）》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109190327A (en) * | 2018-11-23 | 2019-01-11 | 华北电力大学（保定） | Organic Rankine Cycle system analysis optimization method, device and equipment |
| CN109190327B (en) * | 2018-11-23 | 2022-11-22 | 华北电力大学（保定） | Method, device and equipment for analyzing and optimizing organic Rankine cycle system |
| CN110263990A (en) * | 2019-06-10 | 2019-09-20 | 山东大学 | Vortex compounding machine flow torque prediction method and system neural network based |
| CN113255211A (en) * | 2021-05-14 | 2021-08-13 | 湘潭大学 | BP neural network and multi-objective optimization based organic Rankine cycle working medium screening method |
| CN113255211B (en) * | 2021-05-14 | 2022-04-26 | 湘潭大学 | ORC working medium screening method based on BP neural network and multi-objective optimization |
| CN114543256A (en) * | 2022-02-09 | 2022-05-27 | 青岛海尔空调电子有限公司 | Household charging method and device for multi-split air conditioner and multi-split air conditioner |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108564172A (en) | Organic Rankine Cycle output power predicting method based on BP neural network | |
| Zhao et al. | Integrated simulation and control strategy of the diesel engine–organic Rankine cycle (ORC) combined system | |
| Tian et al. | Performance prediction of a cryogenic organic Rankine cycle based on back propagation neural network optimized by genetic algorithm | |
| CN109447236A (en) | A kind of method for diagnosing faults of hybrid vehicle heat management system | |
| CN102135021B (en) | Method for predicting shaft power of industrial extraction condensing steam turbine | |
| CN105116730B (en) | Hydrogen-fuel engine electronic spark advance angle and optimizing system and its optimization method based on Particle Group Fuzzy Neural Network | |
| Ping et al. | Nonlinear modeling and multi-scale influence characteristics analysis of organic Rankine cycle (ORC) system considering variable driving cycles | |
| CN110414089A (en) | The simulated prediction method of vehicle PEMS discharge based on Engine Universal Characteristics | |
| CN107729658A (en) | A kind of fuzzy intelligence composition decomposition extreme value response phase method of reliability Optimum Design | |
| CN109272174A (en) | Combustion turbine exhaustion system condition prediction technique based on Recognition with Recurrent Neural Network | |
| Okoji et al. | Evaluating the thermodynamic efficiency of the cement grate clinker cooler process using artificial neural networks and ANFIS | |
| Lin | Prediction of temperature distribution on piston crown surface of dual-fuel engines via a hybrid neural network | |
| Taghavifar et al. | Towards modeling of combined cooling, heating and power system with artificial neural network for exergy destruction and exergy efficiency prognostication of tri-generation components | |
| CN113466691A (en) | Prediction method for power generation efficiency of two-stage compression expansion generator | |
| CN110263990B (en) | Neural network-based flow torque prediction method and system for vortex type compound machine | |
| Sadatsakkak et al. | Implementation of artificial neural-networks to model the performance parameters of Stirling engine | |
| CN111931436A (en) | Burner nozzle air quantity prediction method based on numerical simulation and neural network | |
| CN111079920A (en) | Method for predicting uneven flow coefficient of outlet of turbine gas collecting cavity | |
| CN105373701A (en) | Electromechanical equipment association degree determination method | |
| Wang et al. | Research on anomaly detection and positioning of marine nuclear power steam turbine unit based on isolated forest | |
| CN100366876C (en) | Online analysis method and system for operation efficiency of combined gas-steam cycle power station | |
| Tao et al. | Development of integrative data intelligence models for thermo-economic performances prediction of hybrid organic rankine plants | |
| Zhang et al. | Operating conditions monitoring of vehicle internal combustion engine waste heat utilization systems based on support vector machines | |
| CN112670997A (en) | Electric heating energy source system time sequence probability load flow calculation method considering photovoltaic uncertainty | |
| Chen et al. | Organic Rankine Cycle Performance Analysis Based on WGCNA |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180921 |
|
| WD01 | Invention patent application deemed withdrawn after publication |