CN115759403B - Dynamic combination prediction model construction method for water circulation process in cold region - Google Patents
Dynamic combination prediction model construction method for water circulation process in cold region Download PDFInfo
- Publication number
- CN115759403B CN115759403B CN202211442607.6A CN202211442607A CN115759403B CN 115759403 B CN115759403 B CN 115759403B CN 202211442607 A CN202211442607 A CN 202211442607A CN 115759403 B CN115759403 B CN 115759403B
- Authority
- CN
- China
- Prior art keywords
- cold region
- water circulation
- constructing
- time
- model
- 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 69
- 238000000034 method Methods 0.000 title claims abstract description 54
- 230000008569 process Effects 0.000 title claims abstract description 28
- 238000010276 construction Methods 0.000 title description 2
- 238000004364 calculation method Methods 0.000 claims abstract description 10
- 238000001556 precipitation Methods 0.000 claims abstract description 10
- 230000006403 short-term memory Effects 0.000 claims abstract description 10
- 230000007787 long-term memory Effects 0.000 claims abstract description 4
- 239000002689 soil Substances 0.000 claims description 23
- 230000015654 memory Effects 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 12
- 230000008018 melting Effects 0.000 claims description 9
- 238000002844 melting Methods 0.000 claims description 9
- 230000005855 radiation Effects 0.000 claims description 5
- 230000000717 retained effect Effects 0.000 claims description 5
- 230000008595 infiltration Effects 0.000 claims description 4
- 238000001764 infiltration Methods 0.000 claims description 4
- 230000035699 permeability Effects 0.000 claims description 4
- 239000005413 snowmelt Substances 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 claims description 3
- 230000001364 causal effect Effects 0.000 claims description 3
- 230000003068 static effect Effects 0.000 claims description 3
- 230000008022 sublimation Effects 0.000 claims description 3
- 238000000859 sublimation Methods 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 238000004148 unit process Methods 0.000 claims description 2
- 230000008859 change Effects 0.000 abstract description 5
- 230000008014 freezing Effects 0.000 description 3
- 238000007710 freezing Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 2
- 238000010257 thawing Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 102100024873 Ceramide-1-phosphate transfer protein Human genes 0.000 description 1
- 101710189399 Ceramide-1-phosphate transfer protein Proteins 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000008020 evaporation Effects 0.000 description 1
- 238000001704 evaporation Methods 0.000 description 1
- 239000003337 fertilizer Substances 0.000 description 1
- 239000005457 ice water Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
Abstract
The invention discloses a method for constructing a dynamic combination prediction model aiming at a water circulation process in a cold region. The method comprises the following steps: s1: collecting basic data of relevant factors of water circulation in a cold region, and dividing the basic data into a plurality of limiting factor groups and a plurality of limiting factor groups according to influence limiting conditions; s2: constructing a dynamic Bayesian network to predict the value of the multi-limiting factor group at the end of the period; s3: constructing a long-term and short-term memory network model to predict the value of a few limiting factor group at the end of a period; s4: constructing an influence relation among factors related to water circulation in a cold region, and calculating influence probability among the factors; s5: constructing a cold region precipitation model; s6: and constructing a cold region water circulation model. The method and the device define random uncertainty of climate change on cold region water circulation prediction and correlation among factors, and perform autoregressive combined prediction based on the factor-limited condition, so that the calculation accuracy and the application range of the model method are improved.
Description
Technical Field
The invention belongs to the technical field of agricultural water management, and particularly relates to a method for constructing a dynamic combination prediction model for a water circulation process in a cold region.
Background
The hydrology in cold region is an important component of hydrology science, is a special branch of hydrology, and has the dual characteristics of freezing circle science and hydrology. The natural environment in the cold region is severe, mostly the natural environment is a natural region, and the uncertainty of the observation of the characteristic process in the cold region is increased, so that the deficiency and the lack of the observation data are caused. The water circulation in cold regions is mainly affected by glacier frozen soil, and the response of the cold regions existing in glacier frozen soil to climate change and human activities is different from that of non-cold regions. Compared with the non-cold region hydrologic factors, the cold region hydrologic factors have great difference, ice-water transformation is the biggest characteristic, and solid-liquid transformation is the basic process of the cold region hydrologic, so the cold region hydrologic process is more influenced by factors such as environment. The special thawing rule of glacier snow and the soil water redistribution caused by the soil freezing and thawing process affect the water circulation processes of infiltration, evaporation, precipitation, runoff, soil water and the like in cold areas to different degrees. Soil freezing and river ice sealing are the biggest characteristics of cold region water circulation different from non-cold region, and understanding cold region water circulation process has important meaning to frozen soil and water resource evolution law research.
For the hydrologic cycle model of the cold region, students at home and abroad do many researches, including direct application of the hydrologic model of the non-cold region in the cold region after debugging, application of the hydrologic model of the non-cold region in the cold region after improvement, hydrologic process simulation of the cold region based on a modularized modeling method, development of a land process model of the cold region and the like. However, in the process of predicting the water circulation in the cold region, the interaction relationship among the influencing factors is not fully considered, and particularly, the close dynamic response existing between each factor and time is ignored. Among the factors of the water circulation in the cold region, some factors are influenced by various conditions, and the conditions that some factors are limited are fewer, but the influence of the problem on the model accuracy is rarely considered in the existing prediction model. And the water circulation in the cold region caused by climate change causes more random uncertainty in the water circulation in the cold region, and the calculation accuracy and the application range of the existing model are limited. Therefore, it is needed to comprehensively consider the prediction of the water circulation in the cold region and develop a method for constructing a dynamic combined prediction model aiming at the water circulation process in the cold region.
Disclosure of Invention
The invention aims to solve the problems in the prior art, and provides a method for constructing a dynamic combined prediction model for a water circulation process in a cold region, which considers the random uncertainty of climate change on the water circulation prediction in the cold region and the correlation among factors, and performs autoregressive combined prediction based on the limited conditions of the factors, so that the calculation accuracy and the application range of the model method are improved.
The invention has the following advantages: (1) taking the influence relation among all factors of the water circulation in the cold region into consideration, and innovatively constructing a prediction model aiming at the water circulation elements in the cold region; (2) fully considers the influence of time dynamics on the model method; (3) aiming at different limiting conditions of all factors of the water circulation in the cold region, an autoregressive combined prediction method is constructed, and the model prediction precision is improved; (4) the random uncertainty of future climate change on the water circulation process in the cold region is considered, and the calculation accuracy and the application range of the model are improved.
The technical scheme of the invention is realized as follows: a method for constructing a dynamic combination prediction model aiming at a water circulation process in a cold region is characterized by comprising the following steps:
s1: collecting basic data of relevant factors of water circulation in a cold region, and dividing the basic data into a plurality of limiting factor groups and a plurality of limiting factor groups according to influence limiting conditions;
s2: constructing a dynamic Bayesian network to predict the value of the multi-limiting factor group at the end of the period;
s3: constructing a long-term and short-term memory network model to predict the value of a few limiting factor group at the end of a period;
s4: constructing an influence relation among factors related to water circulation in a cold region, and calculating influence probability among the factors;
s5: when the average air temperature is lower than the critical temperature of snowfall, defining the rainfall as snowfall, accumulating the snowfall into the accumulated snow, increasing the accumulated snow along with the snowfall, and melting and sublimating along with the accumulated snow to reduce, wherein a mass conservation equation is as follows:
S day ＝S 0 +R day -E sub -S mlt
wherein S is day Snow water equivalent in a certain day, mm; s is S 0 Initial snow water equivalent (product of snow density and snow depth) for a certain day, mm; r is R day The snowfall water equivalent weight of a certain day when the average air temperature is lower than the snowfall critical temperature is mm; e (E) sub Snow sublimation amount in a certain day, and mm; s is S mlt Snow melt equivalent weight of a day, mm;
s6: constructing a cold region water circulation model, and calculating the soil water content at the end of a period of time:
in which W is t Soil moisture content at the end of the period of time, mm; w (W) 0 Soil moisture content at the beginning of the period of time, mm; days of t period, d; s is S day Precipitation amount on day i, mm; q (Q) surf Surface runoff on day i, mm; e (E) a The amount of evapotranspiration on day i, mm; q (Q) gw The amount of return water on day i, mm.
Preferably, the base data in S1 includes: initial soil moisture content, precipitation, snow fall, surface runoff, evapotranspiration, infiltration, maximum temperature, minimum temperature, relative humidity, wind scale high wind speed, solar hours, wind scale height, snow density, snow depth, temperature, snow melting rate, porosity, volume weight, permeability, solar radiation related factors.
Preferably, the multiple limiting factor set in S1 includes: initial soil water content, surface runoff, evapotranspiration, snow density and snow melting rate; the few limiting factor group includes: rainfall, snowfall, temperature.
Preferably, the dynamic bayesian network described in S2 is defined as (B, 2 TBN), where B is BN, which defines the a priori distribution P (X1), and 2TBN is a BN comprising two time slices:
in the method, in the process of the invention,is a node on time slice t, +.>For the parent node set on time slice t or t-1, the nodes in the first time slice of 2TBN have no relevant parameters, the nodes of the continuous variable in the second time slice have a relevant conditional probability distribution, and each discrete variable has a conditional probability table defined as +.>The arcs between the time slices are from left to right reflecting the flow of time; the arc within the time slice is arbitrary; directional arcs within the time slices represent "instantaneous" causal relationships; the semantics of a dynamic bayesian network can be defined by "unrolling" the 2TBN time slices until there are T time slices; the resulting joint distribution is:
each time slice in the dynamic Bayesian network is a static Bayesian network, which consists of state variables X i And evidence variable E i Composition; at time slice t, X i By having a limited number of statesNode n of (2) (i,t) E N represents; />Representing the probability distribution of these states at time slices t, which are made up of a number of node sets N 0 ,…,N t A representation; n (N) t Including all random variables associated with time slice t;
the arcs connecting two variables belonging to different time slices represent the time probability dependencies between these variables; dynamic shellfishThe leaf network can model random variables and can represent the influence of the random variables on future distribution of other variables, and the influence is defined as transition probability between the variable state of the time slice t-1 and the variable state of the time slice t; in the model, the next time slice t is conditionally independent with respect to the current time slice t-1, i.e. CPT P (n i,t |pa(n i,t ) A markov characteristic;
furthermore, the conditional probability is equivalent to a markov model if the following conditions are satisfied:
Pa(n i,t )＝n i,t-1 andP(n i,t |n i,t-1 )＝P MC
starting from the observation situation when the time slice t=0, the calculation is performed by dynamic Bayesian network reasoningProbability distribution over states; for calculating->The Bayesian network form is kept compact by an iterative reasoning method, and the time concept is introduced into the Bayesian network by utilizing the reasoning form; in practice, an arbitrary variable X can be calculated based on the probability corresponding to the time slice t-1 alone i The probability distribution at t, the probability distribution of time slice t-1, may be obtained by successive inference calculations.
Preferably, the short-term memory network model in S3 includes three gates for controlling the states of the units, and the first step in the operation of the short-term memory network model is to determine which information needs to be discarded for the cell states, and the operation is processed by a sigmoid unit called "forgetting gate"; the sigmoid unit processes the information by looking at h t-1 And x t The information outputs a vector between 0 and 1, the 0-1 values inside the vector representing which C's in the cell state t-1 The information is retained, or discarded; which is a kind ofWherein 0 indicates that no information is reserved, 1 indicates that both are reserved, and a sigmoid function and a tanh function are used as activation functions respectively;
an input door:
I t ＝σ(W i [h t-1 ,x t ]+b i )
in which W is i For inputting a gate weight matrix, sigma is a Sigmoid function, h t-1 For the output data of the memory cell at the previous time, x t B for input at the current time i Bias for the input gate;
forgetting the door:
f t ＝σ(W f [h t-1 ,x t ]+b f )
in which W is f For forgetting the gate weight matrix, b f Bias for forgetting the door;
output door:
O t ＝σ(W o [h t-1 ,x t ]+b o )
in which W is o To output the gate weight matrix, b o Offset for the output gate;
a memory unit:
in which W is C B is a memory cell weight matrix C Is the bias of the memory cell;
a memory unit c for updating the candidate value:
wherein C is t-1 The candidate value at the previous moment is referred;
the memory cell outputs:
h t ＝o t *tanh(C t )。
preferably, the short-term memory network model described in S3 is optimally trained by a small batch gradient descent algorithm Jing Xiang, with the mean square error function mse as the loss function:
wherein y is i Is the predicted value of the ith moment; y is Y i Is the true value at the i-th moment; n is the number of samples input by the model.
Preferably, the influence relationship between the cold region water circulation related factors in S4 may be established through a dynamic bayesian network.
Drawings
FIG. 1 is a schematic flow diagram of a method for constructing a dynamic combination prediction model for a water circulation process in a cold region;
Detailed Description
A method for constructing a dynamic combination prediction model aiming at a water circulation process in a cold region is characterized by comprising the following steps:
s1: collecting basic data of relevant factors of water circulation in a cold region, and dividing the basic data into a plurality of limiting factor groups and a plurality of limiting factor groups according to influence limiting conditions;
the base data includes: initial soil water content, precipitation, snow fall, surface runoff, evapotranspiration, infiltration, highest temperature, lowest temperature, relative humidity, wind scale high wind speed, sunshine hours, wind scale height, snow density, snow depth, temperature, snow melting rate, porosity, volume weight, permeability and solar radiation related factors;
the multiple limiting factor set includes: initial soil water content, surface runoff, evapotranspiration, snow density and snow melting rate; the few limiting factor group includes: rainfall, snowfall, temperature.
Limiting factors for soil moisture content include: precipitation, porosity, volume weight, permeability, applied fertilizer, etc.; surface runoff limiting factors include: precipitation, solar radiation related factors, air temperature, wind speed, pressure, relative humidity, drainage basin area, etc.; the evapotranspiration limiting factors include: maximum temperature, minimum temperature, relative humidity, wind vane high wind speed, sunshine hours and vane height; snow density limiting factors include: temperature, snowfall area, wind speed, etc.; snow melt rate limiting factors include: solar radiation related factors, temperature, wind speed, etc. The rainfall is mainly affected by sea and land position, topography, barometric zone, wind zone, monsoon, cyclone, frontal surface, underlying surface, ocean current, etc., and when the research area is determined, the law of rainfall and snowfall is greatly affected by time, and is less affected by factors except the region and time.
S2: constructing a dynamic Bayesian network to predict the value of the multi-limiting factor group at the end of the period;
a dynamic bayesian network is defined as (B, 2 TBN), where B1 is BN, which defines the a priori distribution P (X1), and 2TBN is a BN comprising two time slices.
In the method, in the process of the invention,is a node on time slice t, +.>For the parent node set on time slice t or t-1, the nodes in the first time slice of the 2TBN have no relevant parameters, but the nodes of the continuous variable in the second slice have an associated Conditional Probability Distribution (CPD), each discrete variable has a Conditional Probability Table (CPT) defined asThe arcs between time slices are left to right reflecting the flow of time. The arc within the slice is arbitrary. Directed arcs within a time slice represent "instantaneous" causal relationships. The semantics of a dynamic bayesian network can be defined by "unrolling" the 2TBN slots until there are T slots. The resulting joint distribution is:
each time slice in the dynamic Bayesian network is a static Bayesian network, and is composed of state variables X i And evidence variable E i Composition is prepared. At time slice t, X i By having a limited number of statesNode n of (2) (i,t) E N. />Representing the probability distribution of these states at time slices t, these time slices being defined by a number of node sets N 0 ,…,N t And (3) representing. N (N) t Including all random variables associated with time slice t.
The arcs connecting two variables belonging to different time slices represent the time probability dependencies between these variables. The dynamic bayesian network not only can construct a model of random variables, but also can represent the influence of the random variables on future distribution of other variables, and the influence is defined as transition probability between the variable state of the time slice t-1 and the variable state of the time slice t. In this model, the next time slice t is conditionally independent with respect to the current time slice t-1, i.e. CPTP (n) i,t |pa(n i,t ) A markov characteristic is followed. Furthermore, if present:
Pa(n i,t )＝n i,t-1 andP(n i,t |n i,t-1 )＝P MC
this conditional probability is equivalent to a markov model. Starting from the observation situation when the time slice t=0, the calculation is performed by dynamic Bayesian network reasoningProbability distribution over states. Is thatCalculate->The Bayesian network form is kept compact by an iterative reasoning method, and the time concept is introduced into the Bayesian network by utilizing the reasoning form. In practice, an arbitrary variable X can be calculated based on the probability corresponding to the time slice t-1 alone i The probability distribution at t, the probability distribution of time slice t-1, may be obtained by successive inference calculations.
S3: constructing a long-term and short-term memory network model to predict the value of a few limiting factor group at the end of a period;
and (3) adopting a long-short-term memory network model (LSTM) as a prediction model combining factors with few influence on the limiting conditions, and adopting a small-batch gradient descent algorithm to perform optimization training on the LSTM model. The mean square error function mse is used as a loss function:
wherein y is i Is the predicted value of the ith moment; y is Y i Is the true value at the i-th moment; n is the number of samples input by the model.
LSTMs can be used for selectively forgetting or retaining past information and judging long information, so that LSTMs can well process the condition of large data volume. The core of LSTMs is the cell state, represented by the horizontal line through the cell. LSTM networks are capable of deleting or adding information of cell states through a structure called a gate. The LSTM contains three gates inside that control the state of the cells. The first step of LSTM is to decide which information the cell state needs to discard. This part of the operation is handled by a sigmoid unit called a "forget gate". It is by looking at h t-1 And x t The information outputs a vector between 0 and 1, the 0-1 value inside this vector representing which C's in the cell state t-1 Information is retained or how much information is discarded. 0 means that no information is retained, 1 means that both are retained, using the sigmoid function and the tanh function as activation functions.
An input door:
I t ＝σ(W i [h t-1 ,x t ]+b i )
in which W is i For inputting a gate weight matrix, sigma is a Sigmoid function, h t-1 For the output data of the memory cell at the previous time, x t B for input at the current time i Is the bias of the input gate.
Forgetting the door:
f t ＝σ(W f [h t-1 ,x t ]+b f )
in which W is f For forgetting the gate weight matrix, b f Is the bias of the forgetting gate.
Output door:
O t ＝σ(W o [h t-1 ,x t ]+b o )
in which W is o To output the gate weight matrix, b o To output the gate bias.
A memory unit:
in which W is C B is a memory cell weight matrix C Is the bias of the memory cell.
A memory unit c for updating the candidate value:
wherein C is t-1 Refers to the candidate value at the previous time.
The memory cell outputs:
h t ＝o t *tanh(C t )
s4: constructing an influence relation among factors related to water circulation in a cold region, and calculating influence probability among the factors;
and (3) predicting influence factors of the water circulation in the cold region through the steps S2 and S3, inputting predicted data into the dynamic Bayesian network again, and exploring influence relations among the relevant factors of the water circulation in the cold region to obtain influence probabilities among the factors.
S5: when the average air temperature is lower than the critical temperature of snowfall, defining the rainfall as snowfall, accumulating the snowfall into the accumulated snow, increasing the accumulated snow along with the snowfall, and melting and sublimating along with the accumulated snow to reduce, wherein a mass conservation equation is as follows:
S day ＝S 0 +R day -E sub -S mlt
wherein S is day Snow water equivalent in a certain day, mm; s is S 0 Initial snow water equivalent (product of snow density and snow depth) for a certain day, mm; r is R day The snowfall water equivalent weight of a certain day when the average air temperature is lower than the snowfall critical temperature is mm; e (E) sub Snow sublimation amount in a certain day, and mm; s is S mlt Snow melt equivalent weight of a day, mm;
s6: constructing a cold region water circulation model, and calculating the soil water content at the end of a period of time:
in which W is t Soil moisture content at the end of the period of time, mm; w (W) 0 Soil moisture content at the beginning of the period of time, mm; days of t period, d; s is S day Precipitation amount on day i, mm; q (Q) surf Surface runoff on day i, mm; e (E) a The amount of evapotranspiration on day i, mm; q (Q) gw The amount of return water on day i, mm.
Claims (6)
1. A method for constructing a dynamic combination prediction model aiming at a water circulation process in a cold region is characterized by comprising the following steps: s1: collecting basic data of relevant factors of water circulation in a cold region, and dividing the basic data into a plurality of limiting factor groups and a plurality of limiting factor groups according to influence limiting conditions; the multiple limiting factor set includes: initial soil water content, surface runoff, evapotranspiration, snow density and snow melting rate; the few limiting factor group includes: rainfall, snowfall, and temperature;
s2: constructing a dynamic Bayesian network to predict the value of the multi-limiting factor group at the end of the period;
s3: constructing a long-term and short-term memory network model to predict the value of a few limiting factor group at the end of a period;
s4: constructing an influence relation among factors related to water circulation in a cold region, and calculating influence probability among the factors;
s5: when the average air temperature is lower than the critical temperature of snowfall, defining the rainfall as snowfall, accumulating the snowfall into the accumulated snow, increasing the accumulated snow along with the snowfall, and melting and sublimating along with the accumulated snow to reduce, wherein a mass conservation equation is as follows:
S day ＝S 0 +R day -E sub -S mlt
wherein S is day Snow water equivalent in a certain day, mm; s is S 0 Initial snow water equivalent of a certain day, mm; r is R day The snowfall water equivalent weight of a certain day when the average air temperature is lower than the snowfall critical temperature is mm; e (E) sub Snow sublimation amount in a certain day, and mm; s is S mlt Snow melt equivalent weight of a day, mm;
s6: constructing a cold region water circulation model, and calculating the soil water content at the end of a period of time:
in which W is t Soil moisture content at the end of the period of time, mm; w (W) 0 Soil moisture content at the beginning of the period of time, mm; days of t period, d; s is S day Precipitation amount on day i, mm; q (Q) surf Surface runoff on day i, mm; e (E) a The amount of evapotranspiration on day i, mm; q (Q) gw The amount of return water on day i, mm.
2. The method for constructing the dynamic combination prediction model for the water circulation process in the cold region according to claim 1, which is characterized by comprising the following steps: the basic data in S1 includes: initial soil moisture content, precipitation, snow fall, surface runoff, evapotranspiration, infiltration, maximum temperature, minimum temperature, relative humidity, wind scale high wind speed, solar hours, wind scale height, snow density, snow depth, temperature, snow melting rate, porosity, volume weight, permeability, solar radiation related factors.
3. The method for constructing the dynamic combination prediction model for the water circulation process in the cold region according to claim 1, which is characterized by comprising the following steps: the dynamic bayesian network described in S2 is defined as (B, 2 TBN), where B is BN, which defines the a priori distribution P (X1), and 2TBN is a BN comprising two time slices:
in the method, in the process of the invention,is a node on time slice t, +.>For the parent node set on time slice t or t-1, the nodes in the first time slice of 2TBN have no relevant parameters, the nodes of the continuous variable in the second time slice have a relevant conditional probability distribution, and each discrete variable has a conditional probability table defined as +.>The arcs between the time slices are from left to right reflecting the flow of time; the arc within the time slice is arbitrary; directional arcs within the time slices represent "instantaneous" causal relationships; the semantics of a dynamic bayesian network can be defined by "unrolling" the 2TBN time slices until there are T time slices; the resulting joint distribution is:
each time slice in the dynamic Bayesian network is a static Bayesian networkCollaterals, defined by state variable X i And evidence variable E i Composition; at time slice t, X i By having a limited number of statesNode n of (2) (i,t) E N represents; />Representing the probability distribution of these states at time slices t, which are made up of a number of node sets N 0 ,…,N t A representation; n (N) t Including all random variables associated with time slice t;
the arcs connecting two variables belonging to different time slices represent the time probability dependencies between these variables; the dynamic Bayesian network can model the random variable and can represent the influence of the random variable on the future distribution of other variables, and the influence is defined as the transition probability between the variable state of the time slice t-1 and the variable state of the time slice t; in the model, the next time slice t is conditionally independent with respect to the current time slice t-1, i.e. CPT P (n i,t |pa(n i,t ) A markov characteristic;
furthermore, the conditional probability is equivalent to a markov model if the following conditions are satisfied:
starting from the observation situation when the time slice t=0, the calculation is performed by dynamic Bayesian network reasoningProbability distribution over states; for calculating->The Bayesian network form is kept compact by an iterative reasoning method, and the time concept is introduced into the Bayesian network by utilizing the reasoning form; in practice, onlyAccording to the probability corresponding to the time slice t-1, any variable X can be calculated i The probability distribution at t, the probability distribution of time slice t-1, may be obtained by successive inference calculations.
4. The method for constructing the dynamic combination prediction model for the water circulation process in the cold region according to claim 1, which is characterized by comprising the following steps: s3, a short-term memory network model internally comprises three gates for controlling unit states, wherein the first step of the operation of the short-term memory network model is to determine which information needs to be discarded in the cell state, and the operation is processed by a sigmoid unit called a forgetting gate; the sigmoid unit processes the information by looking at h t-1 And x t The information outputs a vector between 0 and 1, the 0-1 values inside the vector representing which C's in the cell state t-1 The information is retained, or discarded; wherein 0 indicates that no information is reserved, 1 indicates that both are reserved, and a sigmoid function and a tanh function are used as activation functions respectively;
an input door:
I t ＝σ(W i [h t-1 ,x t ]+b i )
in which W is i For inputting a gate weight matrix, sigma is a Sigmoid function, h t-1 For the output data of the memory cell at the previous time, x t B for input at the current time i Bias for the input gate;
forgetting the door:
f t ＝σ(W f [h t-1 ,x t ]+b f )
in which W is f For forgetting the gate weight matrix, b f Bias for forgetting the door;
output door:
O t ＝σ(W o [h t-1 ,x t ]+b o )
in which W is o To output the gate weight matrix, b o Offset for the output gate;
a memory unit:
in which W is C B is a memory cell weight matrix C Is the bias of the memory cell;
a memory unit c for updating the candidate value:
wherein C is t-1 The candidate value at the previous moment is referred;
the memory cell outputs:
h t ＝o t *tanh(C t )。
5. the method for constructing the dynamic combination prediction model for the water circulation process in the cold region according to claim 1, which is characterized by comprising the following steps: the short-term memory network model in S3 is optimized and trained through a small-batch gradient descent algorithm, and a mean square error function mse is used as a loss function:
wherein y is i Is the predicted value of the ith moment; y is Y i Is the true value at the i-th moment; n is the number of samples input by the model.
6. The method for constructing the dynamic combination prediction model for the water circulation process in the cold region according to claim 1, which is characterized by comprising the following steps: and S4, establishing an influence relationship among the cold region water circulation related factors through a dynamic Bayesian network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211442607.6A CN115759403B (en) | 2022-11-15 | 2022-11-15 | Dynamic combination prediction model construction method for water circulation process in cold region |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211442607.6A CN115759403B (en) | 2022-11-15 | 2022-11-15 | Dynamic combination prediction model construction method for water circulation process in cold region |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115759403A CN115759403A (en) | 2023-03-07 |
| CN115759403B true CN115759403B (en) | 2023-12-15 |
Family
ID=85372899
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211442607.6A Active CN115759403B (en) | 2022-11-15 | 2022-11-15 | Dynamic combination prediction model construction method for water circulation process in cold region |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115759403B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107203810A (en) * | 2017-05-22 | 2017-09-26 | 河海大学 | A kind of precipitation Forecasting Methodology based on depth network |
| CN110598242A (en) * | 2019-07-24 | 2019-12-20 | 浙江大学 | Novel hydrological model based on gridding watershed and classification calibration |
| CN110633859A (en) * | 2019-09-18 | 2019-12-31 | 西安理工大学 | Hydrological sequence prediction method for two-stage decomposition integration |
| CN111310968A (en) * | 2019-12-20 | 2020-06-19 | 西安电子科技大学 | LSTM neural network circulation hydrological forecasting method based on mutual information |
| CN111639748A (en) * | 2020-05-15 | 2020-09-08 | 武汉大学 | Watershed pollutant flux prediction method based on LSTM-BP space-time combination model |
| CN111833202A (en) * | 2020-07-14 | 2020-10-27 | 中国水利水电科学研究院 | Farmland evapotranspiration short-term prediction method considering crop coefficient dynamic change and rainfall |
| CN112183721A (en) * | 2020-09-16 | 2021-01-05 | 河海大学 | Construction method of combined hydrological prediction model based on self-adaptive differential evolution |
| CN114818367A (en) * | 2022-05-18 | 2022-07-29 | 珠江水利委员会珠江水利科学研究院 | Runoff simulation analysis method, system and medium suitable for alpine region |
| CN114819395A (en) * | 2022-05-19 | 2022-07-29 | 浙江大学 | Industry medium and long term load prediction method based on long and short term memory neural network and support vector regression combination model |
| CN115221800A (en) * | 2022-09-20 | 2022-10-21 | 武汉大学 | Extended period runoff set prediction method integrating natural gas generator and deep learning |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AR109623A1 (en) * | 2018-02-16 | 2019-01-09 | Pescarmona Enrique Menotti | PROCESS AND SYSTEM OF ANALYSIS AND HYDROLOGICAL MANAGEMENT FOR BASINS |
| KR102308751B1 (en) * | 2021-02-19 | 2021-10-05 | 주식회사 에스아이에이 | Method for prediction of precipitation based on deep learning |
-
2022
- 2022-11-15 CN CN202211442607.6A patent/CN115759403B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107203810A (en) * | 2017-05-22 | 2017-09-26 | 河海大学 | A kind of precipitation Forecasting Methodology based on depth network |
| CN110598242A (en) * | 2019-07-24 | 2019-12-20 | 浙江大学 | Novel hydrological model based on gridding watershed and classification calibration |
| CN110633859A (en) * | 2019-09-18 | 2019-12-31 | 西安理工大学 | Hydrological sequence prediction method for two-stage decomposition integration |
| CN111310968A (en) * | 2019-12-20 | 2020-06-19 | 西安电子科技大学 | LSTM neural network circulation hydrological forecasting method based on mutual information |
| CN111639748A (en) * | 2020-05-15 | 2020-09-08 | 武汉大学 | Watershed pollutant flux prediction method based on LSTM-BP space-time combination model |
| CN111833202A (en) * | 2020-07-14 | 2020-10-27 | 中国水利水电科学研究院 | Farmland evapotranspiration short-term prediction method considering crop coefficient dynamic change and rainfall |
| CN112183721A (en) * | 2020-09-16 | 2021-01-05 | 河海大学 | Construction method of combined hydrological prediction model based on self-adaptive differential evolution |
| CN114818367A (en) * | 2022-05-18 | 2022-07-29 | 珠江水利委员会珠江水利科学研究院 | Runoff simulation analysis method, system and medium suitable for alpine region |
| CN114819395A (en) * | 2022-05-19 | 2022-07-29 | 浙江大学 | Industry medium and long term load prediction method based on long and short term memory neural network and support vector regression combination model |
| CN115221800A (en) * | 2022-09-20 | 2022-10-21 | 武汉大学 | Extended period runoff set prediction method integrating natural gas generator and deep learning |
Non-Patent Citations (3)
| Title |
|---|
| 基于LSTM网络鄱阳湖抚河流域径流模拟研究;姜淞川;陆建忠;陈晓玲;刘子旋;;华中师范大学学报(自然科学版)(01);全文 * |
| 寒区水文循环模型研究进展;李保琦;肖伟华;王义成;孙青言;刘卓娅;;西南民族大学学报(自然科学版)(04);全文 * |
| 考虑气象因子不确定性的概率降水预报研究;邢贞相;张涵;付强;宫兴龙;李衡;;灌溉排水学报(10);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115759403A (en) | 2023-03-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Milan et al. | Development of adaptive neuro fuzzy inference system–Evolutionary algorithms hybrid models (ANFIS-EA) for prediction of optimal groundwater exploitation | |
| Hipni et al. | Daily forecasting of dam water levels: comparing a support vector machine (SVM) model with adaptive neuro fuzzy inference system (ANFIS) | |
| Bae et al. | Monthly dam inflow forecasts using weather forecasting information and neuro-fuzzy technique | |
| Buyukyildiz et al. | Estimation of the change in lake water level by artificial intelligence methods | |
| Akrami et al. | Improving rainfall forecasting efficiency using modified adaptive neuro-fuzzy inference system (MANFIS) | |
| Keshtegar et al. | Subset modeling basis ANFIS for prediction of the reference evapotranspiration | |
| CN110363349B (en) | ASCS-based LSTM neural network hydrological prediction method and system | |
| CN112989538B (en) | Control method and control device for urban drainage system | |
| CN112541839B (en) | Reservoir storage flow prediction method based on neural differential equation | |
| CN108764515A (en) | A kind of reservoir operation Application of risk decision method of Coupled Numerical meteorological model DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM | |
| Özger | Comparison of fuzzy inference systems for streamflow prediction | |
| Perea et al. | Prediction of applied irrigation depths at farm level using artificial intelligence techniques | |
| Rahgoshay et al. | Simulation of daily suspended sediment load using an improved model of support vector machine and genetic algorithms and particle swarm | |
| Latif et al. | Evaluation of deep learning algorithm for inflow forecasting: a case study of Durian Tunggal Reservoir, Peninsular Malaysia | |
| Deswal et al. | Artificial neural network based modeling of evaporation losses in reservoirs | |
| Yadav et al. | Prediction of suspended sediment yield by artificial neural network and traditional mathematical model in Mahanadi river basin, India | |
| Li et al. | Deep learning with long short-term memory networks for air temperature predictions | |
| CN116523269A (en) | Model predictive control-based river network dynamic intelligent regulation and control method | |
| Nguyen et al. | Water level forecasting using neuro-fuzzy models with local learning | |
| JP7473524B2 (en) | Model generation method, water level prediction system, inflow prediction system, and downstream river water level prediction system | |
| CN115759403B (en) | Dynamic combination prediction model construction method for water circulation process in cold region | |
| CN114676882A (en) | Hydrological multi-model time-varying weight combined forecasting method | |
| Nazeri-Tahroudi et al. | Estimation of dew point temperature in different climates of Iran using support vector regression | |
| Katipoğlu et al. | Prediction of flood routing results in the Central Anatolian region of Türkiye with various machine learning models | |
| Sojitra et al. | Comparative study of daily rainfall forecasting models using adaptive-neuro fuzzy inference system (ANFIS) |
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 |