CN108985313A - AGC system discrimination method based on big data and Bayesian neural network - Google Patents
AGC system discrimination method based on big data and Bayesian neural network Download PDFInfo
- Publication number
- CN108985313A CN108985313A CN201810503450.0A CN201810503450A CN108985313A CN 108985313 A CN108985313 A CN 108985313A CN 201810503450 A CN201810503450 A CN 201810503450A CN 108985313 A CN108985313 A CN 108985313A
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- value
- output
- bayesian neural
- 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
- 238000013531 bayesian neural network Methods 0.000 title claims abstract description 36
- 238000012850 discrimination method Methods 0.000 title claims abstract description 16
- 238000000034 method Methods 0.000 claims abstract description 42
- 238000004364 calculation method Methods 0.000 claims abstract description 4
- 210000002569 neuron Anatomy 0.000 claims abstract description 4
- 238000013178 mathematical model Methods 0.000 claims abstract description 3
- 238000009499 grossing Methods 0.000 claims description 10
- 238000013459 approach Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 abstract description 12
- 238000012549 training Methods 0.000 description 12
- 238000012360 testing method Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 6
- 238000013473 artificial intelligence Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 239000000446 fuel Substances 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000000052 comparative effect Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000009826 distribution Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 230000001154 acute effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000000418 atomic force spectrum Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000013021 overheating Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000033764 rhythmic process Effects 0.000 description 1
- 230000005619 thermoelectricity Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Primary Health Care (AREA)
- Biomedical Technology (AREA)
- Public Health (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Probability & Statistics with Applications (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Water Supply & Treatment (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention relates to a kind of AGC system discrimination method based on big data and Bayesian neural network, comprising: 1, acquire the historical data of AGC system, pretreatment obtains sample data；2, Bayesian neural network is initialized；3, the input and output of hidden layer, each neuron of output layer are calculated, calculate the difference of reality output and Bayesian neural network output, and according to MSE criterion calculation error；4, whether error in judgement reaches requirement, if so, carry out step 6, otherwise, between output layer and hidden layer weight and threshold be modified, between input layer and hidden layer weight and threshold be modified, update each connection weight, study number adds 1；5, step 3~4 are repeated, until reaching error requirements or maximum study number；6, Bayesian neural network, the mathematical model recognized are calculated.Compared with classic BP neural network identification method, the present invention possesses better identification precision and faster convergence rate.
Description
Technical field
The present invention relates to information control technology field, more particularly, to a kind of based on big data and Bayesian neural network
AGC system discrimination method.
Background technique
End the end of the year in 2017, China's capacity of installed generator is 17.7 hundred million kilowatts, and thermoelectricity installed capacity accounts for total installation of generating capacity
62.2%, thermal power generation is still the main forms of electricity generation in China, however as economy enter new normality, face resource and ring
The double constraints in border, the situation that Thermal Power Generation Industry faces are more and more severeer.AGC is one of dispatching of power netwoks advanced technology means,
Its main task is to realize power network schedule automation Energy Management System (Energy Management System, abbreviation EMS)
With the closed-loop control between generator group coordination and control system (Coordination Control System, abbreviation CCS).It is extra-high
The fast development of voltage electric grid, AC and DC transmission technology and new energy more improves the complexity of dispatching of power netwoks and control, and
The fired power generating unit for having depth peak modulation capacity is therefore, to establish optimization AGC system model using energy consumption sacrifice as cost, promoted
AGC system Control platform is the most important thing of work at present.
In recent years, intelligence manufacture temperature is surging, and big data is the hot spot direction of each area research, target first is that according to
Data analysis result adjust automatically control strategy and way to manage keep power plant's production long-term to guarantee that fired power generating unit operates normally
In safety, economical and environmentally friendly in operating status.With the breakthrough development of deep neural network technology, artificial intelligence is in the world
The situation of persistently overheating is presented in range, 2017, " artificial intelligence " was put into the government work report in two Conferences for the first time, this is not
Great attention of the China to Artificial Intelligence Development is illustrated only, more indicates that " artificial intelligence " formally enters " Chinese rhythm ".Closely
Several years, big data caused the highest attention of all trades and professions, relied on the artificial intelligence of big data that can more play unlimited potentiality, can
To say that big data opens a new epoch.
Artificial intelligent recognition method is research emphasis in recent years, wherein such as neural network, genetic algorithm, particle group optimizing
Algorithm scheduling algorithm all again system be in achieve great research achievement, but there is also some shortcomings.It is directed to although having at present
AGC system carries out the research of the intelligent algorithms such as neural network modeling, but how sufficiently to combine power plant's big data and retrograde intelligence again
Can identification aspect still without providing practical guidance.Therefore by acquisition power plant's correlation operation data, power plant's big data background is studied
The lower feasibility that nerve root network modeling method is applied to AGC system identification, to the control performance for further increasing AGC system
Also significant.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide one kind based on big data with
The AGC system discrimination method of Bayesian neural network.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of AGC system discrimination method based on big data and Bayesian neural network, comprising the following steps:
S1, the historical data for acquiring AGC system, obtain sample data after pre-processing to historical data；
S2, Bayesian neural network is initialized；
S3, calculate hidden layer, output layer each neuron is output and input, calculate reality output and Bayesian neural network
The difference of output, and according to MSE criterion calculation error；
Whether S4, error in judgement reach requirement, if reaching requirement, carry out step S6, otherwise, to output layer and imply
Layer between weight and threshold be modified, between input layer and hidden layer weight and threshold be modified, update each company
Weight is connect, study number adds 1；
S5, step S3~S4 is repeated, until reaching error requirements or maximum study number；
S6, the final result for calculating Bayesian neural network hidden layer and output layer, the mathematical model recognized.
Preferably, the step S1 is specifically included:
S11, the historical data for acquiring AGC system, using the noise in smoothing method removal historical data；
S12, zero initialization and normalized are carried out to historical data；
S13, it selects to obtain sample data from historical data using nearest neighbour method.
Preferably, smoothing method uses LOESS smoothing method in the step S11, and this method specifically includes:
S111, with arbitrary number strong point x in historical dataiCentered on determine a section, interval width depend on participation office
The observed value number that portion returns；
S112, the weight for defining all the points in the section；
S113, the point in the section is fitted to conic section, obtains match value yi；
S114, the processing that step S111~S113 is carried out to point each in historical data, obtain one group of smooth point, xiIt is flat
Sliding point is exactly xiFitting the match point (x on the straight line comei,yi), all the points are connected with short straight line, obtain LOESS recurrence
Curve obtains LOESS sharpening result according to LOESS regression curve.
Preferably, the weight in the step S112 is by cube weighting function T (u) definition:
Wherein, u=Δi(x)/Δ(q)(x), Δi(x)=| xi- x | indicate the distance of point x to xi, xiIndicate historical data
Arbitrary number strong point, x indicate historical data in remove xiExcept other data points, Δ(q)(x) Δ is indicatedi(x) maximum value.
Preferably, include: according to the process that LOESS regression curve obtains LOESS sharpening result in the step S114
The quadratic sum for defining the difference that error function is match value and actual value, returns LOESS using gradient descent method
Curve approaches y=θixi, y expression inputs or output data smoothed curve, θiIndicate xiThe slope for locating smoothed curve, to obtain
LOESS sharpening result.
Preferably, zero initialization process specifically includes in the step S12:
Wherein, u* (k), y* (k), u (k), y (k) respectively indicate the value after input data zero initializes, at the beginning of output data zero
The value before value, the initialization of input data zero, the value after the initialization of output data zero after beginningization, N is constant, indicates access evidence
Initial value of the mean value of top N as data.
Compared with prior art, the present invention uses bayesian algorithm, converts statistical concept for uncertainty, will weigh
Value optimization problem is converted into solution normal law error minimum problems and improves algorithm to avoid the case where falling into local optimum
Rapidity and accuracy, compared with classic BP neural network identification method, possess better identification precision and faster receive
Speed is held back, is had important practical significance to the automatic control level for improving fired power generating unit.
Detailed description of the invention
Fig. 1 is that the present invention is based on the AGC system discrimination method flow diagrams of big data and Bayesian neural network；
Fig. 2 is certain the 1000MW extra-supercritical unit AGC system lofty tone gate valve position input curve figure acquired in embodiment；
Fig. 3 is certain the 1000MW extra-supercritical unit AGC system feedwater flow input curve figure acquired in embodiment；
Fig. 4 is certain the total fuel quantity input curve figure of 1000MW extra-supercritical unit AGC system acquired in embodiment；
Fig. 5 is certain the 1000MW extra-supercritical unit AGC system actual power output curve diagram acquired in embodiment；
Fig. 6 is certain the 1000MW extra-supercritical unit AGC system centrum's temperature output curve diagram acquired in embodiment；
Fig. 7 is certain the 1000MW extra-supercritical unit AGC system main steam pressure output curve diagram acquired in embodiment；
Fig. 8 is the lofty tone gate valve position sample data curve graph obtained in embodiment using the method for the present invention；
Fig. 9 is the feedwater flow sample data curve graph obtained in embodiment using the method for the present invention；
Figure 10 is the total fuel quantity sample data curve graph obtained in embodiment using the method for the present invention；
Figure 11 is the actual power sample data curve graph obtained in embodiment using the method for the present invention；
Figure 12 is the centrum's temperature sample data curve graph obtained in embodiment using the method for the present invention；
Figure 13 is the main steam pressure sample data curve graph obtained in embodiment using the method for the present invention；
Figure 14 is the training of actual power when being recognized using the method for the present invention and classical BP algorithm to AGC system in embodiment
Comparative result figure；
Figure 15 is the instruction of centrum's temperature when being recognized using the method for the present invention and classical BP algorithm to AGC system in embodiment
Practice comparative result figure；
Figure 16 is the instruction of main steam pressure when being recognized using the method for the present invention and classical BP algorithm to AGC system in embodiment
Practice comparative result figure；
Figure 17 is practical function obtained by the method for the present invention and classical BP algorithm when being verified in embodiment with remaining data
Rate test result comparison diagram；
Figure 18 is intermediate point obtained by the method for the present invention and classical BP algorithm when being verified in embodiment with remaining data
Temperature test result comparison diagram；
Figure 19 is main steam obtained by the method for the present invention and classical BP algorithm when being verified in embodiment with remaining data
Pressure testing results comparison diagram.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention
Premised on implemented, the detailed implementation method and specific operation process are given, but protection scope of the present invention is not limited to
Following embodiments.
A kind of AGC system discrimination method based on big data and Bayesian neural network, as shown in Figure 1, passing through acquisition AGC
The history big data of system selects lofty tone gate valve position, feedwater flow and total fuel quantity as mode input amount, actual power, in
Between put temperature and main steam pressure as model output, first using data smoothing, removal trend term, normalize and etc. into
Then line number Data preprocess is selected the sample data set that can characterize system performance using nearest neighbour method, utilizes Bayesian neural network
Model is trained the relationship between rear mining data, and establishes the AGC model based on Bayesian neural network, specifically include with
Lower step:
S1, the historical data for acquiring AGC system, obtain sample data after pre-processing to historical data, specifically include:
S11, the historical data for acquiring AGC system, using the noise in smoothing method removal historical data, the present embodiment
In, LOESS smoothing method is specifically used, smoothing method does not depend on the overall distribution and its parameter of data, and application is more flexible, tool
Body includes:
S111, with arbitrary number strong point x in historical dataiCentered on determine a section, width depend on q=f × n,
Wherein, q is the observed value number for participating in local regression, and f is that the number for the observed value for participating in local regression accounts for observed value number
Ratio, n is observed value number, in practical application, first selecting f value, the value of q is determined further according to f, n, increases f, q, smooth journey
Degree increases, otherwise reduces；
The weight of all the points in S112, interval of definition is defined by cube weighting function T (u):
Wherein, u=Δi(x)/Δ(q)(x), Δ is enabledi(x)=| xi- x | indicate point x to xiDistance, xiIndicate historical data
Arbitrary number strong point, x indicate historical data in remove xiExcept other data points, by Δi(x) it is arranged as Δ from small to large(i)
(x), Δ(q)(x) Δ is indicatedi(x) maximum value；
S113, the point in section is fitted to conic section, influence of the bigger point of weight to result is bigger, outside section
Point weight is 0, obtains match value yi；
S114, the processing that step S111~S113 is carried out to point each in historical data, obtain one group of smooth point, xiIt is flat
Sliding point is exactly xiFitting the match point (x on the straight line comei,yi), all the points are connected with short straight line, obtain LOESS recurrence
Curve defines the quadratic sum for the difference that error function is match value and actual value, makes LOESS regression curve using gradient descent method
Approach y=θixi, y expression inputs or output data smoothed curve, θiIndicate xiThe slope for locating smoothed curve, to obtain LOESS
Sharpening result.
S12, zero initialization and normalized, zero initialization process formula are carried out to historical data are as follows:
Wherein, u*(k)、y*(k), u (k), y (k) respectively indicate the value after input data zero initializes, at the beginning of output data zero
The value before value, the initialization of input data zero, the value after the initialization of output data zero after beginningization, N is constant, indicates access evidence
Initial value of the mean value of top N as data；
To the formula of data normalized are as follows:
Wherein, xminMinimum value before indicating to normalize in data, xmaxMaximum value before indicating to normalize in data,
xgiI-th of data after indicating normalization, xsIndicate the individual data point to be normalized；
S13, it selects to obtain sample data from historical data using nearest neighbour method, comprising:
S131, in data according to same intervals select 200 data groups be stored as initial training sample set X, due to number
It is big according to amount, acquisition time is long, it is believed that be obvious different between data, be not neighbour；
S132, when inputting new test sample xniWhen, calculate the Europe between the test sample and each initial training sample
Formula distance Ed:
In formula, XiIndicate arbitrary initial training sample value；
S133, judge test sample xniWith XiEuclidean distance whether be smaller value α, α is obtained by many experiments
The constant value that identification effect can be made best.If so, illustrating X in sample setiExisting xniNeighbour's value, then xniIt does not need to be added
Training sample set Xi；Conversely, then illustrating that the sample can embody different system features, need xniIt is selected into XiTo improve sample
Collection；
S134 repeats S132~S133, obtains final sample collection after traversing total data.
S2, Bayesian neural network is initialized, gives each connection weight at random, set error function, given and calculate essence
Degree, learning rate, maximum study number, and initialize study number.
S3, calculate hidden layer, output layer each neuron is output and input, calculate reality output and Bayesian neural network
The difference of output, and according to MSE criterion calculation error E.
Whether S4, error in judgement reach requirement, if reaching requirement, carry out step S6, otherwise, to output layer and imply
Layer between weight and threshold be modified, between input layer and hidden layer weight and threshold be modified, update each company
Weight is connect, study number adds 1.
S5, step S3~S4 is repeated, until reaching error requirements or maximum study number.
S6, the final result for calculating Bayesian neural network hidden layer and output layer save network, the mathematics recognized
Model.
After training obtains the Bayesian neural network of AGC system through the above steps, with remaining sampled data to being established
Network is tested.
Classical BP neural network setting in initial weight, initial learning rate value, the hidden layer number of plies, hidden layer unit number etc.
It sets and is all instructed without accurate theoretical foundation, exist and easily fall into the drawbacks such as local optimum, pace of learning be slow.The application uses pattra leaves
This algorithm improvement neural network carries out AGC system identification.The final purpose of neural metwork training is to find suitable weight vector ω,
So that root-mean-square error E obtains minimum value.Bayesian algorithm is to convert uncertainty to statistical concept, by weight optimizing
Problem is converted into solution normal law error minimum problems and improves the quick of algorithm to avoid the case where falling into local optimum
Property and accuracy.According to bayesian principle, probabilistic model is converted by traditional problem analysis, weight matrix ω will be asked to make posteriority
Distribution probability obtains max problem and is converted into normal law error functional minimum value problem of asking, that is, utilizes bayesian theory by weight
Optimization problem converts to solve minimum problems.
Embodiment
To verify the AGC system discrimination method identification precision with higher and faster convergence speed that the application is proposed
Degree, acquire certain 1000MW extra-supercritical unit history data, select lofty tone gate valve position, feedwater flow and total fuel quantity as
Mode input amount, actual power, centrum's temperature and main steam pressure acquire 16381 groups of data, adopt as model output
0.01s is divided between sample, then the input observed and curve of output are respectively as shown in Fig. 2~7.
The pretreatment such as zero initialization, data de-noising, normalization is carried out to all data according to the application mentioned discrimination method
Afterwards, 1000 groups of data are chosen as sample data using nearest neighbour method, the sample data curve of selection is as depicted in figures 8-13.By sample
Sampled data except notebook data is as test data, respectively using classical BP algorithm and the improved nerve net of the application Bayes
Network algorithm is trained and tests, and network parameter is arranged by 1 data of table when training:
1 neural network algorithm parameter setting of table
| Hidden nodes | Maximum frequency of training | Learning rate | Minimum error values |
| 30 | 3000 | 0.00006 | 0.65*10^-35 |
After 3000 study, BP neural network algorithm and Bayesian neural network algorithm have been restrained, training result
With sample curve of output respectively as shown in Figure 14, Figure 15, Figure 16.It will be evident that classical BP neural network is changing from figure
Trend can train preferable as a result, still not in dramatic decrease or ascent stage and sample data acute variation when smaller
The identification effect of Bayesian neural network can be reached, although the worst error value of Bayesian neural network is larger, from entirety
From the point of view of still better than classical BP neural network.After model training is good, the result that remainder data is verified is respectively such as Figure 17, figure
18, shown in Figure 19.
From Picture study it is found that two methods test effect have it is significantly different.Actual power and main steam pressure force curve
Fluctuation compared with centrum's temperature curve is significantly smaller, thus in the prediction of the two output valves two methods show compared with
Good, although wherein the identification trend of classic BP is in the main true, but accuracy is not as good as Bayesian neural network.And in intermediate point temperature
It spends in the prediction of output valve, it is exactly the opposite, it is particularly evident in fluctuation larger part.
The learning error of Bayesian neural network drops to 153 from 2780 rapidly in preceding 1000 generation error, and effect is obvious, with
It is slowly varying up to training terminates near optimal value afterwards.Final error is 7.4576.Under identical network structure, from instruction
Practice result apparently, Bayesian neural network clearly compared with classical BP neural network advantage, test phase error still
It is smaller, and Bayesian neural network convergence rate is very fast.
Claims (6)
1. a kind of AGC system discrimination method based on big data and Bayesian neural network, which is characterized in that including following step
It is rapid:
S1, the historical data for acquiring AGC system, obtain sample data after pre-processing to historical data；
S2, Bayesian neural network is initialized；
S3, calculate hidden layer, output layer each neuron is output and input, calculate reality output and Bayesian neural network export
Difference, and according to MSE criterion calculation error；
Whether S4, error in judgement reach requirement, if reaching requirement, carry out step S6, otherwise, to output layer and hidden layer it
Between weight and threshold be modified, between input layer and hidden layer weight and threshold be modified, update each connection weight
Value, study number add 1；
S5, step S3~S4 is repeated, until reaching error requirements or maximum study number；
S6, the final result for calculating Bayesian neural network hidden layer and output layer, the mathematical model recognized.
2. the AGC system discrimination method according to claim 1 based on big data and Bayesian neural network, feature exist
In the step S1 is specifically included:
S11, the historical data for acquiring AGC system, using the noise in smoothing method removal historical data；
S12, zero initialization and normalized are carried out to historical data；
S13, it selects to obtain sample data from historical data using nearest neighbour method.
3. the AGC system discrimination method according to claim 2 based on big data and Bayesian neural network, feature exist
In smoothing method uses LOESS smoothing method in the step S11, and this method specifically includes:
S111, with arbitrary number strong point x in historical dataiCentered on determine a section, interval width depend on participate in local regression
Observed value number；
S112, the weight for defining all the points in the section；
S113, the point in the section is fitted to conic section, obtains match value yi；
S114, the processing that step S111~S113 is carried out to point each in historical data, obtain one group of smooth point, xiSmooth point
It is exactly xiFitting the match point (x on the straight line comei,yi), all the points are connected with short straight line, LOESS is obtained and returns song
Line obtains LOESS sharpening result according to LOESS regression curve.
4. the AGC system discrimination method according to claim 3 based on big data and Bayesian neural network, feature exist
In the weight in the step S112 is by cube weighting function T (u) definition:
Wherein, u=Δi(x)/Δ(q)(x), Δi(x)=| xi- x | indicate the distance of point x to xi, xi indicates appointing for historical data
Meaning data point, x indicate to remove x in historical dataiExcept other data points, Δ(q)(x) Δ is indicatedi(x) maximum value.
5. the AGC system discrimination method according to claim 3 based on big data and Bayesian neural network, feature exist
In including: according to the process that LOESS regression curve obtains LOESS sharpening result in the step S114
The quadratic sum for defining the difference that error function is match value and actual value, makes LOESS regression curve using gradient descent method
Approach y=θixi, y expression inputs or output data smoothed curve, θiIndicate xiThe slope for locating smoothed curve, to obtain LOESS
Sharpening result.
6. the AGC system discrimination method according to claim 2 based on big data and Bayesian neural network, feature exist
In zero initialization process specifically includes in the step S12:
Wherein, u*(k)、y*(k), u (k), y (k) respectively indicate the value after input data zero initializes, output data zero initializes
Rear value, input data zero initialize before value, output data zero initialize after value, N is constant, indicates the preceding N of access evidence
Initial value of the mean value of position as data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810503450.0A CN108985313A (en) | 2018-05-23 | 2018-05-23 | AGC system discrimination method based on big data and Bayesian neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810503450.0A CN108985313A (en) | 2018-05-23 | 2018-05-23 | AGC system discrimination method based on big data and Bayesian neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108985313A true CN108985313A (en) | 2018-12-11 |
Family
ID=64542611
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810503450.0A Pending CN108985313A (en) | 2018-05-23 | 2018-05-23 | AGC system discrimination method based on big data and Bayesian neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108985313A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111415010A (en) * | 2020-03-20 | 2020-07-14 | 广东电网有限责任公司阳江供电局 | Bayesian neural network-based wind turbine generator parameter identification method |
| CN111555363A (en) * | 2020-04-09 | 2020-08-18 | 广西大学 | AGC real-time control strategy based on deep learning under big data environment |
| WO2020173270A1 (en) * | 2019-02-25 | 2020-09-03 | 日本电气株式会社 | Method and device used for parsing data and computer storage medium |
| CN112034340A (en) * | 2019-06-03 | 2020-12-04 | 中国人民解放军63756部队 | Method for screening fault characteristics of measurement and control antenna motor |
| CN115700494A (en) * | 2022-09-16 | 2023-02-07 | 哈尔滨工业大学 | Rail transit monitoring data cleaning method and system based on Bayesian inference |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107358060A (en) * | 2017-09-06 | 2017-11-17 | 大连理工大学 | A kind of method estimated wind power prediction error burst based on HMM |
| CN107742000A (en) * | 2017-08-31 | 2018-02-27 | 国网江西省电力公司电力科学研究院 | Boiler combustion oxygen content modeling method |
-
2018
- 2018-05-23 CN CN201810503450.0A patent/CN108985313A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107742000A (en) * | 2017-08-31 | 2018-02-27 | 国网江西省电力公司电力科学研究院 | Boiler combustion oxygen content modeling method |
| CN107358060A (en) * | 2017-09-06 | 2017-11-17 | 大连理工大学 | A kind of method estimated wind power prediction error burst based on HMM |
Non-Patent Citations (3)
| Title |
|---|
| WILLIAM S. CLEVELAND: "Robust Locally Weighted Regression and Smoothing Scatterplots", 《JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION》 * |
| 杰夫•H•吉文斯 等著: "《计算统计(第2版)》", 31 December 2017 * |
| 马湧: "钢铁企业蒸汽压力调控优化***研究", 《中国博士学位论文全文数据库 工程科技Ⅰ辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020173270A1 (en) * | 2019-02-25 | 2020-09-03 | 日本电气株式会社 | Method and device used for parsing data and computer storage medium |
| CN112034340A (en) * | 2019-06-03 | 2020-12-04 | 中国人民解放军63756部队 | Method for screening fault characteristics of measurement and control antenna motor |
| CN111415010A (en) * | 2020-03-20 | 2020-07-14 | 广东电网有限责任公司阳江供电局 | Bayesian neural network-based wind turbine generator parameter identification method |
| CN111415010B (en) * | 2020-03-20 | 2022-11-22 | 广东电网有限责任公司阳江供电局 | Bayesian neural network-based wind turbine generator parameter identification method |
| CN111555363A (en) * | 2020-04-09 | 2020-08-18 | 广西大学 | AGC real-time control strategy based on deep learning under big data environment |
| CN115700494A (en) * | 2022-09-16 | 2023-02-07 | 哈尔滨工业大学 | Rail transit monitoring data cleaning method and system based on Bayesian inference |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108985313A (en) | AGC system discrimination method based on big data and Bayesian neural network | |
| CN106529166B (en) | A kind of System in Optimal Allocation of Regional Water Resources method based on MAEPSO algorithm | |
| CN108616120B (en) | Non-invasive power load decomposition method based on RBF neural network | |
| CN107482692B (en) | Active control method, device and system for wind power plant | |
| CN111008504B (en) | Wind power prediction error modeling method based on meteorological pattern recognition | |
| CN109256810A (en) | Consider that blower is contributed and does not know the Multipurpose Optimal Method of cost | |
| CN109636009B (en) | Method and system for establishing neural network model for determining line loss of power grid | |
| CN106529719A (en) | Method of predicting wind power of wind speed fusion based on particle swarm optimization algorithm | |
| CN110837915B (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
| CN106528912A (en) | Method for estimating frequency regulation capacity of wind power plant | |
| CN112990627B (en) | Power quality evaluation method | |
| CN109636054A (en) | Solar energy power generating amount prediction technique based on classification and error combination prediction | |
| CN113555881A (en) | Wind power plant frequency modulation fan sequencing method | |
| CN103440535A (en) | Multi-target plant level load optimization method based on immune optimization and fuzzy decision | |
| CN115173465A (en) | Wind, light, water, fire and storage integrated coupling mechanism analysis method based on Copula theory | |
| CN110197296A (en) | A kind of unit load prediction technique based on Time Series Similarity | |
| CN115641231A (en) | Typical scene extraction method and system for power system, electronic equipment and medium | |
| CN109840335A (en) | Based on the radial forging pit prediction optimization method for strengthening T-S fuzzy neural network | |
| CN113361454A (en) | Deep learning non-invasive load monitoring method based on unsupervised optimization | |
| CN113705098A (en) | Air duct heater modeling method based on PCA and GA-BP network | |
| CN108321801A (en) | Method and system for making day-ahead power generation plan of energy base system | |
| CN112001639A (en) | Adjustable capacity evaluation method for energy demand of comprehensive energy system and storage medium | |
| CN115630311A (en) | Wind, photovoltaic and hydrogen scene reduction method considering correlation of uncertain factors | |
| Mishra et al. | Ramping behaviour analysis of wind farms | |
| CN115102190B (en) | Parameter optimization method for in-station/station network oscillation suppression of grid-connected system of photovoltaic power station |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181211 |