CN116050630A - Lake multi-depth temperature prediction method and model driven by mechanism and data in combined mode - Google Patents

Lake multi-depth temperature prediction method and model driven by mechanism and data in combined mode Download PDF

Info

Publication number
CN116050630A
CN116050630A CN202310078182.3A CN202310078182A CN116050630A CN 116050630 A CN116050630 A CN 116050630A CN 202310078182 A CN202310078182 A CN 202310078182A CN 116050630 A CN116050630 A CN 116050630A
Authority
CN
China
Prior art keywords
depth
data
sequence
model
lake
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
Application number
CN202310078182.3A
Other languages
Chinese (zh)
Inventor
王莉
陈玲玲
许柏宁
曹昱博
宁泽飞
苗昊
麻巍祥
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Taiyuan University of Technology
Original Assignee
Taiyuan University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Taiyuan University of Technology filed Critical Taiyuan University of Technology
Priority to CN202310078182.3A priority Critical patent/CN116050630A/en
Publication of CN116050630A publication Critical patent/CN116050630A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/08—Fluids
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
    • 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
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Computing Systems (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Strategic Management (AREA)
  • Software Systems (AREA)
  • Human Resources & Organizations (AREA)
  • Artificial Intelligence (AREA)
  • Economics (AREA)
  • Mathematical Physics (AREA)
  • Quality & Reliability (AREA)
  • Biophysics (AREA)
  • General Business, Economics & Management (AREA)
  • Medical Informatics (AREA)
  • Tourism & Hospitality (AREA)
  • Operations Research (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Algebra (AREA)
  • Fluid Mechanics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a mechanism and data combined driving lake multi-depth temperature prediction method and model, belonging to the technical field of lake depth temperature prediction; the problem that the dependency of the depth sequence relation cannot be fully mined in the prior art is solved; the method comprises the following steps: inputting meteorological characteristic data into a physical model, and obtaining prediction data based on the physical model through simulation of the physical model; splicing meteorological characteristic data and predicted data based on a physical model, inputting the meteorological characteristic data and the predicted data as characteristics of a data driving model, preprocessing the data to obtain a time sequence characteristic matrix and a depth sequence characteristic matrix, and respectively capturing time sequence information and depth sequence information to obtain time sequence predicted data and depth sequence predicted data; inputting predicted data, time sequence predicted data and depth sequence predicted data based on a physical model into a fully-connected layer to predict lake temperatures at different depths at t time; the method is applied to lake depth temperature prediction.

Description

Lake multi-depth temperature prediction method and model driven by mechanism and data in combined mode
Technical Field
The invention provides a mechanism and data combined driving lake multi-depth temperature prediction method and model, and belongs to the technical field of lake depth temperature prediction.
Background
Lake temperature is an important factor affecting the ecological environment of lakes and affects the production activities of humans to some extent, such as water resistance, flood control, fishery, etc. Water temperature is the main driving factor for vertical stratification of lakes, so it significantly affects the transport of mass (including nutrients and dissolved oxygen), energy and momentum in the body of water. Higher lake temperatures can cause carbon stored in the lake sedimentary formations to convert to methane and carbon dioxide, thereby accelerating global warming in a feedback effect. Not only does the process of lake-gas interactions change regional climate characteristics, but interactions between different atmospheric loops in the climate system can also affect regional and global climate forecast accuracy. Therefore, it is important to accurately predict lake temperature.
Water temperature is the most important environmental condition in the biological community of lakes, and it has been shown that lake water temperature is affected by a number of factors, since lakes exchange energy between the atmosphere and the atmosphere. Lake temperature is also particularly susceptible to invasive species and land use development, which can lead to deterioration of water quality and loss of ecosystem integrity. The temperature of the entire lake is considerably affected due to physical processes (e.g., thermal layering, mixing processes), geochemical processes (e.g., chemical reaction rates, oxygen solubility), and ecological processes (e.g., metabolism, growth, and reproduction of organisms). The underground environment of the lake is also quite complex, and the temperature in different depths, different moments and different places of the lake becomes phantom and is unrealistic to measure the temperature manually. Therefore, accurate and effective prediction of lake temperature is very difficult.
Traditional research methods are mainly based on physical and statistical methods, for example, a physical-based air2water model can provide a robust prediction of lake surface temperature under the condition that only air temperature is known; the GLM model based on the process can simulate the fluid dynamics of lakes, reservoirs and wetlands, simulate the sub-model details of the vertical mixing and inflow-outflow dynamics of the surface heat exchange and the ice cover dynamics, and is suitable for different lake types with larger difference. Many parameters in a physical or process-based model require a large amount of observation data to calibrate, increasing the accuracy of the model requires a large computational cost and a large amount of observation data, while a deep learning model shows its good application effect in many fields, so that a data-driven method is introduced to the task of lake temperature prediction, for example, using a deep learning architecture of an entity-aware long-short-term memory (EA-LSTM) network to predict lake japanese surface temperature, but the data-driven model requires a large amount of data with high quality, and ignores process assumptions behind the data, which is poor in interpretability and may lead to false or inaccurate predictions. Thus, many students consider using a physically guided deep learning model to make predictions of lake temperature. The PGNN model is proposed by a skilled person, the model takes the output of a physical model as the characteristic input of a neural network, and a loss function based on physics is designed according to the relation of lake depth and density, so that the physical consistency of a prediction result is ensured. Other technicians have proposed PGRNN models that are pre-trained using uncalibrated physical model output data, so that the model can achieve good predictive results using less observation data. Like PGNNs, PGRNNs design a loss function based on the law of conservation of energy to normalize the learning of neural networks. Another technician proposes a PAG model that hard codes physical constraints in the LSTM architecture, preserving the predicted monotonicity and physical consistency of LSTM. However, the fusion mode of the physical model and the data model is single, a large lifting space still exists, and great challenges still exist in improving the lake temperature prediction accuracy.
Disclosure of Invention
The invention provides a multi-depth temperature prediction method and model for a lake, which are driven by a mechanism and data in a combined way, and aims to solve the problems that the dependency of depth sequence relation cannot be fully excavated, the utilization of a lake temperature physical model is insufficient, the temperature values of different depths of the lake at different times cannot be accurately predicted and the like in the prior art.
In order to solve the technical problems, the invention adopts the following technical scheme: a lake multi-depth temperature prediction method driven by mechanism and data combination comprises the following steps:
s1: inputting meteorological characteristic data into a physical model, and obtaining prediction data based on the physical model through simulation of the physical model;
s2: splicing meteorological characteristic data and predicted data based on a physical model, inputting the meteorological characteristic data and the predicted data as characteristics of a data driving model, preprocessing the data to obtain a time sequence characteristic matrix and a depth sequence characteristic matrix, and respectively capturing time sequence information and depth sequence information to obtain time sequence predicted data and depth sequence predicted data;
s3: and inputting predicted data, time sequence predicted data and depth sequence predicted data based on the physical model into the fully-connected layer to predict lake temperatures at different depths at the moment t.
The object in the step S1The GLM model is adopted as the physical model, the GLM model uses a training set to calibrate parameters, and after the calibrated GLM model is obtained, the model is input into meteorological characteristic data X acquired in one day f Outputting predicted data based on the physical model
Figure BDA0004066645580000021
The step S2 specifically includes:
predictive data to be based on physical model
Figure BDA0004066645580000022
And weather characteristic data X f Splicing to obtain a feature matrix, performing a series of matrix transformations on the feature matrix to respectively obtain the input X of the time sequence prediction module time And input X of depth sequence prediction module depth The depth sequence prediction module and the time sequence prediction module both adopt an LSTM architecture;
input sequence X of the time sequence prediction module time Characteristic data representing depth d is fixed, time t-nΔt to t:
Figure BDA0004066645580000023
input sequence X of the depth sequence prediction module depth Characteristic data representing time tdetermined, depth 0 to d:
Figure BDA0004066645580000031
inputting the obtained time sequence prediction module into X time And depth sequence prediction module input X depth The time sequence prediction module outputs predicted values based on time sequence relation through LSTM operation in the respective modules
Figure BDA0004066645580000032
The depth sequence prediction module outputs a predicted value +.>
Figure BDA0004066645580000033
The step S3 specifically includes:
outputting the prediction of the physical model with a certain t and a certain depth d
Figure BDA0004066645580000034
Predictive value based on time series relation
Figure BDA0004066645580000035
And a predicted value based on depth sequence relation +.>
Figure BDA0004066645580000036
Passing through a Dense layer, the number of neurons of the Dense layer is set to 1, so as to obtain a temperature predicted value with a certain final time t and a certain depth d>
Figure BDA0004066645580000037
/>
Figure BDA0004066645580000038
In the above formula: f-function is a linear activation function, w time 、w depth 、w phy Representing the weight, b represents the bias term.
Updating model parameters by using a loss function, and defining the loss function
Figure BDA0004066645580000039
Is true value y t,d And predictive value->
Figure BDA00040666455800000310
Root mean square error between:
Figure BDA00040666455800000311
wherein s is the number of samples;
designing loss function loss based on law of conservation of energy by using PGRNN simultaneously Ec The two are added to the final loss function:
Figure BDA0004066645580000041
wherein lambda is Ec Coefficients of a loss function that is the law of conservation of energy;
for each complete prediction of all samples of a batch, a penalty value is calculated and back-propagated to update the sample parameters.
A mechanism and data jointly driven lake multi-depth temperature prediction model comprising:
the system comprises a physical model prediction module and a data driving model prediction module, wherein the data driving model prediction module comprises a time sequence prediction module and a depth sequence prediction module;
the input of the physical model prediction module is a meteorological feature, so that a prediction output based on a physical model is obtained, the prediction output of the physical model and the meteorological feature of the original input are spliced, and the prediction output is used as the input of the data driving model prediction module through data preprocessing, so that a prediction value based on a time sequence relation and a prediction value based on a depth sequence relation are obtained;
and inputting the predicted output of the physical model with a certain time and a certain depth into the full-connection layer together with the predicted value based on the time sequence relation and the predicted value based on the depth sequence relation to obtain the final predicted value of the lake temperature.
The time sequence prediction module is used for capturing a time sequence relation in a time sequence, and an LSTM unit is adopted to obtain a temperature predicted value based on the time sequence relation.
The depth sequence prediction module is used for capturing the depth sequence relation of the lake and obtaining a temperature predicted value based on the depth sequence relation by adopting an LSTM unit.
And (3) the predicted output of the physical model with a certain time and a certain depth passes through a full-connection layer together with the predicted value based on the time sequence relation and the predicted value based on the depth sequence relation, and finally the predicted value of the lake temperature with a certain time and a certain depth is output.
Compared with the prior art, the invention has the following beneficial effects:
1) The method provided by the invention is a mechanism and data combined driving method, can fully utilize a mechanism model to solve the problem of high quality data required by a data driving model, and combines the powerful representation capability of deep learning to perform more accurate prediction;
2) The method provided by the invention has generalization, can be applied to temperature prediction of scenes such as oceans, marshes and the like, and the thought of combined driving of mechanism and data can be applied to other complex scenes such as energy management, traffic planning, field equipment monitoring management and the like;
3) The method provided by the invention has the advantages of advancement, stability and practicability, high prediction accuracy and remarkable performance improvement on two lake data sets.
Drawings
The invention is further described below with reference to the accompanying drawings:
FIG. 1 is a flow chart of a prediction method of the present invention;
FIG. 2 is a schematic diagram of a prediction method according to the present invention.
Detailed Description
1-2, the invention provides a mechanism and data combined driving multi-depth lake temperature prediction method, which fuses a physical model and a data driving model. Firstly, inputting meteorological feature data into a physical model, obtaining prediction data based on the physical model through physical model simulation, then splicing the meteorological feature data and the prediction data based on the physical model, preprocessing the data to obtain a time sequence feature matrix and a depth sequence feature matrix, and respectively capturing time sequence information and depth sequence information to obtain time sequence prediction data and depth sequence prediction data. And finally, inputting output predicted values of the physical model prediction module, the time sequence prediction module and the depth sequence prediction module into a full-connection layer to obtain a final predicted result, wherein the final purpose is to predict lake temperatures at different depths at the moment t.
As shown in FIG. 1, the method for predicting the temperature of the multi-depth lake driven by the combination of the mechanism and the data comprises the following specific steps:
step one, 9 characteristics X such as weather conditions, depth information, time stamp and the like in one day are obtained f As input to the physical model GLM, the physical model is calibrated using training data to obtain a predicted output based on the physical model
Figure BDA0004066645580000051
Outputting the prediction based on the physical model
Figure BDA0004066645580000052
And X is f Splicing in time and depth, the LSTM architecture is used for both sequence predictions, except for the input sequence X based on time sequence predictions time Characteristic data representing depth d is fixed, time t-nΔt to t:
Figure BDA0004066645580000053
input sequence X based on depth sequence prediction module depth Characteristic data representing time tdetermined, depth 0 to d:
Figure BDA0004066645580000054
therefore, a series of matrix transformations are required to be performed on the feature matrix to obtain the inputs X of the time-series prediction module and the depth-series prediction module, respectively time And X depth 。
Step three, inputting the obtained time sequence prediction module into a matrix X time And depth sequence prediction module input matrix X depth The time sequence prediction module outputs a time sequence relation based on LSTM operation in each modulePredictive value
Figure BDA0004066645580000055
The depth sequence prediction module outputs a predicted value +.>
Figure BDA0004066645580000056
Outputting the prediction of the physical model with a certain t and a certain depth d
Figure BDA0004066645580000057
Predicted value based on time series relation ++>
Figure BDA0004066645580000058
And a predicted value based on depth sequence relation +.>
Figure BDA0004066645580000059
Passing through a Dense layer, the number of neurons of the Dense layer is set to 1, so as to obtain a temperature predicted value with a certain final time t and a certain depth d>
Figure BDA0004066645580000061
Figure BDA0004066645580000062
/>
In the above formula: f-function is a linear activation function, w time 、w depth 、w phy Representing the weight, b represents the bias term.
Loss function
Figure BDA0004066645580000063
Defined as the true value y t,d And predictive value->
Figure BDA0004066645580000064
Root mean square error between:
Figure BDA0004066645580000065
wherein s is the number of samples;
designing loss function loss based on law of conservation of energy by using PGRNN simultaneously Ec The two are added to the final loss function:
Figure BDA0004066645580000066
wherein lambda is Ec Coefficients of a loss function that is the law of conservation of energy;
for each complete prediction of all samples of a batch, a penalty value is calculated and back-propagated to update the sample parameters.
FIG. 2 is a block diagram of a mechanism and data combined driving multi-depth lake temperature prediction model of the invention, which mainly comprises three modules and a final combination part, and the specific implementation modes are as follows:
module I, physical model prediction module
Among the lake temperature prediction problems, the most widely used physical model is the GLM model, the parameters of which have very definite physical significance, which can simulate the fluid dynamics of lakes, reservoirs and wetlands, and can simulate the sub-model details of the surface heat exchange and ice cover dynamics vertical mixing and inflow-outflow dynamics, and the predicted values of which have physical knowledge characterization. Therefore, the invention utilizes the simulation output of the GLM physical model to improve the quantity and quality of the data required by the data driven model.
The specific process is as follows:
the GLM model has a plurality of parameters, the parameters need to be calibrated by using a training set, the data set is divided into a training set and a testing set, the GLM model is operated by using the training set for each possible parameter value combination, and 9 characteristic X such as weather conditions, depth information, time stamps and the like in one day are input f And selecting a group of parameter values with the smallest error with the observed value to obtain a calibrated GLM model. Then using the GLM model after test set operation calibration to obtainPrediction output based on physical model
Figure BDA0004066645580000071
Obtaining a simulation output based on a physical model
Figure BDA0004066645580000072
Then, the physical model and the data driving model are fused to jointly drive the modeling of lake temperature, and the fusion place mainly comprises two parts: (1) Prediction output based on physical model
Figure BDA0004066645580000073
And combining according to different time and different depths to obtain a new data set, selecting matrix data with different time and different depths as input of a prediction module based on a time sequence, and outputting a lake temperature predicted value with a time sequence relation. And selecting matrix data with different depths and certain time as input of a depth sequence prediction module, and outputting lake temperature predicted values with depth sequence relations. Therefore, the prediction module based on the depth sequence and the prediction module based on the time sequence can learn physical knowledge, so that the predicted values of the two modules are more accurate. (2) Prediction output of physical model with constant time t and depth d
Figure BDA0004066645580000074
Predicted value to be associated with time series based relationship +.>
Figure BDA0004066645580000075
Predicted values based on depth sequence relationships
Figure BDA0004066645580000076
Inputting all the layers together to obtain final temperature predicted value +.>
Figure BDA0004066645580000077
By such an operation, the final temperature predicted value is obtained by taking the time series relation and the depth series relation into considerationAnd knowledge of physical laws, thereby obtaining more accurate prediction results.
Module two, time sequence prediction module
In order to capture the time sequence relation in the time sequence, the invention captures the time sequence information by using the LSTM unit in the time sequence prediction module, wherein the LSTM is a variant model of the RNN, inherits the advantages of the RNN, has a long-time memory function, solves the problems of gradient elimination and gradient explosion in the long-sequence training process, and has certain advantages in the sequence modeling problem. The specific method and process are as follows:
prediction output based on physical model
Figure BDA0004066645580000078
And X is f Processing the spliced matrix, selecting a characteristic matrix with a certain depth d and a time t-nDeltat to t as an input sequence X based on a time sequence prediction module time ：
Figure BDA0004066645580000079
Based on a time sequence prediction module, the invention selects LSTM to mine the time sequence relation of lake temperature and output a predicted value based on the time sequence relation
Figure BDA00040666455800000710
The mathematical model of the LSTM neural network is as follows:
i t ＝σ(W i ·[h t-1 ，x t ]+b i )；
f t ＝σ(W f ·[h t-1 ，x t ]+b f )；
o t ＝σ(W o ·[h t-1 ，x t ]+b o )；
Figure BDA0004066645580000081
Figure BDA0004066645580000082
h t ＝o t *tanh(C t )；
in the above formula: i.e t 、f t And o t Respectively represent an input door, a forget door and an output door, W i 、W f And W is equal to o Weight matrix of input gate, forget gate and output gate, b i 、b f And b o Bias terms of three gates, h t-1 Represents the output of LSTM at time t-1, x t The input at time t is indicated,
Figure BDA0004066645580000083
information indicating new input at time t, C t The cell state at time t is represented, and σ represents the sigmod function.
Module three, depth sequence prediction module
For lake temperature prediction problems, it is often necessary to predict temperatures at different depths in a lake. In order to fully capture the sequence relation of the lake depth, similar to a time sequence prediction module, the invention also uses LSTM units to model the temperature predictions of the lake at different depths. The specific method and process are as follows:
prediction output based on physical model
Figure BDA0004066645580000084
And X is f Processing the spliced matrix, selecting characteristic data with certain time t and depth 0 to d as an input sequence X based on a depth sequence prediction module depth ：
Figure BDA0004066645580000085
L based on the same structure as the temporal sequence prediction module is selected by the depth sequence prediction moduleSTM excavates lake temperature depth sequence relation, outputs predicted value based on depth sequence relation
Figure BDA0004066645580000086
The structure of the LSTM neural network is consistent with that of the time sequence prediction module, and the hidden neurons are set to be 20./>
Binding portion:
after the prediction output of the three modules is obtained, the prediction output of a physical model with a certain t and a certain depth d is obtained
Figure BDA0004066645580000091
Predicted value based on time series relation ++>
Figure BDA0004066645580000092
And a predicted value based on depth sequence relation +.>
Figure BDA0004066645580000093
Passing through a Dense layer, the number of neurons of the Dense layer is set to 1, so as to obtain a temperature predicted value with a certain final time t and a certain depth d>
Figure BDA0004066645580000094
Figure BDA0004066645580000095
Wherein f-function is a linear activation function, w time 、w depth 、w phy Representing the weight, b represents the bias term.
The specific implementation cases are as follows:
the invention selects the data set of two evaluation models, the first lake data set is Mendota lake, the lake area is about 40 square kilometers, the depth range is 0-25 meters, the whole data of the Mendota lake comprises 35242 temperature observations in 2009-2017, and the invention selects the temperature observation values from 0m to 25m (the interval is 0.5m, and the total depth is 50). The second lake dataset was a Sparkling lake, with 36 depths in the range of 0-18m, and the entire dataset also included temperature observations at different depths in the 2009-2017 year lake. The characteristic data set comprises characteristics of weather conditions and the like every day from 2009 to 2017, and the characteristics comprise 10 characteristics of days in one year, depth, short wave radiation, long wave radiation and the like.
To verify the performance of the model of the invention, it was compared with 3 reference models, which are respectively: based on the process model (PB), the deep learning model (DL), the cyclic neural network (PGRNN) is physically guided. The experimental results are shown in table 1, which demonstrates the behavior of the model of the present invention and the baseline model in both data sets.
Figure BDA0004066645580000096
Table 1 a comparison of the model of the invention with the baseline model in two datasets;
it can be seen from table 1 that the model of the present invention performs significantly better on both data sets than the other three baseline models.
The invention provides a time sequence model design and a depth sequence model design. The input feature matrix is processed and transposed to obtain time sequence input data and depth sequence input data respectively, and then two different LSTM units are used for prediction respectively, so that the sequence relation of lake temperature in two dimensions of time and depth is captured.
The mechanism and data combined driving fusion mechanism provided by the invention optimizes the mechanism model by utilizing the historical data, and the mechanism model result is not only used as a characteristic of the data driving model, but also fused with the result of the data driving model, so that the final temperature prediction accuracy is improved.
The specific structure of the invention needs to be described that the connection relation between the component modules adopted by the invention is definite and realizable, and besides the specific description in the embodiment, the specific connection relation can bring about corresponding technical effects, and on the premise of not depending on execution of corresponding software programs, the technical problems of the invention are solved, the types of the components, the modules and the specific components, the connection modes of the components and the expected technical effects brought by the technical characteristics are clear, complete and realizable, and the conventional use method and the expected technical effects brought by the technical characteristics are all disclosed in patents, journal papers, technical manuals, technical dictionaries and textbooks which can be acquired by a person in the field before the application date, or the prior art such as conventional technology, common knowledge in the field, and the like, so that the provided technical scheme is clear, complete and the corresponding entity products can be reproduced or obtained according to the technical means.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.

Claims (9)

1. A lake multi-depth temperature prediction method driven by mechanism and data is characterized by comprising the following steps of: the method comprises the following steps:
s1: inputting meteorological characteristic data into a physical model, and obtaining prediction data based on the physical model through simulation of the physical model;
s2: splicing meteorological characteristic data and predicted data based on a physical model, inputting the meteorological characteristic data and the predicted data as characteristics of a data driving model, preprocessing the data to obtain a time sequence characteristic matrix and a depth sequence characteristic matrix, and respectively capturing time sequence information and depth sequence information to obtain time sequence predicted data and depth sequence predicted data;
s3: and inputting predicted data, time sequence predicted data and depth sequence predicted data based on the physical model into the fully-connected layer to predict lake temperatures at different depths at the moment t.
2. The method for predicting the lake multi-depth temperature driven by combination of mechanism and data according to claim 1, wherein the method comprises the following steps: the physical model in the step S1 adopts a GLM model, the GLM model uses a training set to calibrate parameters, and after the calibrated GLM model is obtained, the model is input into meteorological characteristic data X acquired in one day f Outputting predicted data based on the physical model
Figure FDA0004066645570000011
3. The method for predicting the lake multi-depth temperature driven by combination of mechanism and data according to claim 2, wherein the method comprises the following steps: the step S2 specifically includes:
predictive data to be based on physical model
Figure FDA0004066645570000012
And weather characteristic data X f Splicing to obtain a feature matrix, performing a series of matrix transformations on the feature matrix to respectively obtain the input X of the time sequence prediction module time And input X of depth sequence prediction module depth The depth sequence prediction module and the time sequence prediction module both adopt an LSTM architecture;
input sequence X of the time sequence prediction module time Characteristic data representing depth d is fixed, time t-nΔt to t:
Figure FDA0004066645570000013
input sequence X of the depth sequence prediction module depth Characteristic data representing time tdetermined, depth 0 to d:
Figure FDA0004066645570000014
inputting the obtained time sequence prediction module into X time Depth of sumInput X of degree sequence prediction module depth The time sequence prediction module outputs predicted values based on time sequence relation through LSTM operation in the respective modules
Figure FDA0004066645570000015
The depth sequence prediction module outputs a predicted value +.>
Figure FDA0004066645570000016
4. A method for mechanism and data driven lake multi-depth temperature prediction in accordance with claim 3, wherein: the step S3 specifically includes:
outputting the prediction of the physical model with a certain t and a certain depth d
Figure FDA0004066645570000021
Predictive value based on time series relation
Figure FDA0004066645570000022
And a predicted value based on depth sequence relation +.>
Figure FDA0004066645570000023
Passing through a Dense layer, the number of neurons of the Dense layer is set to 1, so as to obtain a temperature predicted value with a certain final time t and a certain depth d>
Figure FDA0004066645570000024
Figure FDA0004066645570000025
In the above formula: f-function is a linear activation function, w time 、w depth 、w phy Representing the weight, b represents the bias term.
5. The method for predicting the lake multi-depth temperature driven by combination of mechanism and data according to claim 4, wherein the method comprises the following steps: updating model parameters by using a loss function, and defining the loss function
Figure FDA0004066645570000026
Is true value y t,d And predictive value->
Figure FDA0004066645570000027
Root mean square error between:
Figure FDA0004066645570000028
wherein s is the number of samples;
designing loss function loss based on law of conservation of energy by using PGRNN simultaneously Ec The two are added to the final loss function:
Figure FDA0004066645570000029
wherein lambda is Ec Coefficients of a loss function that is the law of conservation of energy;
for each complete prediction of all samples of a batch, a penalty value is calculated and back-propagated to update the sample parameters.
6. A lake multi-depth temperature prediction model driven by mechanism and data is characterized in that: comprising the following steps:
the system comprises a physical model prediction module and a data driving model prediction module, wherein the data driving model prediction module comprises a time sequence prediction module and a depth sequence prediction module;
the input of the physical model prediction module is a meteorological feature, so that a prediction output based on a physical model is obtained, the prediction output of the physical model and the meteorological feature of the original input are spliced, and the prediction output is used as the input of the data driving model prediction module through data preprocessing, so that a prediction value based on a time sequence relation and a prediction value based on a depth sequence relation are obtained;
and inputting the predicted output of the physical model with a certain time and a certain depth into the full-connection layer together with the predicted value based on the time sequence relation and the predicted value based on the depth sequence relation to obtain the final predicted value of the lake temperature.
7. The mechanism and data driven lake multi-depth temperature prediction model of claim 6, wherein: the time sequence prediction module is used for capturing a time sequence relation in a time sequence, and an LSTM unit is adopted to obtain a temperature predicted value based on the time sequence relation.
8. The mechanism and data driven lake multi-depth temperature prediction model of claim 7, wherein: the depth sequence prediction module is used for capturing the depth sequence relation of the lake and obtaining a temperature predicted value based on the depth sequence relation by adopting an LSTM unit.
9. The mechanism and data driven lake multi-depth temperature prediction model of claim 8, wherein: and (3) the predicted output of the physical model with a certain time and a certain depth passes through a full-connection layer together with the predicted value based on the time sequence relation and the predicted value based on the depth sequence relation, and finally the predicted value of the lake temperature with a certain time and a certain depth is output.
CN202310078182.3A 2023-02-08 2023-02-08 Lake multi-depth temperature prediction method and model driven by mechanism and data in combined mode Pending CN116050630A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310078182.3A CN116050630A (en) 2023-02-08 2023-02-08 Lake multi-depth temperature prediction method and model driven by mechanism and data in combined mode

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310078182.3A CN116050630A (en) 2023-02-08 2023-02-08 Lake multi-depth temperature prediction method and model driven by mechanism and data in combined mode

Publications (1)

Publication Number Publication Date
CN116050630A true CN116050630A (en) 2023-05-02

Family

ID=86129377

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310078182.3A Pending CN116050630A (en) 2023-02-08 2023-02-08 Lake multi-depth temperature prediction method and model driven by mechanism and data in combined mode

Country Status (1)

Country Link
CN (1) CN116050630A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117875193A (en) * 2024-03-12 2024-04-12 北京师范大学 Lake thermodynamic process simulation method and simulation system

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117875193A (en) * 2024-03-12 2024-04-12 北京师范大学 Lake thermodynamic process simulation method and simulation system
CN117875193B (en) * 2024-03-12 2024-05-28 北京师范大学 Lake thermodynamic process simulation method and simulation system

Similar Documents

Publication Publication Date Title
Daw et al. Physics-guided neural networks (pgnn): An application in lake temperature modeling
Jia et al. Physics-guided machine learning for scientific discovery: An application in simulating lake temperature profiles
CN107688871B (en) Water quality prediction method and device
CN109142171B (en) Urban PM10 concentration prediction method based on feature expansion and fusing with neural network
Navone et al. Predicting Indian monsoon rainfall: a neural network approach
CN112651665B (en) Surface water quality index prediction method and device based on graph neural network
CN112488415A (en) Power load prediction method based on empirical mode decomposition and long-and-short-term memory network
CN114399073A (en) Ocean surface temperature field prediction method based on deep learning
CN116050630A (en) Lake multi-depth temperature prediction method and model driven by mechanism and data in combined mode
CN113627093A (en) Underwater mechanism cross-scale flow field characteristic prediction method based on improved Unet network
CN114462718A (en) CNN-GRU wind power prediction method based on time sliding window
CN115438582A (en) PM2.5 concentration prediction method combining multiple elements and attention
Xiong et al. Ai-goms: Large ai-driven global ocean modeling system
Agga et al. Short-term load forecasting: based on hybrid CNN-LSTM neural network
CN114034334A (en) Method for identifying pollution source and flow of karst pipeline
CN113484882A (en) GNSS sequence prediction method and system of multi-scale sliding window LSTM
Mu et al. PIRT: A physics-informed red tide deep learning forecast model considering causal-inferred predictors selection
CN116483036B (en) Transformer-based self-encoder soft measurement modeling method
CN117114168A (en) Long-time-scale sea surface temperature intelligent forecasting method
CN116522648A (en) Lake algae state prediction method and application
CN116628444A (en) Water quality early warning method based on improved meta-learning
CN116245018A (en) Sea wave missing measurement data forecasting method based on bivariate long-short-term memory algorithm
Zhou et al. Water quality prediction approach based on t-SNE and SA-BiLSTM
Li et al. Deepphysinet: Bridging deep learning and atmospheric physics for accurate and continuous weather modeling
Bano-Medina et al. Transferability and explainability of deep learning emulators for regional climate model projections: Perspectives for future applications

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