CN110414742A - A kind of freshwater line intelligent Forecasting of more gaging station linkages - Google Patents
A kind of freshwater line intelligent Forecasting of more gaging station linkages Download PDFInfo
- Publication number
- CN110414742A CN110414742A CN201910712704.4A CN201910712704A CN110414742A CN 110414742 A CN110414742 A CN 110414742A CN 201910712704 A CN201910712704 A CN 201910712704A CN 110414742 A CN110414742 A CN 110414742A
- Authority
- CN
- China
- Prior art keywords
- water level
- gaging
- value
- none
- layer
- 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
- 239000013505 freshwater Substances 0.000 title claims abstract description 17
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 78
- 230000000644 propagated effect Effects 0.000 claims abstract description 13
- 238000001914 filtration Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 4
- 238000000034 method Methods 0.000 claims description 36
- 230000008569 process Effects 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 7
- 230000009467 reduction Effects 0.000 claims description 7
- 210000004205 output neuron Anatomy 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 5
- 230000007787 long-term memory Effects 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000011144 upstream manufacturing Methods 0.000 abstract description 7
- 238000013528 artificial neural network Methods 0.000 abstract description 5
- 230000000306 recurrent effect Effects 0.000 abstract description 3
- 238000012549 training Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 241001269238 Data Species 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 125000006850 spacer group Chemical group 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012300 Sequence Analysis Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
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
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention provides a kind of freshwater line intelligent Forecasting of more gaging station linkages, comprising: prepares sample data, the filling of vacancy value and the processing of SG filtering are carried out to it;Construct the three gaging stations linkage prediction model based on GRU;The three gaging stations linkage prediction model constructed in step S2 is trained and is predicted using propagated forward and backpropagation.Technical solution of the present invention constructs forecast model of water level using GRU (gating cycle unit) Recognition with Recurrent Neural Network, and unlike the prior art, this programme not only allows for the time relationship of water level value, while also contemplating the spatial relationship between gaging station.In order to predict the water level value of some gaging station, this programme not only relies only on the historical data of the gaging station, it assists establishing prediction model using the historical data at its neighbouring water levels of upstream and downstream station simultaneously, to reduce the error of single station data, the universal law for preferably reflecting SEA LEVEL VARIATION, improves the accuracy of water level forecast.
Description
Technical field
The present invention relates to water level forecast technical fields, specifically, more particularly to a kind of interior river water of more gaging station linkages
Position intelligent Forecasting.
Background technique
China's cruiseway is resourceful, including the water systems such as the Changjiang river, the Zhujiang River and Beijing-Hangzhou Grand Canal, total cruising range reach
12.70 ten thousand km, have become the important component of each regional complex transportation system in China.The water level of cruiseway is as boat
The main indicator of road scale maintenance is an important factor for instructing the reasonable prestowage of ship and ensure marine operation safety.Comprehensively and
When, accurately perceive water level information along cruiseway, and rational prediction water level short term variations trend is logical for promoting navigation channel
Row ability ensures that navigation safety and science carry out the most important of navigation channel maintenance.However, since cruiseway is mostly width
The thalweg of alternate rhizoma nelumbinis shape braided rcach, local section is swung, and main secondary channel replaces growth and decline, some inland rivers are combined by multi-reservoir adjusts
The factors such as degree, basin changes and precipitation influence, and the features such as non-stationary, non-linear are presented in navigation channel SEA LEVEL VARIATION.In view of inland river nature ring
The property complicated and changeable in border, therefore, it is difficult to which accurately and effectively Physics-mathematics model is set up in the prediction for SEA LEVEL VARIATION trend.
Recently as the development of telemetry remote control technology, by the setting of gaging station along the line, many navigation channel departments are
The multidate information of water level can be grasped comprehensively, in time, also grasped a large amount of fine-grained history waterlevel datas.Utilize big number
According to the method with artificial intelligence, model is established based on history waterlevel data, water level will be become to calculate to following water level
The new tool of prediction.
History waterlevel data substantially belongs to time series data.Time series is typically referred to the number of certain statistical indicator
Value, in chronological sequence sequence arrangement are formed by ordered series of numbers.Time series data itself contains enough information content, can therefrom seek
Its timing and regularity are looked for, obtains its development process, direction and the trend that reflect, and further with such
It pushes away or extends, predict the level that statistical indicator of lower a period of time is likely to be breached.This time Sequence Analysis Method can be predicted in freshwater line
It is upper to obtain preferable effect.How scientificlly and effectively these history waterlevel datas to be pre-processed, how to establish efficiently and accurately
Prediction model, researcher carry out further investigation and extensively experiment.The forecast model of water level of one efficiently and accurately is for mentioning
Traffic capacity of waterway is risen, navigation channel comprehensive service capability is promoted, even takes precautions against natural calamities and be all of great significance.
In view of the above-mentioned problems, application number 201810464065.X, patent name is a kind of water levels of upstream and downstream of step hydropower station
Prediction technique, the invention are to predict power station water level, and but merely with the water level in upstream power station, lower station water level value is same
It is easy observation.If the water level value in upstream and downstream power station is used in combination, the powerful learning ability of LSTM may learn two power stations
SEA LEVEL VARIATION correlativity improves accuracy.Application number 201810689104.6, patent name are to be followed based on cost-sensitive LSTM
The voltage-stablizer water level prediction method of ring neural network, only optimizes from training method, and core is sensitive weight variable pair
Master mould loss function improves, and there are many more optimization methods, such as data normalized, regularization.Prediction accuracy is also
There is biggish room for promotion.In addition, being directed to same river, upstream is bound to have an impact downstream water place value, same lower section
Water level value can also be delayed to a certain extent reacts the variation tendency of upper water place value, i.e. the spatially water between neighbouring gaging station
There is incidence relations for position.
Summary of the invention
Predict that the present invention proposes a kind of method based on Recognition with Recurrent Neural Network, a large amount of from what is be collected into for freshwater line
Learning knowledge in historical data, and three gaging stations linkage prediction model is established, realize the prediction more accurate to water level.Model
Net effect is up to: the water level value of input gaging station (such as 20 days) within the next few days, export its following (such as 5 days) accuracy in several days compared with
High forecast level.This programme constructs the learning model of water level forecast using GRU (gating cycle unit) Recognition with Recurrent Neural Network,
Unlike the prior art, this programme not only allows for the time relationship of water level value, while also contemplating the space between gaging station
Relationship.In order to predict the water level value of some gaging station, this programme not only relies only on the historical data of the gaging station, while utilizing it
The historical data at neighbouring water levels of upstream and downstream station assists establishing prediction model, to reduce the error of single station data, preferably
The universal law for reflecting SEA LEVEL VARIATION, improves the accuracy of water level forecast.
The technological means that the present invention uses is as follows:
A kind of freshwater line intelligent Forecasting of more gaging station linkages, comprising the following steps:
Step S1: preparing sample data, and the filling of vacancy value and the processing of SG filtering are carried out to it;
Step S2: three gaging stations linkage prediction model of the building based on GRU;
Step S3: the three gaging stations linkage prediction model constructed in step S2 is carried out using propagated forward and backpropagation
Training and prediction.
Further, the hollow missing value filling of the step S1 is filled using average value, i.e., the water level of 4 points around vacancy item
It is averaged, such as water level value x1,x2,x3,x4,x5, wherein x3For vacancy item, then Filling power are as follows:
Further, SG filtering treatment process includes: in the step S1
Step S11: to the data point in window, being fitted using following formula, and k-1 is fitting number:
P (n)=a0+a1n+a2n2+...+ak-1nk-1
Wherein, n value is 0, ± 1, ± 2 ...;P (n) indicates the good water level of noise reduction, parameter a in formula0,a1,...,ak-1It is logical
Least square method is crossed to determine;
Step S12: the residual error of least square method fitting is set as:
Wherein, window width 2m+1, i.e. above formula n successively take-m,-m+1, and -2, -1,0,1,2 ..., m.Wherein, x (n)
Window is determined to reach noise reduction smooth effect by continuous moving window to entire water level sample data set for sample observations
Mouth width degree and fitting number.
Further, three gaging station linkage prediction model includes input layer, the first GRU layers, the 2nd GRU layers and defeated
Layer out;
The input dimension [(None, X, Y)] of the input layer, wherein None indicates the data volume of primary input network, X
Indicate step-length, Y is characterized quantity;
Described first GRU layers of activation primitive is set as tanh, connects input layer, and output is [(None, A, B)], wherein
The numerical value of None is identical as input layer, and A indicates step-length, and B indicates this layer of output neuron number;
Described 2nd GRU layers of activation primitive is set as relu, GRU layer of connection the first, and output is [(None, C)], wherein
The numerical value of None is identical as input layer, and C is this layer of output neuron number, docks output layer;
The output layer is full articulamentum, does not set activation primitive, is exported dimension [(None, Z)], it is defeated that None represents network
Data volume out, Z are characterized quantity.
Further, the propagated forward process used in the step S3 is as follows:
rt=σ (Wr[ht-1,xt])
zt=σ (Wz[ht-1,xt])
yt=σ (Woht)
Wherein, [ht-1,xt] indicate two vector ht-1,xtIt is longitudinal spliced together, ht-1Indicate the hiding information of water level,
xtIndicate the waterlevel data of input, Wr,Wz,Wh,WoFor network parameter, that is, weight information;rt, ztIt respectively indicates associated gate, update door
Value, codomain be (0,1), pass through control door switch restricted information inflow;htIndicate recall info (comprising to current time
The information of water level before step), work as ztWhen=0, there is ht=ht-1Even if the value of recall info remains to by several time steps
It remains unchanged, that is, has long-term memory function.
Further, the back-propagation process used in the step S3 is as follows:
Step S31: to the parameter W in the propagated forwardr,Wz,Wh,WoIt is updated, wherein WoOnly at the last one
Spacer step includes, and Wr,Wz,WhInclude in each time step;Wr,Wz,WhIt is to be spliced by two weights, complete form is such as
Under:
Wr=Wrz+Wrh
Wz=Wzx+Wzh
Step S32: loss function is defined as the square error between sample water level and forecast level:
So then have:
δt=δh,tzttanh'
Wherein, y is model output value,For sample value, δ*Indicate each error term,
Further, the mistake three gaging stations linkage prediction model constructed in step S2 being trained in the step S3
It further include by Adam optimization method, reducing J (θ) constantly until restrain the step of in every layer of addition L2 canonical in journey;It adds
Cost function after entering canonical is as follows:
Wherein, m indicates sample size, and λ=0.001, θ are model parameter,Indicate sample water level and prediction
Square error between water level.
Compared with the prior art, the invention has the following advantages that
1, the freshwater line intelligent Forecasting of more gaging station linkages provided by the invention, in terms of instructing ship's navigation,
When drauht is deeper, it can be driven into when forecast level value is big, avoid stranded generation, to ensure navigation safety simultaneously
Ship-lifting efficiency of navigation.
2, the freshwater line intelligent Forecasting of more gaging station linkages provided by the invention, in terms of sea-route management, the depth of water
Line is exactly lifeline.Water level forecast can instruct sea-route management department, in conjunction with hydrological observation as a result, correct publication bulletin, it is ensured that linchpin
Navigation safety in area.
3, forecast level value is significant to power station, when water level value is high, brings biggish flow, can generate more
Electric energy.Forecast level is known in advance, technical staff's reasonable arrangement in power station can be helped to work, and improves working efficiency.
4, forecast level value also can be applicable in flood forecasting, predict water level in advance, carry out flood control measure, can greatly reduce
Various losses.
To sum up, apply the technical scheme of the present invention the accurate prediction realized to freshwater line, in ship's navigation, sea-route management
And many aspects such as flood control have important application, have apparent economic and social benefit.
The present invention can be widely popularized in fields such as water level forecasts based on the above reasons.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to do simply to introduce, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without any creative labor, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the method for the present invention flow chart.
Fig. 2 is SG filtering treatment process figure provided in an embodiment of the present invention.
Fig. 3 is comparison diagram of the water level sample data provided in an embodiment of the present invention before and after SG filtering.
Fig. 4 is the upstream and downstream spatial relation graph of three gaging station provided in an embodiment of the present invention.
Fig. 5 is three gaging station provided in an embodiment of the present invention linkage prediction model structure chart.
Fig. 6 is the structural schematic diagram of GRU provided in an embodiment of the present invention.
Fig. 7 is three gaging station provided in an embodiment of the present invention linkage prediction model structure expanded view.
Fig. 8 is GRU layers of schematic diagram provided in an embodiment of the present invention.
Fig. 9 is three gaging station provided in an embodiment of the present invention linkage water level forecast result schematic diagram.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, below in conjunction in the embodiment of the present invention
Attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is only
The embodiment of a part of the invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
The model that the present invention protects all should belong in member's every other embodiment obtained without making creative work
It encloses.
It should be noted that description and claims of this specification and term " first " in above-mentioned attached drawing, "
Two " etc. be to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should be understood that using in this way
Data be interchangeable under appropriate circumstances, so as to the embodiment of the present invention described herein can in addition to illustrating herein or
Sequence other than those of description is implemented.In addition, term " includes " and " having " and their any deformation, it is intended that cover
Cover it is non-exclusive include, for example, the process, method, system, product or equipment for containing a series of steps or units are not necessarily limited to
Step or unit those of is clearly listed, but may include be not clearly listed or for these process, methods, product
Or other step or units that equipment is intrinsic.
As shown in Figure 1, the present invention provides a kind of freshwater line intelligent Forecasting of more gaging station linkages, including it is following
Step:
Step S1: preparing sample data, pre-process to it, i.e. the filling of vacancy value and the processing of SG filtering;
Step S2: three gaging stations linkage prediction model of the building based on GRU;
Step S3: the three gaging stations linkage prediction model constructed in step S2 is carried out using propagated forward and backpropagation
Training and prediction.
Embodiment 1
When analyzing time series data, the quality of data is very big on the influence of the result of analysis, and a quality is low
Under time series data no matter how ingenious temporal model is, be all difficult to reach ideal analytical effect.The present embodiment is to sample data
It is pre-processed as follows.
Specifically, the hollow missing value filling of step S1 is filled using average value, 4 points around vacancy item are used in the present embodiment
Water level be averaged, such as water level value x1,x2,x3,x4,x5, wherein x3For vacancy item, then Filling power are as follows:
Illustrate smoothing process by taking smooth (window width 5) as an example at 5 points in the present embodiment.As shown in Fig. 2, step S1
Middle SG filtering treatment process includes:
Step S11: to the data point in window, being fitted using following formula, and k-1 is fitting number:
P (n)=a0+a1n+a2n2+...+ak-1nk-1
Wherein, n value is 0, ± 1, ± 2 ....P (n) indicates the good water level of noise reduction, parameter a in formula0,a1,...,ak-1It is logical
Least square method is crossed to determine;
Step S12: the residual error of least square method fitting is set as:
Wherein, window width 2m+1, i.e. above formula n successively take-m,-m+1, and -2, -1,0,1,2 ..., m.Wherein, x (n)
Window is determined to reach noise reduction smooth effect by continuous moving window to entire water level sample data set for sample observations
Mouth width degree and fitting number.
To entire water level sample data set, by continuous moving window, to reach noise reduction smooth effect.Fig. 3 shows water
Comparison of the position sample data before and after SG filtering, the window width of used SG filter are 9, and fitting number is 5.It can by Fig. 3
See, the filtered sample data waveform of SG is more steady, the essence for disclosing time series data is more advantageous to, to be effectively reduced
Influence of the random error to last prediction result.Subsequent water level forecast comparative analysis is also demonstrated compared to untreated original number
According to the pretreatment of SG filter significantly improves the precision of prediction.The last water level forecast result of this programme is filtered by SG
Device carries out the prediction result after data prediction.
Embodiment 2
Using the Changjiang river as example, the three gaging stations linkage prediction model of this programme is illustrated.As shown in figure 4, the Changjiang river edge
Nanjing, Wuhu and three ground of Anqing of line are equipped with gaging station, and wherein the position in Wuhu is between Nanjing and Anqing.
In the present embodiment, three gaging stations based on GRU of building link prediction model, as shown in figure 5, include input layer,
First GRU layers, the 2nd GRU layers and output layer;
The input dimension [(None, 20,3)] of input layer, wherein None indicates the data volume batch_ of primary input network
Size (such as: being set as 128), 20 indicate step-length, i.e., using 20 days in the past waterlevel datas as input;3 are characterized quantity;I.e.
The water level value inputted daily includes 3 features, i.e. tactic 3 water level values of the gaging station of upper, middle and lower reaches 3.
First GRU layers of activation primitive is set as tanh, connects input layer, and output is [(None, 20,512)], wherein
None is batch_size (numerical value is identical as input layer), and 20 indicate step-length (timestep), and 512 indicate this layer of output neuron
Number;
2nd GRU layers of activation primitive is set as relu, GRU layer of connection the first, and output is [(None, 512)], and None is
Batch_size (numerical value is identical as input layer), 512 be this layer of output neuron number, docks output layer;
Output layer is full articulamentum, does not set activation primitive, is exported dimension [(None, 15)], and None represents network output
Data volume batch_size, 15 are characterized quantity.That is 5 days 15 water level values after 3 gaging stations are respectively predicted.
Contain 2 GRU layers in said structure, the specific structure of GRU is as shown in fig. 6, in figure, ht-1To receive upper one
The output information of a GRU unit, xtFor the input water level value of current time step, [h can be usedt-1,xt] indicate the vertical of the two vectors
To splicing;htIt indicates recall info (memory comprising water level information before being walked to current time);rt, ztRespectively indicate associated gate,
The value of door is updated, codomain is (0,1), passes through the inflow of the switch restricted information of control door.Relationship in figure between each parameter is as follows
The description of face formula:
rt=σ (Wr[ht-1,xt])
zt=σ (Wz[ht-1,xt])
yt=σ (Woht)
Wherein, [ht-1,xt] indicate two vector ht-1,xtIt is longitudinal spliced together, ht-1Indicate the hiding information of water level,
xtIndicate the waterlevel data of input, Wr,Wz,Wh,WoFor network parameter, that is, weight information.
As can be seen from the above formula that working as ztWhen=0, there is ht=ht-1, indicate the recall info after several time steps
Value remain to remain unchanged, that is, possess long-term memory function.ztSpecific value will be by minimizing cost function by model training
And it obtains.The dependence that GRU can guarantee that span is very big in RNN model is unaffected, thus solve the problems, such as gradient disappearance, this
It is not available for standard neural network structure.
If setting batch_size=128, then the input of three above-mentioned gaging stations linkage forecast model of water level structure is
(128,20,3) are calculated output (128,15) by model as a result, its expanded schematic diagram is as shown in Figure 7.
Embodiment 3
The training of model is divided into two parts, propagated forward and backpropagation.The purpose of propagated forward is provide model defeated
Out, backpropagation is used to update network weight.According to above-mentioned model structure, propagated forward and backpropagation emphasis are mainly in GRU
On model, it is 20 (as shown in Figure 8) that time step, which is assumed below, and the training to GRU layers is illustrated.
Propagated forward in step S3 finally obtains output y as shown in fig. 6, determined by following formulat。
rt=σ (Wr[ht-1,xt])
zt=σ (Wz[ht-1,xt])
yt=σ (Woht)
Wherein, [ht-1,xt] indicate two vector ht-1,xtIt is longitudinal spliced together, ht-1Indicate the hiding information of water level,
xtIndicate the waterlevel data of input, Wr,Wz,Wh,WoFor network parameter, that is, weight information.rt, ztIt respectively indicates associated gate, update door
Value, codomain be (0,1), pass through control door switch restricted information inflow;htIndicate recall info (comprising to current time
The information of water level before step), work as ztWhen=0, there is ht=ht-1Even if the value of recall info remains to by several time steps
It remains unchanged, that is, has long-term memory function.
The back-propagation process used in step S3 is as follows:
Step S31: to the parameter W in the propagated forwardr,Wz,Wh,WoIt is updated, wherein WoOnly at the last one
Spacer step includes, and Wr,Wz,WhInclude in each time step;Wr,Wz,WhIt is to be spliced by two weights, complete form is such as
Under:
Wr=Wrz+Wrh
Wz=Wzx+Wzh
Step S32: loss function is defined as the square error between sample water level and forecast level:
So then have:
δt=δh,tzttanh'
Wherein, y is model output value,For sample value, δ*Indicate each error term,
Each parameter of model can be updated in back-propagation process by above formula.
In the training process, to avoid serious over-fitting, in every layer of addition L2 canonical, wherein λ=0.001.It is added just
Cost function after then is as follows:
Wherein, m indicates sample size, and θ is model parameter,It indicates between sample water level and forecast level
Square error.By Adam optimization method, reduce J (θ) constantly until convergence, that is, learn θ.
Embodiment 4
Waterlevel data when in the present solution, Anqing, Wuhu and 30 years daily 8 each three, Nanjing gaging station is utilized carries out
Model training and test.
It in actual prediction, enables batch_size=1 (changing 128 in Fig. 7 into 1), i.e. input dimension [1,20,3], it is defeated
Dimension [1,15] out.It can be obtained the output of 15 water level values, this 15 water level values contain three gaging stations each 5 days backward pre-
Measured value.5 days water level forecast values for wherein belonging to middle portion station (such as Wuhu) are only taken out, i.e. realization prediction task.
Fig. 9 illustrates the test result (interception wherein 200 days) of three gaging stations linkage prediction model.In figure, red lines
Indicate actual observed value, black lines indicate the value of models fitting, and blue lines indicate predicted value.The song of Wuhu gaging station in figure
Line is in the centre of three groups of lines.The error analysis of model prediction result is as shown in table 1.
Seen from table 1, the mean error between model predication value and measured value is 4.67%, wherein first 3 days opposite mistakes
For difference less than 3%, prediction accuracy is higher compared with other methods.
The analysis of 1 model prediction resultant error of table
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations;To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement;And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (7)
1. a kind of freshwater line intelligent Forecasting of more gaging station linkages, which comprises the following steps:
Step S1: preparing sample data, and the filling of vacancy value and the processing of SG filtering are carried out to it;
Step S2: three gaging stations linkage prediction model of the building based on GRU;
Step S3: the three gaging stations linkage prediction model constructed in step S2 is trained using propagated forward and backpropagation
And prediction.
2. the freshwater line intelligent Forecasting of more gaging station linkages according to claim 1, which is characterized in that the step
The rapid hollow missing value filling of S1 is filled using average value, i.e., the water level of 4 points is averaged around vacancy item, such as water level value x1,x2,x3,
x4,x5, wherein x3For vacancy item, then Filling power are as follows:
3. the freshwater line intelligent Forecasting of more gaging station linkages according to claim 1, which is characterized in that the step
SG filtering treatment process includes: in rapid S1
Step S11: to the data point in window, being fitted using following formula, and k-1 is fitting number:
P (n)=a0+a1n+a2n2+...+ak-1nk-1
Wherein, n value is 0, ± 1, ± 2 ...;P (n) indicates the good water level of noise reduction, parameter a in formula0,a1,...,ak-1By most
Small square law determines;
Step S12: the residual error of least square method fitting is set as:
Wherein, window width 2m+1, i.e. above formula n successively take-m,-m+1, and -2, -1,0,1,2 ..., m;X (n) is sample observation
Value, by continuous moving window, to reach noise reduction smooth effect, determines window width and intends to entire water level sample data set
Close number.
4. the freshwater line intelligent Forecasting of more gaging station linkages according to claim 1, which is characterized in that described three
Gaging station linkage prediction model includes input layer, the first GRU layers, the 2nd GRU layers and output layer;
The input dimension [(None, X, Y)] of the input layer, wherein None indicates the data volume of primary input network, and X is indicated
Step-length, Y are characterized quantity;
Described first GRU layers of activation primitive is set as tanh, connects input layer, and output is [(None, A, B)], wherein None
Numerical value it is identical as input layer, A indicates step-length, and B indicates this layer of output neuron number;
Described 2nd GRU layers of activation primitive is set as relu, GRU layer of connection the first, and output is [(None, C)], wherein None
Numerical value it is identical as input layer, C be this layer of output neuron number, dock output layer;
The output layer is full articulamentum, does not set activation primitive, is exported dimension [(None, Z)], and None represents network output
Data volume, Z are characterized quantity.
5. the freshwater line intelligent Forecasting of more gaging station linkages according to claim 1, which is characterized in that the step
The propagated forward process used in rapid S3 is as follows:
rt=σ (Wr[ht-1,xt])
zt=σ (Wz[ht-1,xt])
yt=σ (Woht)
Wherein, [ht-1,xt] indicate two vector ht-1,xtIt is longitudinal spliced together, ht-1Indicate the hiding information of water level, xtTable
Show the waterlevel data of input, Wr,Wz,Wh,WoFor network parameter, that is, weight information;rt, ztIt respectively indicates associated gate, update door
Value, codomain are (0,1), pass through the inflow of the switch restricted information of control door;htIndicate recall info (comprising walking to current time
The information of water level before), work as ztWhen=0, there is ht=ht-1Even if the value of recall info remains to protect by several time steps
It holds constant, that is, has long-term memory function.
6. the freshwater line intelligent Forecasting of more gaging station linkages according to claim 1, which is characterized in that the step
The back-propagation process used in rapid S3 is as follows:
Step S31: to the parameter W in the propagated forwardr,Wz,Wh,WoIt is updated, wherein WoOnly in the last one time step
Include, and Wr,Wz,WhInclude in each time step;Wr,Wz,WhIt is to be spliced by two weights, complete form is as follows:
Wr=Wrz+Wrh
Wz=Wzx+Wzh
Step S32: loss function is defined as the square error between sample water level and forecast level:
So then have:
δt=δh,tzttanh'
Wherein, y is model output value,For sample value, δ*Indicate each error term,
7. the freshwater line intelligent Forecasting of more gaging station linkages according to claim 1, which is characterized in that the step
It further include in every layer of addition L2 during being trained in rapid S3 to the three gaging stations linkage prediction model constructed in step S2
Canonical reduces J (θ) constantly until restrain the step of by Adam optimization method;Cost function after its addition canonical is such as
Under:
Wherein, m indicates sample size, and λ=0.001, θ are model parameter,Indicate sample water level and forecast level
Between square error.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910712704.4A CN110414742A (en) | 2019-08-02 | 2019-08-02 | A kind of freshwater line intelligent Forecasting of more gaging station linkages |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910712704.4A CN110414742A (en) | 2019-08-02 | 2019-08-02 | A kind of freshwater line intelligent Forecasting of more gaging station linkages |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110414742A true CN110414742A (en) | 2019-11-05 |
Family
ID=68365549
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910712704.4A Pending CN110414742A (en) | 2019-08-02 | 2019-08-02 | A kind of freshwater line intelligent Forecasting of more gaging station linkages |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110414742A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110991776A (en) * | 2020-03-04 | 2020-04-10 | 浙江鹏信信息科技股份有限公司 | Method and system for realizing water level prediction based on GRU network |
CN111062476A (en) * | 2019-12-06 | 2020-04-24 | 重庆大学 | Water quality prediction method based on gated circulation unit network integration |
CN111242344A (en) * | 2019-12-11 | 2020-06-05 | 大连海事大学 | Intelligent water level prediction method based on cyclic neural network and convolutional neural network |
CN111461192A (en) * | 2020-03-25 | 2020-07-28 | 长江水资源保护科学研究所 | River channel water level flow relation determination method based on multi-hydrological station linkage learning |
CN111898250A (en) * | 2020-07-03 | 2020-11-06 | 武汉大学 | Estuary tailing branch of a river prediction method and device based on multi-source data fusion |
CN112528557A (en) * | 2020-11-30 | 2021-03-19 | 北京金水信息技术发展有限公司 | Flood flow prediction system and method based on deep learning |
CN112668711A (en) * | 2020-11-30 | 2021-04-16 | 西安电子科技大学 | Flood flow prediction method and device based on deep learning and electronic equipment |
CN115034497A (en) * | 2022-06-27 | 2022-09-09 | 武汉理工大学 | Multi-site daily water level prediction method and device, electronic equipment and computer medium |
CN115271186A (en) * | 2022-07-18 | 2022-11-01 | 福建中锐网络股份有限公司 | Reservoir water level prediction early warning method based on delay factor and PSO RNN Attention model |
CN116485003A (en) * | 2023-03-03 | 2023-07-25 | 大连海事大学 | Multi-step channel water level prediction method and device based on echo algorithm and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105321120A (en) * | 2014-06-30 | 2016-02-10 | 中国农业科学院农业资源与农业区划研究所 | Assimilation evapotranspiration and LAI (leaf area index) region soil moisture monitoring method |
WO2018014658A1 (en) * | 2016-07-22 | 2018-01-25 | 上海海洋大学 | Ommastrephidaeentral fishing ground prediction method |
CN108764539A (en) * | 2018-05-15 | 2018-11-06 | 中国长江电力股份有限公司 | A kind of water levels of upstream and downstream prediction technique of step hydropower station |
-
2019
- 2019-08-02 CN CN201910712704.4A patent/CN110414742A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105321120A (en) * | 2014-06-30 | 2016-02-10 | 中国农业科学院农业资源与农业区划研究所 | Assimilation evapotranspiration and LAI (leaf area index) region soil moisture monitoring method |
WO2018014658A1 (en) * | 2016-07-22 | 2018-01-25 | 上海海洋大学 | Ommastrephidaeentral fishing ground prediction method |
CN108764539A (en) * | 2018-05-15 | 2018-11-06 | 中国长江电力股份有限公司 | A kind of water levels of upstream and downstream prediction technique of step hydropower station |
Non-Patent Citations (10)
Title |
---|
冯兴东: "《分布式统计计算》", 30 April 2018, pages: 277 - 283 * |
宋知用: "《MATLAB数字信号处理》", 30 November 2016, pages: 116 - 125 * |
朱星明等: "基于人工神经网络的洪水水位预报模型", 《水利学报》 * |
朱星明等: "基于人工神经网络的洪水水位预报模型", 《水利学报》, no. 07, 28 July 2005 (2005-07-28), pages 805 - 811 * |
桑海峰等: "基于双向GRU和注意力机制模型的人体动作预测", 《计算机辅助设计与图形学学报》 * |
桑海峰等: "基于双向GRU和注意力机制模型的人体动作预测", 《计算机辅助设计与图形学学报》, 31 July 2019 (2019-07-31), pages 1 * |
王复明等: "第四届中国水利水电岩土力学与工程学术讨论会暨第七届全国水利工程渗流学术研讨会论文集", 黄河水利出版社, pages: 4 * |
王玮等: "基于门控循环单元神经网络的PM_2.5浓度预测", 《无线互联科技》 * |
王玮等: "基于门控循环单元神经网络的PM_2.5浓度预测", 《无线互联科技》, 28 February 2019 (2019-02-28), pages 2 * |
陈睿鹤: "基于小波变换和GRU深度神经网络的地下水位预测研究", 《中国优秀博硕士学位论文全文数据库》, 15 June 2019 (2019-06-15), pages 2 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111062476A (en) * | 2019-12-06 | 2020-04-24 | 重庆大学 | Water quality prediction method based on gated circulation unit network integration |
CN111242344A (en) * | 2019-12-11 | 2020-06-05 | 大连海事大学 | Intelligent water level prediction method based on cyclic neural network and convolutional neural network |
CN110991776A (en) * | 2020-03-04 | 2020-04-10 | 浙江鹏信信息科技股份有限公司 | Method and system for realizing water level prediction based on GRU network |
CN111461192A (en) * | 2020-03-25 | 2020-07-28 | 长江水资源保护科学研究所 | River channel water level flow relation determination method based on multi-hydrological station linkage learning |
CN111898250A (en) * | 2020-07-03 | 2020-11-06 | 武汉大学 | Estuary tailing branch of a river prediction method and device based on multi-source data fusion |
CN111898250B (en) * | 2020-07-03 | 2022-03-15 | 武汉大学 | Estuary tailing branch of a river prediction method and device based on multi-source data fusion |
CN112668711A (en) * | 2020-11-30 | 2021-04-16 | 西安电子科技大学 | Flood flow prediction method and device based on deep learning and electronic equipment |
CN112528557A (en) * | 2020-11-30 | 2021-03-19 | 北京金水信息技术发展有限公司 | Flood flow prediction system and method based on deep learning |
CN112668711B (en) * | 2020-11-30 | 2023-04-18 | 西安电子科技大学 | Flood flow prediction method and device based on deep learning and electronic equipment |
CN115034497A (en) * | 2022-06-27 | 2022-09-09 | 武汉理工大学 | Multi-site daily water level prediction method and device, electronic equipment and computer medium |
CN115271186A (en) * | 2022-07-18 | 2022-11-01 | 福建中锐网络股份有限公司 | Reservoir water level prediction early warning method based on delay factor and PSO RNN Attention model |
CN115271186B (en) * | 2022-07-18 | 2024-03-15 | 福建中锐网络股份有限公司 | Reservoir water level prediction and early warning method based on delay factor and PSO RNN Attention model |
CN116485003A (en) * | 2023-03-03 | 2023-07-25 | 大连海事大学 | Multi-step channel water level prediction method and device based on echo algorithm and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110414742A (en) | A kind of freshwater line intelligent Forecasting of more gaging station linkages | |
CN104318325B (en) | Many basin real-time intelligent water quality prediction methods and system | |
CN105223937B (en) | Hydropower Stations ecological regulation and control intelligence control system and method | |
CN109285346A (en) | A kind of city road net traffic state prediction technique based on key road segment | |
CN111242344A (en) | Intelligent water level prediction method based on cyclic neural network and convolutional neural network | |
CN106779151B (en) | A kind of line of high-speed railway wind speed multi-point multi-layer coupling prediction method | |
CN103226741B (en) | Public supply mains tube explosion prediction method | |
Minglei et al. | Classified real-time flood forecasting by coupling fuzzy clustering and neural network | |
CN108711847A (en) | A kind of short-term wind power forecast method based on coding and decoding shot and long term memory network | |
CN104376371B (en) | A kind of distribution based on topology is layered load forecasting method | |
CN106355540A (en) | Small- and medium-sized reservoir dam safety evaluating method based on GRA-BP (grey relational analysis and back propagation) neural network | |
Areerachakul et al. | Rainfall-Runoff relationship for streamflow discharge forecasting by ANN modelling | |
CN105678417A (en) | Prediction method and device for tunnel face water inflow of construction tunnel | |
CN111160620A (en) | Short-term wind power prediction method based on end-to-end memory network | |
CN108710964A (en) | A kind of prediction technique of Fuzzy time sequence aquaculture water quality environmental data | |
CN114021836A (en) | Multivariable reservoir water inflow amount prediction system based on different-angle fusion, training method and application | |
CN108615098A (en) | Water supply network pipeline burst Risk Forecast Method based on Bayesian survival analysis | |
CN115271186A (en) | Reservoir water level prediction early warning method based on delay factor and PSO RNN Attention model | |
Adnan et al. | New Artificial Neural Network and Extended Kalman Filter hybrid model of flood prediction system | |
Jia et al. | Water quality prediction method based on LSTM-BP | |
Ranjbar et al. | Framework for a digital twin of the Canal of Calais | |
CN111400973A (en) | Method for constructing flow-water surface width relation curve based on hydrologic monitoring data | |
CN113393027A (en) | Navigation mark drift intelligent prediction method based on deep learning | |
Xiaojian et al. | A traffic flow forecasting model based on BP neural network | |
CN113505492A (en) | Scheduling method of cross-basin water transfer project based on digital twin technology |
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: 20191105 |