CN111553394B - Reservoir water level prediction method based on cyclic neural network and attention mechanism - Google Patents
Reservoir water level prediction method based on cyclic neural network and attention mechanism Download PDFInfo
- Publication number
- CN111553394B CN111553394B CN202010312062.1A CN202010312062A CN111553394B CN 111553394 B CN111553394 B CN 111553394B CN 202010312062 A CN202010312062 A CN 202010312062A CN 111553394 B CN111553394 B CN 111553394B
- Authority
- CN
- China
- Prior art keywords
- water level
- neural network
- reservoir
- model
- recurrent 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.)
- Active
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 112
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 24
- 230000007246 mechanism Effects 0.000 title claims abstract description 23
- 125000004122 cyclic group Chemical group 0.000 title claims abstract description 6
- 230000000306 recurrent effect Effects 0.000 claims abstract description 32
- 238000003062 neural network model Methods 0.000 claims abstract description 26
- 238000004364 calculation method Methods 0.000 claims abstract description 25
- 238000012360 testing method Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims description 15
- 238000012795 verification Methods 0.000 claims description 14
- 230000000694 effects Effects 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 10
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 230000001131 transforming effect Effects 0.000 claims description 2
- 238000009825 accumulation Methods 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 16
- 238000013507 mapping Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 241000233805 Phoenix Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000004941 influx Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 238000011144 upstream manufacturing Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/045—Combinations of networks
-
- 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
-
- 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
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Strategic Management (AREA)
- Computational Linguistics (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Primary Health Care (AREA)
- Evolutionary Biology (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the field of reservoir operation control, and discloses a reservoir water level prediction method based on a recurrent neural network and an attention mechanism, which comprises the following steps: collecting flow and water level data to form a data set, and classifying the data set; converting the water level and flow values of the data set to obtain a standard data set; constructing a cyclic neural network model with an attention mechanism, and defining a target function; selecting data of a standard data set to train the recurrent neural network model; verifying and testing the trained recurrent neural network model; and predicting the water level of the key section by using the trained recurrent neural network model. The calculation precision of the model can be continuously improved along with the accumulation of the reservoir operation data, the dependence on the accuracy of the boundary condition is low, and the problem of low calculation precision of the traditional hydrodynamic method can be effectively solved; in reservoir flood season scheduling, the key section water level can be accurately predicted, and technical support is provided for flood control and prosperity scheduling decisions of large-scale reservoirs or cascade reservoirs.
Description
Technical Field
The invention belongs to the field of reservoir operation control, and particularly relates to a reservoir water level prediction method based on a recurrent neural network and an attention mechanism.
Background
The forecasting calculation of the water level of the key section of the reservoir plays an important role in reservoir flood control dispatching. In the traditional method, under the condition of given warehousing flow and dam front water level, the water level and flow of all sections in a reservoir area are calculated by solving a hydrodynamics equation (saint-wien equation system), and the water level of a key section is contained in the water level. The method has the advantages that the method has complete mathematical theory support, can completely connect all hydraulic parameters together to form a whole by constructing a more detailed river network space distribution structure, is convenient to process and can adapt to various initial conditions; the method has the disadvantages that after the model is established, whether the model meets the requirements needs to be checked, particularly for the calibration and verification of the river course roughness of the sensitivity parameter, a large amount of labor, time and calculation cost are needed, and the requirement on the accuracy of the input boundary condition is high.
For a large reservoir, particularly in a flood season, the accuracy of a hydrodynamic model is not high, and the main reasons can be summarized as the following three points: (1) the warehousing flow of the reservoir is difficult to be accurately counted. For large reservoirs which extend for hundreds of kilometers, the influx flow of rainfall, torrential flood and the like in an interval is difficult to accurately calculate, so that the boundary condition of the model is inaccurate, which is a main cause of errors; and (2) systematic errors exist in the hydrodynamic model. The hydrodynamic model depends on the terrain of the river channel and the position of the section, measurement errors exist in the data, the length of the river channel of the large reservoir is large, the number of the sections is large, and the accumulated errors also influence the prediction effect of the water level; (3) The conditions for the establishment of the holy-weinan equation set may not be met during the flood season. The slow flow is a precondition established by the holy-south equation set, and the flow velocity is usually obviously increased and the applicability of the holy-south equation set is reduced in the rainstorm or large-flow process in the flood season.
Through research of ten years, the water level calculation of the reservoir area of the three gorges reservoir in the Yangtze river basin in China still has a calculation error of 1-3 meters for the water level of the important section long-life station, so that the reason for the calculation error of the hydrodynamic model is extremely difficult to overcome. In order to solve the technical problem, no effective solution is provided at present.
Disclosure of Invention
The invention has the technical problems that the existing method for calculating the key section water level of the large reservoir by solving the hydrodynamic equation is influenced by measurement and statistical errors of warehousing flow, river channel section and the like, the accuracy of the predicted and calculated key section water level is not high, and the reasons causing the errors are extremely difficult to overcome.
The invention aims to solve the problems and provides a reservoir water level prediction method based on a recurrent neural network and an attention mechanism, which is characterized in that knowledge is learned in historical data of reservoir operation, the reservoir inlet and outlet flow and the mapping relation between the water level before the dam and the critical section water level are searched, and the calculation precision of the critical section water level is improved by using historical information; the method is a data-driven method, and is based on a gradient descent optimization algorithm [1] And the correlation between the data is automatically learned, and compared with the traditional hydrodynamic method, the method reduces the dependence on the accuracy of the boundary conditions and only needs to keep the accuracy of the future boundary conditions and the accuracy of the historical boundary conditions basically stable. Meanwhile, the method has evolutionary capability, the calculation accuracy of the water level of the key section can be continuously improved along with the accumulation of historical data, and the problem that the calculation accuracy of the traditional hydrodynamic method is not high can be effectively solved.
The technical scheme of the invention is a reservoir water level prediction method based on a recurrent neural network and an attention mechanism, which comprises the following steps,
step 1: collecting historical data of the flow of the reservoir in and out of a reservoir and the water level of the key section before the dam to form a data set, and classifying the data set;
step 2: respectively transforming the water level and flow numerical values of each type of data set, mapping to an interval [0,1], obtaining a standard data set, and performing data segmentation on each type of standard data set according to the principle that 80% of the standard data set is a training set, 10% of the standard data set is a verification set and 10% of the standard data set is a test set;
and step 3: respectively constructing a recurrent neural network model with an attention mechanism aiming at each type of data set;
step 3.1: establishing a circulating neural network model, taking the flow of the reservoir in and out of the reservoir and the water level before the dam as the input of the neural network model, and taking the output of the neural network model as the predicted water level of the key section;
step 3.2: defining an optimization objective function based on a cyclic neural network model, wherein the optimization objective is that the error between the predicted water level and the actually measured water level is minimum;
and 4, step 4: selecting the training set training recurrent neural network model in the step 2, and solving the objective function in the step 3 to obtain optimized model parameters;
and 5: selecting the verification set in the step 2 to verify the cyclic neural network model, if the verified prediction effect is unqualified, adjusting the hyper-parameters of the model, re-training the model, and executing the step 4; if the verification is passed, executing step 6;
step 6: selecting the test set in the step 2 to test the recurrent neural network model, recording the test result and evaluating the model;
and 7: and taking the real-time flow rate of the inlet and the outlet and the water level before the dam as the input of the model to obtain the output of the model, namely the predicted water level of the key section.
Further, in step 1, the data sets are classified according to different warehousing flows and dam front water level levels according to the water regime.
Further, in step 3, the optimization objective function is a least square function.
Preferably, the optimization objective function includes a least square term and a regular term of the neural network parameters, and the regular term of the neural network parameters and the weight coefficients thereof are used for preventing the network parameters from being over-fitted.
Preferably, in step 4, the objective function in step 3 is solved, and the objective function is solved by using a gradient descent method. Each iteration of the optimization algorithm is performed by selecting a part of samples from the training set instead of the whole training set, so that the convergence speed of the model is increased, and the prediction effect is improved.
Preferably, the Recurrent neural network model uses GRU (Gated redundant Unit) computation nodes.
Compared with the prior art, the invention has the beneficial effects that: by introducing a data-driven thought and utilizing a recurrent neural network with a concern mechanism, the mapping relation between the flow of the reservoir in and out of the reservoir and the water level before the dam to the water level of the critical section is directly learned, the calculation precision of the model can be continuously improved along with the accumulation of the operation data of the reservoir, the dependence on the accuracy of the boundary condition is low, and the problem of low calculation precision of the traditional hydrodynamics method can be effectively solved. In the flood season scheduling of the reservoir, the water level of the critical section can be accurately predicted, and technical support is provided for flood control and prosperity scheduling decisions of a large reservoir or a cascade reservoir.
Drawings
The invention is further illustrated by the following figures and examples.
Fig. 1 is a schematic flow chart of a reservoir water level prediction method according to the present invention.
FIG. 2 is a data classification diagram of different water regimes according to an embodiment.
Fig. 3 is a schematic structural diagram of a recurrent neural network with a mechanism of interest according to an embodiment.
Fig. 4 is a comparison effect graph of the predicted water level and the measured water level of the three gorges reservoir 2009-2019 longevity station in the embodiment.
Fig. 5 is a comparison effect diagram of the predicted water level and the measured water level of the 2011 longevity station in the three gorges reservoir of the embodiment.
Fig. 6 is a comparison effect graph of the predicted water level and the measured water level of the 2012 longevity station of the three gorges reservoir according to the embodiment.
Fig. 7 is a comparison effect graph of the predicted water level and the measured water level of the longevity station in the three gorges reservoir 2014 according to the embodiment.
Fig. 8 is a comparison effect graph of the predicted water level and the measured water level of the 2017 longevity station of the three gorges reservoir in the embodiment.
Detailed Description
The reservoir water level prediction method based on the recurrent neural network and the attention mechanism is mainly applied to large reservoirs with certain operation histories. The three gorges reservoir is the reservoir with the largest reservoir capacity in the Yangtze river basin in China, has more than ten years of history in operation, has abundant accumulated data, has the distance between the upstream Zhu Tuo and the front dam of about 757km, is easy to submerge when large warehousing flow occurs in the flood season and the water storage season in the long-life region at the tail of the reservoir, and is a sensitive region for submerging the tail of the reservoir in the three gorges reservoir region. The embodiment selects the three gorges reservoir to predict the water level of the section of the long-life station of the three gorges reservoir.
As shown in fig. 1, the reservoir water level prediction method based on the recurrent neural network and the attention mechanism includes the following steps,
step 1: and collecting historical data of the flow of the reservoir in and out of the reservoir and the water level before the dam and the critical section. According to the historical operating condition of the reservoir, the flow of the reservoir entering and exiting the reservoir and the water level data of the key section in front of the dam in the past decade are collected to form a historical data set.
The embodiment collects the operation data from 2009, 1, 0 to 2019, 11, 18, 23 of the three gorges reservoir, including the smooth flow of the three gorges, the flow of the three gorges coming out of the reservoir, the flow of the cun station, the flow of the wulong station and the water level of the phoenix mountain, and collects 95376 pieces of data in a group of data per hour. The three gorges smooth flow, the cun-beach flow and the Wulong flow are all warehouse entry flow components, the smooth flow is taken as the main part, and the Fenghuangshan water level represents the dam front water level.
And (4) according to the water regime, manually classifying the collected data according to different warehousing flows and dam front water level grades, and respectively executing the steps 2 to 6 for each class.
The three gorges reservoir has large variation of warehousing flow and dam front water level in one year, firstly, water situations are classified according to different warehousing flow and dam front water level, and then a model and evaluation are respectively constructed for each class. FIG. 2 is a specific case of classification, where z represents the dam front water level in meters (m), q represents the three gorges smooth flow in cubic meters per second (m) 3 /s)。
Step 2: and mapping the water level and flow values to an interval [0,1] of each classified data set to obtain a standard data set, and performing data segmentation on each classified standard data set according to the principle that 80% of the standard data set is a training set, 10% of the standard data set is a verification set and 10% of the standard data set is a test set.
The way the data is normalized is calculated as follows:
standardized water level = (original water level-lowest water level)/(highest water level-lowest water level)
Normalized flow = (original flow-minimum flow)/(maximum flow-minimum flow)
And step 3: and (3) respectively constructing a recurrent neural network model with a concern mechanism aiming at each type of data set, taking the standardized flow rate of the reservoir in and out of the reservoir and the water level before the dam in a time sequence as input, and outputting the water level of the predicted key section. And defining a least square objective function based on the model, so that the predicted water level of the key section is the same as or similar to the actually measured water level.
And 4, step 4: inputting the standardized training set formed in the step 2 into the recurrent neural network model constructed in the step 3, and solving the objective function constructed in the step 3 by using a gradient descent method to obtain optimized model parameters.
Examples the gradient descent method used is described in Liu Yingchao, zhang Jiyuan paper "gradient descent method" published by university of Nanjing university of science.
In the embodiment, a TensorFlow open source software package implementation model is used, and an Adam optimizer in the TensorFlow open source software package implementation model is used for solving the objective function in the step 3, wherein the Adam optimizer is one implementation of a gradient descent method. TensorFlow software is disclosed in the article "TensorFlow: learning functions at scale" Abadi Martini published by Acm Sigplan notes 5, 2016.
And 5: verifying the model on the standard verification set formed in the step 2 according to the model parameters obtained in the step 4, if the predicted effect of verification cannot be expected, adjusting the hyper-parameters in the model, and returning to the step 4 to retrain the model; if the verification set passes the verification, step 6 is executed.
Step 6: and (3) selecting the test set in the step (2) to test the recurrent neural network model, recording the test result and evaluating the model.
And 7: and taking the real-time flow of the inlet and the outlet and the water level before the dam as the input of the model to obtain the water level of the key section predicted by the model.
In step 3, the constructed three-layer recurrent neural network model with the attention mechanism is shown in fig. 3, wherein the bottom part is the recurrent neural network, and the top part is the attention mechanism. The input of the whole model is the reservoir outlet and inlet flow and the dam front water level at the first n moments, the output is the long-life station water level at the (n + 1) th moment, and all the input and output are standardized data. In the embodiment, n is 2, and the time interval between two moments is 6 hours. W, U, b in fig. 3 are network parameters; x is the number of t Is t atThe components of the carved input vector comprise three gorges smooth flow, warehouse-out flow, cun-beach flow, wulong flow and dam front water level; h is the hidden state of the corresponding node. The calculation sequence of the method follows the time sequence, flow and water level information is gradually synthesized from left to right and from bottom to top, wherein each rectangle represents a synthesis calculation node, the commonly used calculation nodes comprise a traditional calculation node, an LSTM (Long Short-Term Memory) calculation node and a GRU (Gated Current Unit) calculation node, the calculation modes of different calculation nodes are shown in Table 1, wherein sigma is a sigmoid function, tanh is a tanh function, and the calculation symbols multiplied by corresponding elements of a vector are adopted, and the GRU calculation node is adopted in the embodiment.
TABLE 1 calculation of different nodes
For the focus mechanism part, output
The method is characterized by comprising the following steps of linearly weighting hidden states of a third layer of a recurrent neural network, wherein weight coefficients of the hidden states are generated through a layer of feedforward neural network, and the specific calculation mode is as follows:
α 1 ,α 2 …α n-1 ,α n are respectively the weight coefficient, beta, of each node of the third layer j For the hidden state of each node of the third layer, W α 、b α 、U α Are all parameters of the feedforward neural network, and e is a natural base number.
Based on the output of the model, an objective function is constructed:
wherein y is i Is the actual water level of the water tank,
the prediction result of the model is obtained, and N is the number of samples; the first term in the target function is a least square term, so that the predicted water level is the same as or similar to the actual water level; the second term is a regular term, lambda is a weight coefficient, and the term is used for limiting the value range of the parameters and preventing the overfitting of the model.
In the examples, the operation data of the three gorges reservoir in 2009-2018 was used for training and verification, and the operation data of 2019 was used for testing. Table 2 shows the distribution of error intervals between the predicted water levels of the training set, the validation set, and the test set in the data sets of different categories according to the example, where z represents the water level before the dam, and q represents the smooth flow of the three gorges. Due to the regulation and storage function of the three gorges reservoir, the annual variation of the water level of the long-life station is about 30m, the maximum prediction error in the method is-1.27 m, and the error rate is only-4.2%; the maximum prediction error in 2019 is-0.69 m, the error rate is-2.3%, and the method has obvious advantages compared with the error of 1-3m in the traditional hydrodynamics method.
TABLE 2 table of predicted water level error intervals for various data sets of the example
| Class interval | Training set | Verification set | Test set |
| z∈[173.51,175.16] | [-0.36m,0.26m] | [-0.26m,0.24m] | [-0.15m,0.26m] |
| z∈[170,173.5] | [-0.25m,0.24m] | [-0.21m,0.23m] | [-0.27m,0.26m] |
| z∈[144.77,169.99],q∈[25000,71200] | [-1.27m,1.11m] | [-1.19m,1.08m] | [-0.69m,0.39m] |
| z∈[165.01,169.99],q∈[995,24999] | [-0.31m,0.26m] | [-0.28m,0.26m] | [-0.21m,0.21m] |
| z∈[160.01,165],q∈[995,24999] | [-0.35m,0.59m] | [-0.32m,0.51m] | [-0.31m,0.20m] |
| z∈[150.01,160],q∈[995,24999] | [-0.90m,0.88m] | [-0.62m,0.52m] | [-0.46m,0.37m] |
| z∈[144.77,150],q∈[995,24999] | [-0.96m,1.01m] | [-0.98m,1.02m] | [-0.39m,0.48m] |
Fig. 4 is a comparison graph of the predicted water level and the measured water level from 1 month 1 in 2009 to 18 months in 2019, wherein the predicted water level curve and the measured water level curve mostly coincide with each other. The elevations of the land utilization line and the immigration migration line of the three gorges reservoir long-life station are respectively 175.7m and 178m, and the water level exceeds the land utilization line in 2011, 2012, 2014 and 2017. Fig. 5 to 8 show water level prediction conditions in 2011, 2012, 2014 and 2017 respectively, and it can be seen that when the water level exceeds the land utilization line, the prediction accuracy is very high, so that the method of the present invention can provide an accurate water level prediction value for real-time scheduling of the three gorges reservoir.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (6)
1. The reservoir water level prediction method based on the recurrent neural network and the attention mechanism is characterized by comprising the following steps,
step 1: collecting historical data of the flow of the reservoir in and out of a reservoir and the water level of the key section before the dam to form a data set, and classifying the data set;
step 2: respectively transforming the water level and flow values of each type of data set to obtain standard data sets, and dividing each type of standard data sets into a training set, a verification set and a test set;
and 3, step 3: aiming at each type of data set, respectively constructing a recurrent neural network model with an attention mechanism, and defining a target function;
step 3.1: establishing a circulating neural network model, taking the flow of the reservoir in and out of the reservoir and the water level before the dam as the input of the neural network model, and taking the output of the neural network model as the predicted key section water level;
step 3.2: defining an optimization objective function based on the recurrent neural network model, wherein the optimization objective is that the error between the predicted water level and the actually measured water level is minimum;
and 4, step 4: selecting the training set training recurrent neural network model in the step 2, and solving the objective function in the step 3 to obtain optimized model parameters;
and 5: selecting the verification set in the step 2 to verify the cyclic neural network model, if the verified prediction effect is unqualified, adjusting the hyper-parameters of the model, re-training the model, and executing the step 4; if the verification is passed, executing the step 6;
step 6: selecting the test set in the step 2 to test the recurrent neural network model, recording the test result and evaluating the model;
and 7: and taking the real-time flow rate of the inlet and the outlet and the water level before the dam as the input of the model to obtain the output of the model, namely the predicted water level of the key section.
2. The reservoir water level prediction method based on the recurrent neural network and the attention mechanism as claimed in claim 1, wherein in step 1, the data sets are classified according to different flow rates and dam front water level levels according to water conditions.
3. The reservoir water level prediction method based on the recurrent neural network and the attention mechanism according to claim 1, wherein in step 3, the optimization objective function is a least square function.
4. The recurrent neural network and attention mechanism-based reservoir water level prediction method of claim 3, wherein the optimization objective function comprises a least square term and a regular term of neural network parameters, and the regular term of neural network parameters and its weight coefficients are used to prevent the network parameters from being over-fitted.
5. The reservoir water level prediction method based on the recurrent neural network and the attention mechanism as claimed in claim 1, wherein in step 4, the objective function of step 3 is solved, and the objective function is solved by a gradient descent method.
6. The recurrent neural network and attention mechanism-based reservoir water level prediction method according to any one of claims 1 to 5, wherein the recurrent neural network model employs GRU calculation nodes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010312062.1A CN111553394B (en) | 2020-04-20 | 2020-04-20 | Reservoir water level prediction method based on cyclic neural network and attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010312062.1A CN111553394B (en) | 2020-04-20 | 2020-04-20 | Reservoir water level prediction method based on cyclic neural network and attention mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111553394A CN111553394A (en) | 2020-08-18 |
| CN111553394B true CN111553394B (en) | 2023-02-07 |
Family
ID=72000199
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010312062.1A Active CN111553394B (en) | 2020-04-20 | 2020-04-20 | Reservoir water level prediction method based on cyclic neural network and attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111553394B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112131989B (en) * | 2020-09-15 | 2021-07-23 | 河海大学 | Millimeter wave rain measurement model parameter obtaining method based on space rainfall data |
| CN112541839B (en) * | 2020-12-23 | 2022-02-11 | 国能大渡河大数据服务有限公司 | Reservoir storage flow prediction method based on neural differential equation |
| CN114169047B (en) * | 2021-11-23 | 2024-01-23 | 中国长江三峡集团有限公司 | Method for calculating reservoir backwater range |
| CN114611778B (en) * | 2022-03-04 | 2022-09-06 | 山东锋士信息技术有限公司 | Reservoir water level early warning method and system based on warehousing flow |
| CN117689501B (en) * | 2024-02-03 | 2024-05-14 | 深圳市广汇源环境水务有限公司 | Reservoir flood prevention dynamic forecasting and early warning method and system platform |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108875161B (en) * | 2018-05-31 | 2022-11-29 | 长江勘测规划设计研究有限责任公司 | Traffic grade prediction method based on convolutional neural network deep learning |
| CN109799533B (en) * | 2018-12-28 | 2021-07-27 | 中国石油化工股份有限公司 | Reservoir prediction method based on bidirectional circulation neural network |
| CN110085327A (en) * | 2019-04-01 | 2019-08-02 | 东莞理工学院 | Multichannel LSTM neural network Influenza epidemic situation prediction technique based on attention mechanism |
| CN110175416A (en) * | 2019-05-30 | 2019-08-27 | 三峡大学 | Three Gorges Reservoir water temperature prediction method based on principal component analysis and neural network |
-
2020
- 2020-04-20 CN CN202010312062.1A patent/CN111553394B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111553394A (en) | 2020-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111553394B (en) | Reservoir water level prediction method based on cyclic neural network and attention mechanism | |
| CN103729550B (en) | Multiple-model integration Flood Forecasting Method based on propagation time cluster analysis | |
| Shiri et al. | Global cross-station assessment of neuro-fuzzy models for estimating daily reference evapotranspiration | |
| CN111369070B (en) | Multimode fusion photovoltaic power prediction method based on envelope clustering | |
| CN111665575B (en) | Medium-and-long-term rainfall grading coupling forecasting method and system based on statistical power | |
| CN107463730A (en) | A kind of streamflow change attribution recognition methods for considering Spatio-temporal Evolution of Land Use | |
| CN105139093A (en) | Method for forecasting flood based on Boosting algorithm and support vector machine | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| CN109711617A (en) | A kind of medium-term and long-term Runoff Forecast method based on BLSTM deep learning | |
| CN109143408B (en) | Dynamic region combined short-time rainfall forecasting method based on MLP | |
| CN110969282A (en) | Runoff stability prediction method based on LSTM composite network | |
| CN113592144A (en) | Medium-and-long-term runoff probability forecasting method and system | |
| CN112396152A (en) | Flood forecasting method based on CS-LSTM | |
| CN115640881A (en) | Reservoir water level correction method of reservoir warehousing runoff forecasting model based on LSTM | |
| CN114492922A (en) | Medium-and-long-term power generation capacity prediction method | |
| CN110765418A (en) | Intelligent set evaluation method and system for basin water and sand research model | |
| CN113050430A (en) | Drainage system control method based on robust reinforcement learning | |
| Xiao et al. | Runoff forecasting using machine-learning methods: case study in the middle reaches of Xijiang River | |
| Agarwal et al. | Flood forecasting and flood flow modeling in a river system using ANN | |
| Wei | Comparing single-and two-segment statistical models with a conceptual rainfall-runoff model for river streamflow prediction during typhoons | |
| CN117114190A (en) | River runoff prediction method and device based on mixed deep learning | |
| CN109919362B (en) | Medium-and-long-term runoff forecasting method considering hydraulic engineering scheduling influence | |
| Nizar et al. | Forecasting of temperature by using LSTM and bidirectional LSTM approach: case study in Semarang, Indonesia | |
| Chang et al. | Development of a real-time forecasting model for turbidity current arrival time to improve reservoir desilting operation | |
| CN110110944A (en) | One kind predicting wind electricity digestion demand quantity algorithm based on modified wavelet neural network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |