CN104239489B - Utilize the method for similarity searching and improved BP forecast level - Google Patents
Utilize the method for similarity searching and improved BP forecast level Download PDFInfo
- Publication number
- CN104239489B CN104239489B CN201410454011.7A CN201410454011A CN104239489B CN 104239489 B CN104239489 B CN 104239489B CN 201410454011 A CN201410454011 A CN 201410454011A CN 104239489 B CN104239489 B CN 104239489B
- Authority
- CN
- China
- Prior art keywords
- water level
- sequence
- month
- value
- data
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3322—Query formulation using system suggestions
- G06F16/3323—Query formulation using system suggestions using document space presentation or visualization, e.g. category, hierarchy or range presentation and selection
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Abstract
The present invention discloses a kind of method using similarity searching and improved BP neural network forecast level, the water level in similar hydrological characteristics month is possessed for 10 years to first five according to the water level of 15 days a few days ago to be predicted and carries out similarity measurement, find out itself and most similar water level period in each year, then using the water level of this most similar water level period of 50 years and latter day as training set, it is predicted using the BP neural network based on genetic algorithm.This method includes data prediction, it is intended to makes up data missing errors etc.;Similarity searching, using dynamic bending distance and sliding window technique, find out the minimum i.e. most like sequence of distance of the water level on the 15th month water level similar to first five 10 years;BP neural network based on genetic algorithm, establish system hierarchy with genetic algorithm and carry out global optimizing, and be predicted using the learning training ability of BP neural network.The present invention can look-ahead water level, provide effective technical support for flood control and disaster relief.
Description
Technical field
The present invention relates to a kind of using similarity searching and based on the improved BP neural network forecast level of genetic algorithm
Technology, more particularly to the similarity searching to water level information and the BP neural network Predicting Technique based on genetic algorithm, belong to
Areas of information technology.
Background technology
It is being the accumulation of feature, as time in time sequencing that time series, which is property value, with moving ahead for epoch, water
Literary data are also in accumulation slowly, and these hydrographic datas possess the features such as a large amount of, species is more, dimension is high, updating decision, how to this
A little data carry out strong analysis, and therefrom obtaining useful information turns into focus of concern.With the development of science and technology
And the accumulation of hydrographic data, people give flood control and disaster relief and more paid close attention to.If for the water of one day or more days in a basin
Potential energy is enough effectively predicted that this will be flood forecasting, and Flood Control Dispatch provides strong technical support.
The method that presently, there are many hydrologic(al) prognosis, but they have some defects.Using it is most wide be hydrology field
Model prediction, but these models typically can be used only in specific basin, and they have unique corresponding relation between each other, i.e., suitable
Answering property is weak, and more focuses on the application of hydrology knowledge, the good popularization that can not be obtained;The limitation of hydrology professional knowledge is excluded,
Be easier to receive is the method for computer realm, and this more emphasizes the analysis to data, a large amount of to former years using various methods
Data analysis reaches the purpose of prediction, such as the prediction based on neutral net, and it is likely to be converging on local minimum, can not be right
Convergence rate is controlled;The prediction of SVMs, large-scale training sample can not be implemented.
The content of the invention
Goal of the invention:For problems of the prior art, to improve the precision and adaptability of water level forecast, this hair
A kind of method using similarity searching and based on the improved BP neural network forecast level of genetic algorithm of bright offer.
Technical scheme:A kind of side using similarity searching and based on the improved BP neural network forecast level of genetic algorithm
Method, including:
A) data prediction:Preprocessing process includes:Data selection (data selection), data cleansing (data
Cleaning), data conversion (data transformation).It is to be predicted a few days ago ten to determine to need data to be processed first
Water level on the five and the water level value in first five 10 years identical month, here it is data selection;For being lacked present in Hydrological Time Series
Noise of becoming estranged needs to be cleaned, and is allowed to not influence the correctness of result;Hydrological Time Series are the high dimensional datas of magnanimity, wherein
The noise spot (short-term fluctuation) contained can influence similitude differentiation, so needing to be smoothed time series data i.e. data
Conversion;
B) similarity measurement:According to dynamic time warping distance, using the water level of 15 days a few days ago to be predicted in 50 years
With searching the most like water level period in month water level, using this 50 groups of similar water levels and corresponding latter day water level as instructing
Practice collection;
The same month water level:Search for the determination in month, according to the water regime of basin water level, annual each season or
The water level information in person's each month has certain characteristic, such as:The Hydrological characteristics and seasonal variations rule of Taihu Lake basin are 5
Month, water table ratio was shallower, and June has then risen slightly.Therefore basin must be taken into full account when it is determined that searching for month
Hydrological characteristics and seasonal variations rule, being included in month of having close Hydrological characteristics of the water level characteristic with 15 days to be searched is searched
Rope scope.
After determining search month, according to dynamic time warping (Dynamic Time Warping, DTW) method, utilize
Sliding window technique, the annual and most like water of water level on the 15th to be searched is found out in the annual search month of 50 years
Position, and originate the date of expiry and be marked, while the water level of similar water level latter day every year is taken out, it can thus obtain five
The training set of ten groups of similar water levels+water level of latter day (equivalent to the water level of day to be predicted).
C) genetic algorithm:Before training, first pass through the genetic operations such as selection cross and variation of the genetic algorithm to chromosome and look for
The initial weight optimal to BP neural network.Also, when BP neural network is absorbed in minimum, genetic algorithm optimization is transferred to again
Network parameter.Best initial weights and threshold value are obtained after required precision is met;
It is described to obtain optimal initial weight:Refer to produce initial population and carry out Population Coding, for participating in the defeated of training
Enter the factor, because scale of neural network is larger, so with real coding, i.e., the base by a real number directly as a chromosome
Because of position.The weight of network is generated according to the conventional method of neutral net, the part of coding has:The company of input layer and hidden layer
Connect the connection weight, hidden layer threshold value, output layer threshold value of weights, hidden layer and output layer.These are joined together to form one
Individual long string, above each position represent the weights and threshold value of network, this just constitutes an individual, produce it is multiple this
The individual of sample just constitutes initial population;
It is described to obtain final best initial weights threshold value:The optimal initial weight threshold of the BP neural network obtained according to individual
Value, train BP neural network to obtain the i.e. desired prediction of system output with training data and export, ideal adaptation angle value and reality are defeated
The Error Absolute Value gone out between desired output is relevant.The low individual of fitness value is subjected to selection cross and variation, meets optimization
Principle is best initial weights threshold value.
D) train to obtain predicted value using BP neural network:The optimal initial weight and satisfaction that will be obtained by genetic algorithm
The best initial weights threshold value of condition substitutes into BP neural network and is trained, and obtains the mistake between the reality output of sample and desired output
Difference, weight matrix is adjusted according still further to normal training philosophy, is trained again, until obtaining final weights, last basis
The water level of 15 days a few days ago to be predicted obtains predicted value.
Beneficial effect:Compared with prior art, it is provided by the present invention improved using similarity searching and genetic algorithm
The method of BP neural network forecast level has very strong adaptability, is not limited by basin, can be obtained using similarity searching
Maximally effective training set, and genetic algorithm improved BP is utilized, avoid it from being absorbed in local minimum, can carry significantly
High training effect.
Brief description of the drawings
Fig. 1 is the method flow diagram of the embodiment of the present invention;
Fig. 2 is the data prediction flow chart of the embodiment of the present invention;
Fig. 3 is the similarity searching execution flow chart of the embodiment of the present invention;
Fig. 4 is the improved BP neural network execution flow chart of genetic algorithm of the embodiment of the present invention;
Fig. 5 is the genetic algorithm execution flow chart of the embodiment of the present invention;
Fig. 6 is the BP neural network execution flow chart of the embodiment of the present invention.
Embodiment
With reference to specific embodiment, the present invention is furture elucidated, it should be understood that these embodiments are merely to illustrate the present invention
Rather than limitation the scope of the present invention, after the present invention has been read, various equivalences of the those skilled in the art to the present invention
The modification of form falls within the application appended claims limited range.
As shown in Figure 1:
A) data prediction:Including data selection, data cleansing, data conversion.
It is water level of the 15 days a few days ago water levels to be predicted with first five 10 years identical month that determining first, which needs data to be processed,
Value, here it is data selection;Need to clean for missing present in Hydrological Time Series and noise, be allowed to not influence to tie
The correctness of fruit;Hydrological Time Series are the high dimensional data of magnanimity, wherein the noise spot (short-term fluctuation) contained can influence phase
Differentiate like property, so needing to be smoothed time series data i.e. data conversion;
B) similarity measurement:According to dynamic time warping distance, using the water level of 15 days a few days ago to be predicted in 50 years
With searching the most like water level period in month water level, using this 50 groups of similar water levels and corresponding latter day water level as instructing
Practice collection;
The same month water level:Search for the determination in month, according to the water regime of basin water level, annual each season or
The water level information in person's each month has certain characteristic, such as:The Hydrological characteristics and seasonal variations rule of Taihu Lake basin are 5
Month, water table ratio was shallower, and June has then risen slightly.Therefore basin must be taken into full account when it is determined that searching for month
Hydrological characteristics and seasonal variations rule, being included in month of having close Hydrological characteristics of the water level characteristic with 15 days to be searched is searched
Rope scope.
It is annual in 50 years using sliding window technique according to dynamic time warping method after determining search month
Search month in find out annual with the most like water level of water level on the 15th to be searched, and originated the date of expiry and marked
Come, while take out the water level of annual similar water level latter day, can thus obtain (50 groups of similar water levels+latter day water level (phase
When in the water level of day to be predicted)) training set.
C) genetic algorithm:Before training, first pass through the genetic operations such as selection cross and variation of the genetic algorithm to chromosome and look for
The initial weight optimal to BP neural network.Also, when BP neural network is absorbed in minimum, genetic algorithm optimization is transferred to again
Network parameter.Best initial weights and threshold value are obtained after certain precision is met;
It is described to obtain optimal initial weight:Refer to produce initial population and carry out Population Coding, for participating in the defeated of training
Enter the factor, because scale of neural network is larger, so with real coding, i.e., the base by a real number directly as a chromosome
Because of position.The weight of network is generated according to the conventional method of neutral net, the part of coding has:The company of input layer and hidden layer
Connect the connection weight, hidden layer threshold value, output layer threshold value of weights, hidden layer and output layer.These are joined together to form one
Individual long string, above each position represent the weights and threshold value of network, this just constitutes an individual, produce it is multiple this
The individual of sample just constitutes initial population;
It is described to obtain final best initial weights threshold value:The optimal initial weight threshold of the BP neural network obtained according to individual
Value, train BP neural network to obtain the i.e. desired prediction of system output with training data and export, ideal adaptation angle value and reality are defeated
The Error Absolute Value gone out between desired output is relevant.The low individual of fitness value is subjected to selection cross and variation, meets optimization
Principle is best initial weights threshold value.
D) BP neural network:By the optimal initial weight obtained by genetic algorithm and the best initial weights threshold value for meeting condition
Substitute into BP neural network to be trained, obtain the error between the reality output of sample and desired output, according still further to normal instruction
Practice principle adjustment weight matrix, be trained again, until obtaining final weights, finally according to 15 days a few days ago to be predicted
Water level obtains predicted value.
If Fig. 2 is data prediction flow chart:
In real life, due to reasons such as scope failure, system hysteresis, one or continuous is caused in hydrological data bank
The shortage of data at multiple time points.Such low quality data can influence the differentiation result of similitude to a certain extent, and then
Influence the accuracy of prediction result, it is therefore necessary to which hydrological observation data are pre-processed.Using interpolation, data filling, sequence
It is smooth to wait transform method to pre-process time series, prepared for similarity searching model.Preprocessing process includes:Data
Selection, data cleansing, data conversion.
It is certain basin 15 days a few days ago water levels to be predicted and first five 10 years identical month that determining first, which needs data to be processed,
Water level value, here it is data selection;
Cleaned for lacking needs present in Hydrological Time Series, be allowed to not influence the correctness of result, for
Shortage of data in time series, its Crack cause are often data acquisition equipment failures in data-gathering process, and network transmission lacks
Lose or manually omit.Ignore these data or simply carry out completion using data, it will usually the result of similarity query is influenceed,
Cause insecure matching result, and then influence the result of prediction.Put down using totality of the time related sequence on the period
Average lacks to fill up;Such as the information of period is corresponded to reference to the multiple survey stations closed on;
Hydrological Time Series are the high dimensional data of magnanimity, wherein the noise spot (short-term fluctuation) contained can influence similitude
Differentiate, so needing to be smoothed time series data i.e. data conversion.More chis are carried out using wavelet transform (DWT)
The conversion of degree, low frequency signal is obtained untill meeting required smoothness.
If Fig. 3 is similarity searching execution flow chart.Comprise the following steps:
Step 101, it is the water level time series of 15 days a few days ago to be predicted by sequence to be matched after determining search month
And sequence to be searched is that the water level time series for searching for month is standardized.Using zero-Mean Method, for original series
C, be standardized as sequence C ', wherein u and v are respectively the average value and standard deviation of the sequence:
Step 102, the length of sliding window is determined, 1/2 or 2 times of sequence length generally to be matched, i.e. [7,21],
And matching length is initialized as 7;
Step 103, judge matching length whether between the minimum value and maximum of sliding window length i.e. [7,
21], if so, then carrying out in next step, otherwise carrying out step 106;
Step 104, judge whether to terminate a search sequence matching with the sliding window of this length, if not then carrying out
In next step, the length of sliding window is otherwise added into a carry out step 103;
Step 105, the sliding window for calculating certain length is often moved back between an obtained sequence and sequence to be matched
DTW distances, and mark corresponding sequence originate final position.Computational length is respectively the DTW between m and n time serieses X and Y
It is as follows apart from process:
(1) value m corresponding to coordinate (i, j) in the matrix M, M between X and Y is constructedijFor xiWith yjBetween Euclidean distance d
(xi,yi);
(2) accumulative matrix R is constructed, is worth calculation formula as follows corresponding to coordinate (i, j):
(3) the minimum accumulated value r of final time series crooked routem,nBe exactly DTW between time series X and Y away from
From.
Calculate certain length sliding window travel through DTW distances all after one search sequence and has recorded after, repeatedly
Step 104;
Step 106, it is the water level time series in the search month of 1 year in 50 years for a search sequence, calculates
The sequence to be matched DTW distance all with it, compare and draw beeline and the i.e. corresponding starting of corresponding starting final position
Date of expiry, the most like sequence of 1 year in 50 years is thus obtained, using this sequence and its latter day water level as one
Individual training sample, other times are by that analogy, then have just obtained all training sets.
As shown in figure 4, the improved BP neural network execution flow chart of genetic algorithm for the present embodiment:
Every chromosome in genetic algorithm initial population is all made up of the initial weight threshold value in BP neural network, is gone forward side by side
Row Population Coding.The low individual of fitness value is subjected to selection cross and variation, meet optimization principles is used as best initial weights threshold value,
It is input to BP neural network to be trained, constantly uses the study of BP algorithm, until obtaining final weights, that is, determine most suitable
The neutral net of conjunction, predicted value was finally obtained according to the water level of 15 days a few days ago to be predicted.
As shown in figure 5, the genetic algorithm execution flow chart for the present embodiment.Comprise the following steps:
Step 201, produce initial population and carry out Population Coding, the input factor for participating in training, compiled using real number
Code, i.e., the gene position by a real number directly as a chromosome.The power of network is generated according to the conventional method of neutral net
Weight, the part of coding have:The connection weight of input layer and hidden layer, the connection weight of hidden layer and output layer, hidden layer
Threshold value, output layer threshold value.These are joined together to form a long string, above each position represent one of network power
Value and threshold value, this just constitutes an individual, produces multiple such individuals and just constitutes initial population;
Step 202, the optimal initial weight threshold value of the BP neural network obtained according to individual, BP god is trained with training data
The Error Absolute Value Δ E between system output, ideal adaptation angle value and reality output and system output is obtained through networkiIt is relevant,
Ideal adaptation angle value f (i) is exactly:
F (i)=M/ Δs Ei (3)
Wherein, M is big number, in order to prevent that fitness value is too small, side that genetic algorithm can so increased to fitness
To evolution;
Step 203, the fitness of colony is analyzed, if meeting optimization principles, directly optimum individual is exported and parameter is
Best initial weights and threshold value enter BP neural network, otherwise carry out in next step;
Step 204, selection calculating is carried out according to each individual fitness value, retains the individual for possessing high fitness value,
Fitness value maximum is directly entered the next generation, without genetic manipulations such as cross and variations, can so prevent its degeneration.Choosing
Selecting probability P (i) is:
Wherein, N is population at individual number;
Step 205, crossover operator acts on whole individual and produces a new generation, and such as i-th of individual is with j-th of individual in k
Position crossover operation be:
Wherein, α is the random number between [0,1];
Step 206, structure variation adjustment is carried out to the individual in colony using mutation operator and produces new individual, by i-th
J-th of gene a of individualijEnter row variation, mutation operation is as follows:
Wherein, amaxAnd aminIt is a respectivelyijMaximum and minimum value, α is the random number between [0,1], and g is current
Iterations, GmaxIt is maximum evolution number.
Individual after selecting cross and variation turns into colony of future generation, repeat step 203.
As shown in fig. 6, for the BP neural network execution flow chart of the present invention, specifically comprise the following steps:
Step 301, network is established:Network topology is determined, the optimal initial weight and satisfaction that will be obtained by genetic algorithm
BP neural network is substituted into after the best initial weights threshold value decoding of condition to be trained;
Step 302, give input vector and target output, first five 10 years annual most like period water levels for input to
Amount, the latter day water level of annual similar water level pattern export as target;
Step 303, the output of hidden layer and output layer and corresponding training error are asked;
Step 304, the error that is obtained according to training every time adjusts the weights of network and threshold value, is set if training error is less than
Determine error and then adjust weight matrix, repeat step 303, by iterating to obtain final network;
Step 305, predicted value was obtained according to the water level input BP neural network of 15 days a few days ago to be predicted.
Claims (2)
- A kind of 1. method using similarity searching and improved BP forecast level, it is characterised in that including:A) data prediction, including:Data selection, data cleansing, data conversion;It is water level value of the 15 days a few days ago water levels to be predicted with first five 10 years identical month that determining first, which needs data to be processed, this It is exactly data selection;Need to clean for missing present in Hydrological Time Series and noise, be allowed to not influence result Correctness is data cleansing;I.e. data conversion is smoothed to time series data, it is smooth using interpolation, data filling, sequence Transform method pre-processes to time series, is prepared for similarity searching model;B) similarity measurement:According to dynamic time warping distance, using sliding window technique, with the water of 15 days a few days ago to be predicted Position 50 years with month water level in search the most like water level period, by this 50 groups of similar water levels and corresponding latter day Water level is as training set;C) genetic algorithm:Before training, first pass through selection cross and variation genetic operation of the genetic algorithm to chromosome and find BP god Through the initial weight that network is optimal;Also, when BP neural network is absorbed in minimum, genetic algorithm optimization network ginseng is transferred to again Number;Best initial weights and threshold value are obtained after certain precision is met;D) BP neural network is reverse transmittance nerve network:By the optimal initial weight obtained by genetic algorithm and meet condition Best initial weights threshold value substitute into BP neural network be trained, obtain the error between the reality output of sample and desired output, Weight matrix is adjusted according still further to normal training philosophy, is trained again, it is last pre- according to treating until obtaining final weights The water level surveyed 15 days a few days ago obtains predicted value;The same month water level:The determination in month is searched for, according to the water regime of basin water level, annual each season or every The water level information in individual month has certain characteristic, therefore the hydrology in basin must be taken into full account when it is determined that searching for month Feature and seasonal variations rule, search model is included in month by what the water level characteristic with 15 days to be searched there are close Hydrological characteristics Enclose;After determining search month, according to dynamic time warping method, using sliding window technique, searched in 50 years annual Find out annual with the most like water level of water level on the 15th to be searched in rope month, and originated the date of expiry and be marked, The water level of annual similar water level latter day is taken out simultaneously, can thus obtain the training of 50 groups of similar water levels+latter day water level Collection;Similarity searching concretely comprises the following steps:Step 101, determine search month after, by sequence to be matched be 15 days a few days ago to be predicted water level time series and Sequence to be searched is that the water level time series for searching for month is standardized;, will for original series C using zero-Mean Method It is standardized as sequence C ', CtRepresent sequence C in t-th of element, be standardized as sequence C ' in t-th of element;Wherein u It is respectively the average value and standard deviation of the sequence with v:Step 102, the length for determining sliding window is [7,21], and matching length is initialized as into 7;Step 103, judge matching length whether between the minimum value and maximum of sliding window length i.e. [7,21], if It is then to carry out in next step, otherwise carrying out step 106;Step 104, judge whether to terminate a search sequence matching with the sliding window of this length, if not then carrying out next Step, otherwise adds a carry out step 103 by the length of sliding window;Step 105, the sliding window for calculating certain length is often moved back DTW between an obtained sequence and sequence to be matched Distance, and mark corresponding sequence to originate final position;Computational length is respectively DTW between z and c time serieses X and Y apart from mistake Journey is as follows:(1) value z corresponding to coordinate (e, r) in the matrix M, M between X and Y is constructederFor xeWith yrBetween Euclidean distance d (xe, yr);(2) accumulative matrix R is constructed, the calculation formula of value is as follows corresponding to coordinate (e, r):(3) the minimum accumulated value o of final time series crooked routez,cIt is exactly the DTW distances between time series X and Y;Calculate certain length sliding window travel through DTW distances all after one search sequence and record after, repeat step 104;Step 106, it is the water level time series in the search month of 1 year in 50 years for a search sequence, calculates and treat With the sequence DTW distance all with it, compare and show that beeline and the i.e. corresponding starting of corresponding starting final position terminate On the date, the most like sequence of 1 year in 50 years is thus obtained, using this sequence and its latter day water level as an instruction Practice sample, other times are by that analogy, then have just obtained all training sets.
- 2. existed as claimed in claim 1 using the method for similarity searching and improved BP forecast level, its feature In described to obtain optimal initial weight:Refer to produce initial population and carry out Population Coding, for participate in the input of training because Son, with real coding, i.e., the gene position by a real number directly as a chromosome, generate the weight of network, the group of coding Into partly having:The connection weight of input layer and hidden layer, the connection weight of hidden layer and output layer, hidden layer threshold value, output layer Threshold value;These are joined together to form a long string, above each position represent the weights and threshold value of network, this An individual is just constituted, multiple such individuals is produced and just constitutes initial population;It is described to obtain final best initial weights threshold value:The optimal initial weight threshold value of the BP neural network obtained according to individual, use Training data training BP neural network obtains the i.e. desired prediction output of system output, and ideal adaptation angle value and reality output are with being Error Absolute Value Δ E between system outputiIt is relevant;The low individual of fitness value is subjected to selection cross and variation, it is former to meet optimization As best initial weights threshold value then.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410454011.7A CN104239489B (en) | 2014-09-05 | 2014-09-05 | Utilize the method for similarity searching and improved BP forecast level |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410454011.7A CN104239489B (en) | 2014-09-05 | 2014-09-05 | Utilize the method for similarity searching and improved BP forecast level |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104239489A CN104239489A (en) | 2014-12-24 |
CN104239489B true CN104239489B (en) | 2018-03-20 |
Family
ID=52227548
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410454011.7A Expired - Fee Related CN104239489B (en) | 2014-09-05 | 2014-09-05 | Utilize the method for similarity searching and improved BP forecast level |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104239489B (en) |
Families Citing this family (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105046321B (en) * | 2015-06-25 | 2018-01-19 | 河海大学 | A kind of method of the Combined model forecast water level based on similarity searching |
CN105426441B (en) * | 2015-11-05 | 2018-10-16 | 华中科技大学 | A kind of automatic preprocess method of time series |
CN106203741B (en) * | 2016-08-10 | 2020-02-21 | 国家电网公司 | Multi-element heterogeneous data cleaning method for power grid load prediction |
CN106326656B (en) * | 2016-08-24 | 2018-11-09 | 东南大学 | A kind of simulating and predicting method of job facilities Severe rainstorm flood level |
CN106407672A (en) * | 2016-09-08 | 2017-02-15 | 山东腾泰医疗科技有限公司 | Mental health evaluation system based on Internet |
CN107301564A (en) * | 2017-06-12 | 2017-10-27 | 河南科技大学 | Abnormal consumer behavior detection method based on clustering algorithm and echo state network |
CN107704973A (en) * | 2017-10-31 | 2018-02-16 | 武汉理工大学 | Water level prediction method based on neutral net Yu local Kalman filtering mixed model |
CN108537247B (en) * | 2018-03-13 | 2022-03-08 | 河海大学 | Time-space multivariate hydrological time sequence similarity measurement method |
CN108846573B (en) * | 2018-06-12 | 2021-04-09 | 河海大学 | Watershed hydrological similarity estimation method based on time series kernel distance |
CN109299812B (en) * | 2018-08-23 | 2021-09-24 | 河海大学 | Flood prediction method based on deep learning model and KNN real-time correction |
CN109272146B (en) * | 2018-08-23 | 2021-10-19 | 河海大学 | Flood prediction method based on deep learning model and BP neural network correction |
CN109558436B (en) * | 2018-11-03 | 2023-03-14 | 北京交通大学 | Airport flight delay cause and effect relationship mining method based on transfer entropy |
CN111327441B (en) * | 2018-12-14 | 2022-07-08 | 中兴通讯股份有限公司 | Traffic data prediction method, device, equipment and storage medium |
CN111340176A (en) * | 2018-12-19 | 2020-06-26 | 富泰华工业(深圳)有限公司 | Neural network training method and device and computer storage medium |
CN109783051B (en) * | 2019-01-28 | 2020-05-29 | 中科驭数(北京)科技有限公司 | Time series similarity calculation device and method |
CN111401599B (en) * | 2019-08-01 | 2022-08-26 | 河海大学 | Water level prediction method based on similarity search and LSTM neural network |
CN110738355B (en) * | 2019-09-19 | 2023-07-04 | 河源职业技术学院 | Urban waterlogging prediction method based on neural network |
CN111291020A (en) * | 2019-11-11 | 2020-06-16 | 中国计量大学 | Dynamic process soft measurement modeling method based on local weighted linear dynamic system |
CN113033861A (en) * | 2019-12-25 | 2021-06-25 | 广东奥博信息产业股份有限公司 | Water quality prediction method and system based on time series model |
CN113762618B (en) * | 2021-09-07 | 2022-03-01 | 中国水利水电科学研究院 | Lake water level forecasting method based on multi-factor similarity analysis |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101625733B (en) * | 2009-08-03 | 2012-08-22 | 杭州电子科技大学 | Tidewater water level and time forecasting method based on neural network |
CN101706335B (en) * | 2009-11-11 | 2012-01-11 | 华南理工大学 | Wind power forecasting method based on genetic algorithm optimization BP neural network |
CN101819407B (en) * | 2010-04-02 | 2011-09-07 | 杭州电子科技大学 | Sewage pump station water level prediction method base on neural network |
CN102880755B (en) * | 2012-09-25 | 2014-10-08 | 河海大学 | Method and system for quantitatively forecasting extreme rainfall |
-
2014
- 2014-09-05 CN CN201410454011.7A patent/CN104239489B/en not_active Expired - Fee Related
Non-Patent Citations (3)
Title |
---|
基于BP神经网络的非线性函数拟合技术报告文档;zhaxi323232;《http://www.docin.com/p-365675598.html》;20120319;第8-14页 * |
基于遗传算法的BP神经网络模型在地下水动态预测中的应用研究;迟宝明等;《工程勘察》;20080901(第9期);第36-39页、图1-2 * |
水文时间序列相似模式挖掘的研究与应用;吴德;《中国优秀硕士学位论文全文数据库信息科技辑》;20070615(第6期);第30-35页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104239489A (en) | 2014-12-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104239489B (en) | Utilize the method for similarity searching and improved BP forecast level | |
CN106372731B (en) | A kind of high wind line of high-speed railway wind speed spatial network structure forecast method | |
CN112733996B (en) | GA-PSO (genetic algorithm-particle swarm optimization) based hydrological time sequence prediction method for optimizing XGboost | |
CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
Dariane et al. | Forecasting streamflow by combination of a genetic input selection algorithm and wavelet transforms using ANFIS models | |
CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
CN110942194A (en) | Wind power prediction error interval evaluation method based on TCN | |
CN112633604B (en) | Short-term power consumption prediction method based on I-LSTM | |
CN110070144A (en) | A kind of lake water quality prediction technique and system | |
CN103778482A (en) | Aquaculture dissolved oxygen short-term prediction method based on multi-scale analysis | |
CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
CN116090625A (en) | Coastal city flood rapid prediction method based on LightGBM and hydrological hydrodynamic model | |
CN113393057A (en) | Wheat yield integrated prediction method based on deep fusion machine learning model | |
CN109408896B (en) | Multi-element intelligent real-time monitoring method for anaerobic sewage treatment gas production | |
CN116341391B (en) | Precipitation prediction method based on STPM-XGBoost model | |
CN106021924B (en) | Sewage online soft sensor method based on more attribute gaussian kernel function fast correlation vector machines | |
CN117034060A (en) | AE-RCNN-based flood classification intelligent forecasting method | |
Hoque et al. | Prediction of groundwater level using artificial neural network and multivariate time series models | |
CN114580762A (en) | Hydrological forecast error correction method based on XGboost | |
CN114862007A (en) | Short-period gas production rate prediction method and system for carbonate gas well | |
Kim et al. | A neuro-genetic approach for daily water demand forecasting | |
CN117669008B (en) | Foundation settlement prediction method and system based on deep learning | |
Jahangir et al. | Application of artificial neural networks to the simulation of climate elements, drought forecast by two indicators of SPI and PNPI, and mapping of drought intensity; case study of Khorasan Razavi | |
CN111325384A (en) | NDVI prediction method combining statistical characteristics and convolutional neural network model | |
CN112529403B (en) | Method for determining construction land area influence factor weight value by using neural network 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 | ||
GR01 | Patent grant | ||
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: 20180320 Termination date: 20210905 |