CN114862035A - Combined bay water temperature prediction method based on transfer learning - Google Patents
Combined bay water temperature prediction method based on transfer learning Download PDFInfo
- Publication number
- CN114862035A CN114862035A CN202210556998.8A CN202210556998A CN114862035A CN 114862035 A CN114862035 A CN 114862035A CN 202210556998 A CN202210556998 A CN 202210556998A CN 114862035 A CN114862035 A CN 114862035A
- Authority
- CN
- China
- Prior art keywords
- water temperature
- data
- day
- prediction model
- bay
- 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.)
- Granted
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 193
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000013526 transfer learning Methods 0.000 title claims abstract description 17
- 238000013508 migration Methods 0.000 claims abstract description 35
- 230000005012 migration Effects 0.000 claims abstract description 35
- 238000003062 neural network model Methods 0.000 claims abstract description 11
- 238000013528 artificial neural network Methods 0.000 claims description 52
- 238000012360 testing method Methods 0.000 claims description 36
- 238000010606 normalization Methods 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 16
- 230000002354 daily effect Effects 0.000 claims description 15
- 230000003203 everyday effect Effects 0.000 claims description 15
- 238000012417 linear regression Methods 0.000 claims description 12
- 238000005259 measurement Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 230000008014 freezing Effects 0.000 claims description 4
- 238000007710 freezing Methods 0.000 claims description 4
- 241001596950 Larimichthys crocea Species 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000001556 precipitation Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 241000764238 Isis Species 0.000 claims 1
- 230000008859 change Effects 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 230000035772 mutation Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 241000251468 Actinopterygii Species 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- XKMRRTOUMJRJIA-UHFFFAOYSA-N ammonia nh3 Chemical compound N.N XKMRRTOUMJRJIA-UHFFFAOYSA-N 0.000 description 1
- 238000009360 aquaculture Methods 0.000 description 1
- 244000144974 aquaculture Species 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 235000015205 orange juice Nutrition 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- 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
- G06N3/08—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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
-
- 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
- G06F2119/08—Thermal analysis or thermal optimisation
-
- 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)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Tourism & Hospitality (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Business, Economics & Management (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- Medical Informatics (AREA)
- Geometry (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Operations Research (AREA)
- Computer Hardware Design (AREA)
- Entrepreneurship & Innovation (AREA)
- Agronomy & Crop Science (AREA)
- Animal Husbandry (AREA)
- Marine Sciences & Fisheries (AREA)
- Mining & Mineral Resources (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a combined gulf water temperature prediction method based on transfer learning, which utilizes a B gulf with more historical observation data to construct an optimal source LSTM neural network model and then is based on the optimal source LSTM neural network model, the optimal migration LSTM neural network model of the A bay with less historical observation data is constructed through migration learning, then the pure difference regression prediction model and the mixed difference regression prediction model are combined, the three models are combined into the migration water temperature online prediction model of the A bay or the source water temperature online prediction model of the B bay in a dynamic weighting mode, not only can the source water temperature online prediction model be used for predicting the water temperature of the bay with more historical observation data, but also the migration water temperature online prediction model can be used for predicting the water temperature of the bay with less historical observation data, the prediction precision is high, the result is reliable, and the problem that the water temperature prediction precision is not high due to insufficient data volume in the prior art is solved.
Description
Technical Field
The invention relates to the technical field of water temperature prediction, in particular to a combined gulf water temperature prediction method based on transfer learning.
Background
Water temperature is one of the key factors influencing the growth condition and growth quality of fishes in aquaculture, and sudden change of water temperature can influence the pH value, dissolved oxygen, ammonia nitrogen and the like in a water body and influence the suitable living environment of the fishes.
The water temperature is easily influenced by multiple factors such as air temperature, aquatic life activities and the like in a cross mode, and has the characteristics of instability, nonlinearity and the like, so that the defects of low prediction precision, poor generalization performance and the like exist in a single prediction method, and the actual requirements of people on fishery production management are difficult to meet.
The existing water temperature prediction methods mainly comprise a traditional mathematical statistics method and an artificial intelligence method, the mathematical method based on the water temperature mechanism analyzes the mechanism and factors influencing the water temperature change, researches the influence of different water layer conditions, space-time distribution, flow velocity, flow and the like on the water temperature, and constructs a water temperature prediction model by using a hydrological meteorological theory, a two-dimensional model, a three-dimensional model and the like, and although the methods can effectively predict the water temperatures of a reservoir and a water tank, the prediction model is complex, has more parameters and is difficult to obtain; a non-mechanism method based on data mining generally constructs a mathematical model through a large amount of data, obtains the relation between water temperature and relevant influence factors, captures the change rule of the water temperature, and mostly uses a machine learning algorithm aiming at the mathematical model constructed by a large amount of samples, but a single machine learning algorithm can only predict the change trend of the water temperature and cannot effectively predict the water temperature mutation point. Machine learning often requires a large amount of training data to better develop its effects. However, in many practical scenarios, the target domain does not have enough training data, and the constructed general model is difficult to predict accurately for a specific scenario. At present, a water temperature prediction method capable of effectively predicting the water temperature change trend and the water temperature mutation point is still lacked, and meanwhile, the contradiction between machine learning and less training data and the contradiction between a general model and personalized requirements are faced.
Disclosure of Invention
The invention aims to provide a combined gulf water temperature prediction method based on transfer learning, which is high in precision and reliable in result, can better predict the water temperature value of 7 days in the future of a gulf, helps a target model to acquire knowledge learned by a source model through the transfer learning technology so as to improve the prediction performance of the model, solves the problem of low water temperature prediction precision caused by insufficient data volume in the prior art, and can provide data, information and service guarantee for marine fishery production, scientific research and management.
In order to achieve the purpose, the invention adopts the following technical scheme:
a combined bay water temperature prediction method based on transfer learning comprises the following steps:
s1, collecting water temperature data and meteorological data of an A bay and a B bay respectively, and preprocessing the data respectively, wherein the historical observation data of the B bay is more than that of the A bay;
s2, respectively constructing pure differential regression prediction models of the A bay and the B bay based on the respective forecast air temperature differential values and the actual measured water temperature values; respectively constructing a mixed differential regression prediction model of the A bay and the B bay based on the respective forecast air temperature differential value, the actual measurement water temperature differential value and the actual measurement water temperature value;
s3, aiming at the Bay B, constructing an LSTM neural network model, and selecting an LSTM neural network prediction model of an optimal source through training and testing; combining the optimal source LSTM neural network prediction model, the pure differential regression prediction model of the B bay and the mixed differential regression prediction model into a source water temperature online prediction model based on a dynamic weighting mode; predicting the water temperature of the bay B according to the source water temperature online prediction model;
s4, aiming at the gulf A, selecting the optimal source LSTM neural network prediction model of the gulf B, and selecting the optimal migration LSTM neural network prediction model through migration learning and testing; combining the optimal migration LSTM neural network prediction model, the pure differential regression prediction model of the A bay and the mixed differential regression prediction model into a migration water temperature online prediction model based on a dynamic weighting mode; and predicting the water temperature of the A bay according to the online migration water temperature prediction model.
Preferably, the water temperature data in step S1 includes daily maximum water temperature, daily minimum water temperature, and daily average water temperature; the meteorological data comprises meteorological observation data and meteorological forecast data; the meteorological observation data comprise air temperature statistics, air pressure statistics, relative humidity statistics, wind speed statistics, wind direction statistics and precipitation, and the time resolution is 1 day; the weather forecast data includes future 7-day weather forecast data updated every day, and the future 7-day weather forecast data includes day highest temperature and day lowest temperature of the future 7 days.
Preferably, the method for constructing the pure differential regression prediction model in step S2 is as follows:
setting the predicted temperature values of the weather forecast released on the k day asMeasured value of water temperature is x k If the current time is t, the predicted value of the water temperature of the day 1 isThe calculation formula of (2) is as follows:
predicted value of water temperature of nth (n is more than or equal to 2 and less than or equal to 7) dayThe calculation formula of (2) is as follows:
wherein x is t-1 The measured value of the water temperature on the t-1 day is measured;the air temperature forecast value of the day is acquired for the t +1-i days;obtaining the air temperature forecast value of the day for the t-i days;an air temperature forecast for the nth day in the future acquired for the t-th day;the temperature forecast value of the day is acquired on the t day; w is a p1 ,w p2 ,w p3 And and selecting the weight by constructing a linear regression equation.
Preferably, the construction method of the mixed difference regression prediction model in step S2 is:
setting the predicted temperature values of the weather forecast released on the k day asMeasured value of water temperature is x k If the current time is t, the predicted value of the water temperature of the day 1 isThe calculation formula of (2) is as follows:
predicted value of water temperature of nth (n is more than or equal to 2 and less than or equal to 7) dayThe calculation formula of (2) is as follows:
wherein x is t-1 The measured value of the water temperature on the t-1 day is measured; x is the number of t-2 The measured value of the water temperature on the t-2 th day;obtaining the temperature forecast value of the day for the t day;obtaining the air temperature forecast value of the day for the t-1 day;obtaining an air temperature forecast value for the nth day in the future for the t day; w is a m1 ,w m2 Andand selecting the weight by constructing a linear regression equation.
Preferably, w is chosen by means of a linear regression equation p1 ,w p2 ,w p3 ,w m1 ,w m2 And the concrete mode is as follows:
the predicted temperature values of the weather forecast released on the kth day are respectively set asMeasured value of water temperature is x k When the current time is t, according to the historical water temperature and meteorological forecast data of the bay, a linear regression equation is established according to the following formulas, and w is calculated p1 ,w p2 ,w p3 ,w m1 ,w m2 And
wherein x is t+1-n The measured value of the water temperature on the t +1-n days.
Preferably, the specific process of step S3 is:
s21, adopting water temperature data and meteorological data of the Bay B, carrying out normalization processing on the water temperature data, and dividing the water temperature data and meteorological observation data into a training set and a test set according to the proportion of 0.8: 0.2;
s22, combining meteorological observation data and water temperature data of a plurality of past days to obtain combined data based on a training set part of the water temperature data, and constructing and obtaining a source LSTM neural network prediction model by using the combined data as an input factor and the water temperature data of 7 days in the future as an output factor, wherein the water temperature data of the plurality of past days refers to water temperature data with different time lengths; the combined data refers to the combination of water temperature data serving as fixed input parameters and other meteorological observation elements in a traversing mode;
s23, inputting the combined data into a trained network based on the test set part of the water temperature data, and obtaining a daily frequency water temperature prediction result of 7 days in the future through inverse normalization;
s24, comparing the predicted result with the measured data, and giving an early warning of the water temperature according to the accuracy rate and the root mean squareError RMSE, mean absolute error MAE and coefficient of determination R 2 Counting the test result of the model as a test standard, and selecting an optimal source LSTM neural network prediction model;
s25, setting the initial weight ratio of the optimal source LSTM neural network prediction model and the pure differential regression prediction model and the mixed differential regression prediction model of the B bay to be 1:1:1, taking root mean square error RMSE as a statistical index, carrying out statistics on the prediction accuracy of the three models every day, enabling the weight to be larger when the accuracy is larger, and dynamically updating the weight of the combined model every day.
Preferably, the specific process of step S4 is:
s31, obtaining the optimal source LSTM neural network prediction model of the Bay B constructed in the steps S21 to S24;
s32, adopting the water temperature data and the meteorological data of the A bay to carry out normalization processing on the data of the A bay, and dividing the data into a training set and a testing set according to the proportion of 0.5: 0.5;
s33, freezing a partial network layer of the optimal source LSTM neural network prediction model, and finely adjusting the network based on the training set part of the A bay so as to realize transfer learning of the optimal source LSTM neural network prediction model;
s34, obtaining a daily frequency water temperature prediction result of 7 days in the future by reverse normalization based on the test set part of the gulf A;
s35, comparing the predicted result with the measured data, and determining the accuracy, Root Mean Square Error (RMSE), average absolute error (MAE) and coefficient R of the water temperature early warning level 2 As a test standard, counting the test result of the model, and selecting an optimal migration LSTM neural network prediction model;
s36, setting the initial weight ratio of the optimal migration LSTM neural network prediction model and the pure differential regression prediction model and the mixed differential regression prediction model of the A gulf as 1:1:1, taking root mean square error RMSE as a statistical index, carrying out statistics on the prediction accuracy of the three models every day, enabling the weight to be larger when the accuracy is larger, and dynamically updating the weight of the combined model every day.
Preferably, the calculation formula of the normalization processing in step S21 and step S32 is:
wherein X is the original measured data, X normal Is normalized data, X mean And X std Mean and variance, respectively, for bay B.
Preferably, the denormalization formula in step S23 and step 34 is:
X final =X pre *X std +X mean
wherein, X pre Is the original predicted value, X final Is the final predicted value of inverse normalization, X mean And X std Mean and variance, respectively, for bay B;
the calculation formula of the accuracy of the water temperature early warning grade in the steps S24 and S35 is as follows:
wherein, the division of the water temperature grades refers to the suitable growth temperature of the large yellow croaker, and the total number of the grades is 7;
and testing all the trained models by taking the root mean square error RMSE as the most important index, and selecting the model with the highest precision in the verification set and the test set as the optimal LSTM neural network prediction model or the migration LSTM neural network prediction model.
After adopting the technical scheme, compared with the background technology, the invention has the following advantages: in order to predict the change trend of water temperature, an optimal source LSTM neural network model is constructed by using a B bay with more historical observation data, and an optimal migration LSTM neural network model of an A bay with less historical observation data is constructed by migration learning based on the optimal source LSTM neural network model; in order to accurately forecast the mutation point of the water temperature, the actually measured water temperature and the forecast air temperature data are differentiated, a pure differential regression prediction model is constructed based on the forecast air temperature difference value and the actually measured water temperature value, a mixed differential regression prediction model is constructed based on the forecast air temperature difference value, the actually measured water temperature difference value and the actually measured water temperature value, and the optimal source LSTM neural network model or the optimal migration LSTM neural network model is combined with the pure differential regression prediction model and the mixed differential regression prediction model; the three models are combined into a migration water temperature online prediction model of the gulf A or a source water temperature online prediction model of the gulf B in a dynamic weighting mode, water temperature prediction can be carried out on the gulf with more historical observation data through the source water temperature online prediction model, water temperature prediction can be carried out on the gulf with less historical observation data through the migration water temperature online prediction model, the prediction precision is high, the result is reliable, water temperature values of the gulf in the future 7 days can be well predicted, the target model is helped to obtain knowledge learned by the source model through a migration learning technology to improve the prediction performance of the model, the problem that the water temperature prediction precision is low due to insufficient data volume in the prior art is solved, and data, information and service guarantee can be provided for marine fishery production, scientific research and management.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a block flow diagram of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Referring to fig. 1 to 2, a combined gulf water temperature prediction method based on transfer learning includes the following steps:
s1, collecting water temperature data and meteorological data of an A bay and a B bay respectively, and preprocessing the data respectively, wherein the historical observation data of the B bay is more than that of the A bay;
the water temperature data in the step S1 includes daily maximum water temperature, daily minimum water temperature, and daily average water temperature; the meteorological data comprises meteorological observation data and meteorological forecast data; the meteorological observation data comprise air temperature statistics, air pressure statistics, relative humidity statistics, wind speed statistics, wind direction statistics and precipitation, and the time resolution is 1 day; the weather forecast data comprises future 7-day weather forecast data updated every day, and the future 7-day weather forecast data comprises highest day temperature and lowest day temperature of 7 days in the future;
s2, respectively constructing pure differential regression prediction models of the A bay and the B bay based on the respective forecast air temperature differential values and the actual measurement water temperature values; respectively constructing a mixed differential regression prediction model of the A bay and the B bay based on the respective forecast air temperature difference value, the actual measurement water temperature difference value and the actual measurement water temperature value;
the method for constructing the pure differential regression prediction model in the step S2 includes:
setting the predicted temperature values of the weather forecast released on the k day asMeasured value of water temperature is x k If the current time is t, the predicted value of the water temperature of the day 1 isThe calculation formula of (2) is as follows:
predicted value of water temperature of nth (n is more than or equal to 2 and less than or equal to 7) dayThe calculation formula of (2) is as follows:
wherein x is t-1 The measured value of the water temperature on the t-1 day is measured;the air temperature forecast value of the day is acquired for the t +1-i days;obtaining the air temperature forecast value of the day for the t-i days;an air temperature forecast for the nth day in the future acquired for the t-th day;obtaining the temperature forecast value of the day for the t day; w is a p1 ,w p2 ,w p3 And selecting the weight by constructing a linear regression equation;
the construction method of the mixed difference regression prediction model in the step S2 comprises the following steps:
setting the predicted temperature values of the weather forecast released on the k day asMeasured value of water temperature is x k If the current time is t, the predicted value of the water temperature of the day 1 isThe calculation formula of (2) is as follows:
predicted value of water temperature of nth (n is more than or equal to 2 and less than or equal to 7) dayThe calculation formula of (2) is as follows:
wherein x is t-1 Measured value of water temperature on t-1 day;x t-2 The measured value of the water temperature on the t-2 th day;obtaining the temperature forecast value of the day for the t day;obtaining the air temperature forecast value of the day for the t-1 day;obtaining an air temperature forecast value for the nth day in the future for the t day; w is a m1 ,w m2 Andselecting the weight by constructing a linear regression equation;
selecting w by means of a linear regression equation p1 ,w p2 ,w p3 ,w m1 ,w m2 Andthe specific mode is as follows:
setting the predicted temperature values of the weather forecast released on the k day asMeasured value of water temperature is x k When the current time is t, according to the historical water temperature and meteorological forecast data of the bay, a linear regression equation is established according to the following formulas, and w is calculated p1 ,w p2 ,w p3 ,w m1 ,w m2 And
wherein x is t+1-n The measured value of the water temperature on the t +1-n days is obtained;
s3, aiming at the Bay B, constructing an LSTM neural network model, and selecting an LSTM neural network prediction model of an optimal source through training and testing; combining the optimal source LSTM neural network prediction model, the pure differential regression prediction model of the Bay B and the mixed differential regression prediction model into a source water temperature online prediction model based on a dynamic weighting mode; predicting the water temperature of the bay B according to the source water temperature online prediction model;
the specific process of step S3 is:
s21, adopting water temperature data and meteorological data of the Bay B, carrying out normalization processing on the water temperature data, and dividing the water temperature data and meteorological observation data into a training set and a test set according to the proportion of 0.8: 0.2;
s22, combining meteorological observation data and water temperature data of a plurality of past days to obtain combined data based on a training set part of the water temperature data, and constructing and obtaining a source LSTM neural network prediction model by using the combined data as an input factor and the water temperature data of 7 days in the future as an output factor, wherein the water temperature data of the plurality of past days refers to water temperature data with different time lengths; the combined data is combined with other meteorological observation elements in a traversing mode by taking water temperature data as a fixed input parameter;
s23, inputting the combined data into a trained network based on the test set part of the water temperature data, and obtaining a daily frequency water temperature prediction result of 7 days in the future through inverse normalization;
s24, comparing the predicted result with the measured data, and determining the accuracy, Root Mean Square Error (RMSE), average absolute error (MAE) and coefficient R of the water temperature early warning level 2 Counting the test result of the model as a test standard, and selecting an optimal source LSTM neural network prediction model;
s25, setting the initial weight ratio of the optimal source LSTM neural network prediction model and the pure differential regression prediction model and the mixed differential regression prediction model of the B bay to be 1:1:1, taking root mean square error RMSE as a statistical index, carrying out statistics on the prediction accuracy of the three models every day, enabling the weight to be larger when the accuracy is larger, and dynamically updating the weight of the combined model every day. If the prediction accuracy of the LSTM neural network prediction model is the highest on the next day, the weight of the LSTM neural network prediction model is 1/2 when the weight is (1+1)/(3+1), 1/4 when the weight of the pure differential regression prediction model and the weight of the mixed differential regression prediction model are both 1/(3+1), and the prediction results of the combination model on the next day on the water temperature in the seven days in the future are obtained by multiplying the model prediction values by the corresponding weights and adding the weights; if the prediction accuracy of the mixed differential regression prediction model on the third day is the highest, the weight of the mixed differential regression prediction model is 2/5 (1+1)/(3+2), the weight of the pure differential regression prediction model is 1/5 (1/(3 + 2)), the weight of the LSTM neural network prediction model is 2/5 (2/(3 + 2)), the predicted value of the model is multiplied by the corresponding weights and added, and the prediction result of the combined model on the water temperature on the seventh day in the future on the third day is obtained, and the rest is done;
s4, aiming at the gulf A, selecting the optimal source LSTM neural network prediction model of the gulf B, and selecting the optimal migration LSTM neural network prediction model through migration learning and testing; combining the optimal migration LSTM neural network prediction model, the pure differential regression prediction model of the A bay and the mixed differential regression prediction model into a migration water temperature online prediction model based on a dynamic weighting mode; predicting the water temperature of the gulf A according to the migration water temperature online prediction model;
the specific process of step S4 is:
s31, obtaining the optimal source LSTM neural network prediction model of the Bay B constructed in the steps S21 to S24;
s32, adopting the water temperature data and the meteorological data of the A bay to carry out normalization processing on the data of the A bay, and dividing the data into a training set and a testing set according to the proportion of 0.5: 0.5;
s33, freezing partial network layers of the LSTM neural network prediction model of the optimal source (the LSTM neural network prediction model has two network layers, and the mode of freezing the first network layer is adopted), and finely tuning the network based on the training set part of the A bay so as to realize transfer learning of the LSTM neural network prediction model of the optimal source;
s34, obtaining a daily frequency water temperature prediction result of 7 days in the future by reverse normalization based on the test set part of the gulf A;
s35, comparing the predicted result with the measured data, and determining the accuracy, Root Mean Square Error (RMSE), average absolute error (MAE) and coefficient R of the water temperature early warning level 2 Counting the test result of the model as a test standard, and selecting an optimal migration LSTM neural network prediction model;
s36, setting the initial weight ratio of the optimal migration LSTM neural network prediction model and the pure differential regression prediction model and the mixed differential regression prediction model of the A gulf as 1:1:1, taking root mean square error RMSE as a statistical index, carrying out statistics on the prediction accuracy of the three models every day, enabling the weight to be larger when the accuracy is larger, and dynamically updating the weight of the combined model every day. If the prediction accuracy of the LSTM neural network prediction model is the highest on the next day, the weight of the LSTM neural network prediction model is 1/2 when the weight is (1+1)/(3+1), 1/4 when the weight of the pure differential regression prediction model and the weight of the mixed differential regression prediction model are both 1/(3+1), and the prediction results of the combination model on the next day on the water temperature in the seven days in the future are obtained by multiplying the model prediction values by the corresponding weights and adding the weights; if the prediction accuracy of the mixed differential regression prediction model on the third day is the highest, the weight of the mixed differential regression prediction model is 2/5 (1+1)/(3+2), the weight of the pure differential regression prediction model is 1/5 (1/(3 + 2)), the weight of the LSTM neural network prediction model is 2/5 (2/(3 + 2)), the predicted value of the model is multiplied by the corresponding weights and added, and the prediction result of the combined model on the water temperature on the seventh day in the future on the third day is obtained, and the rest is done;
the calculation formula of the normalization processing in step S21 and step S32 is:
wherein X is the original measured data, X normal Is normalized data, X mean And X std Mean and variance, respectively, for bay B;
the inverse normalization formula in step S23 and step S34 is:
X final =X pre *X std +X mean
wherein, X pre Is the original predicted value, X final Is the final predicted value of inverse normalization, X mean And X std Mean and variance, respectively, for bay B;
the calculation formula of the water temperature early warning grade accuracy in the steps S24 and S35 is as follows:
wherein, the water temperature grades are divided according to the suitable growth temperature of the large yellow croaker, and 7 grades are provided in total, which is shown in table 1;
TABLE 1 Water temperature ratings
The results of the grade prediction are illustrated in table 2.
TABLE 2 grade Warning results Explanation
Note: red low: red low temperature; orange low: orange low temperature; low yellow: yellow and low temperature; yellow height: yellow high temperature;
orange juice: orange high temperature; red height: red high temperature
And testing all the trained models by taking the root mean square error RMSE as the most important index, and selecting the model with the highest precision in the verification set and the test set as the optimal LSTM neural network prediction model or the migration LSTM neural network prediction model.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (9)
1. A combined bay water temperature prediction method based on transfer learning is characterized by comprising the following steps:
s1, collecting water temperature data and meteorological data of an A bay and a B bay respectively, and preprocessing the data respectively, wherein the historical observation data of the B bay is more than that of the A bay;
s2, respectively constructing pure differential regression prediction models of the A bay and the B bay based on the respective forecast air temperature differential values and the actual measurement water temperature values; respectively constructing a mixed differential regression prediction model of the A bay and the B bay based on the respective forecast air temperature differential value, the actual measurement water temperature differential value and the actual measurement water temperature value;
s3, aiming at the Bay B, constructing an LSTM neural network model, and selecting an LSTM neural network prediction model of an optimal source through training and testing; combining the optimal source LSTM neural network prediction model, the pure differential regression prediction model of the Bay B and the mixed differential regression prediction model into a source water temperature online prediction model based on a dynamic weighting mode; predicting the water temperature of the bay B according to the source water temperature online prediction model;
s4, aiming at the gulf A, selecting the optimal source LSTM neural network prediction model of the gulf B, and selecting the optimal migration LSTM neural network prediction model through migration learning and testing; combining the optimal migration LSTM neural network prediction model, the pure differential regression prediction model of the A bay and the mixed differential regression prediction model into a migration water temperature online prediction model based on a dynamic weighting mode; and predicting the water temperature of the A bay according to the online migration water temperature prediction model.
2. The combined gulf water temperature prediction method of claim 1 based on transfer learning, wherein: the water temperature data in the step S1 includes daily maximum water temperature, daily minimum water temperature, and daily average water temperature; the meteorological data comprises meteorological observation data and meteorological forecast data; the meteorological observation data comprise air temperature statistics, air pressure statistics, relative humidity statistics, wind speed statistics, wind direction statistics and precipitation, and the time resolution is 1 day; the weather forecast data includes future 7-day weather forecast data updated every day, and the future 7-day weather forecast data includes day highest temperature and day lowest temperature of the future 7 days.
3. The combined gulf water temperature prediction method based on transfer learning of claim 1, wherein the pure differential regression prediction model in step S2 is constructed by:
setting the predicted temperature values of the weather forecast released on the k day asMeasured value of water temperature is x k If the current time is t, the predicted value of the water temperature of the day 1 isThe calculation formula of (2) is as follows:
predicted value of water temperature of nth (n is more than or equal to 2 and less than or equal to 7) dayIs calculated byThe formula is as follows:
wherein x is t-1 The measured value of the water temperature on the t-1 day is measured;the temperature forecast value of the day is acquired on the t +1-i days;obtaining the air temperature forecast value of the day for the t-i days;obtaining an air temperature forecast value for the nth day in the future for the t day;obtaining the temperature forecast value of the day for the t day; w is a p1 ,w p2 ,w p3 Andand selecting the weight by constructing a linear regression equation.
4. The combined gulf water temperature prediction method based on transfer learning of claim 3, wherein the mixed difference regression prediction model in step S2 is constructed by:
setting the predicted temperature values of the weather forecast released on the k day asMeasured value of water temperature is x k If the current time is t, the predicted value of the water temperature of the day 1 isIs calculated byThe formula is as follows:
predicted value of water temperature of nth (n is more than or equal to 2 and less than or equal to 7) dayThe calculation formula of (2) is as follows:
wherein x is t-1 The measured value of the water temperature on the t-1 day is measured; x is the number of t-2 The measured value of the water temperature on the t-2 days is shown;obtaining the temperature forecast value of the day for the t day;obtaining the air temperature forecast value of the day for the t-1 day;obtaining an air temperature forecast value for the nth day in the future for the t day; w is a m1 ,w m2 Andand selecting the weight by constructing a linear regression equation.
5. The combined gulf water temperature prediction method of claim 4, wherein w is selected by linear regression equation p1 ,w p2 ,w p3 ,w m1 ,w m2 Andthe specific mode is as follows:
setting the predicted temperature values of the weather forecast released on the k day asMeasured value of water temperature is x k When the current time is t, according to the historical water temperature and meteorological forecast data of the bay, a linear regression equation is established according to the following formulas, and w is calculated p1 ,w p2 ,w p3 ,w m1 ,w m2 And
wherein x is t+1-n The measured value of the water temperature on the t +1-n days.
6. The combined gulf water temperature prediction method based on transfer learning of claim 2, wherein the specific process of step S3 is as follows:
s21, adopting water temperature data and meteorological data of the Bay B, carrying out normalization processing on the water temperature data, and dividing the water temperature data and meteorological observation data into a training set and a test set according to the proportion of 0.8: 0.2;
s22, combining meteorological observation data and water temperature data of a plurality of past days to obtain combined data based on a training set part of the water temperature data, and constructing and obtaining a source LSTM neural network prediction model by using the combined data as an input factor and the water temperature data of 7 days in the future as an output factor, wherein the water temperature data of the plurality of past days refers to water temperature data with different time lengths; the combined data is combined with other meteorological observation elements in a traversing mode by taking water temperature data as a fixed input parameter;
s23, inputting the combined data into a trained network based on the test set part of the water temperature data, and obtaining a daily frequency water temperature prediction result of 7 days in the future through inverse normalization;
s24, comparing the predicted result with the measured data, and determining the accuracy, Root Mean Square Error (RMSE), average absolute error (MAE) and coefficient R of the water temperature early warning level 2 Counting the test result of the model as a test standard, and selecting an optimal source LSTM neural network prediction model;
s25, setting the initial weight ratio of the optimal source LSTM neural network prediction model and the pure differential regression prediction model and the mixed differential regression prediction model of the B bay to be 1:1:1, taking root mean square error RMSE as a statistical index, carrying out statistics on the prediction accuracy of the three models every day, enabling the weight to be larger when the accuracy is larger, and dynamically updating the weight of the combined model every day.
7. The combined gulf water temperature prediction method based on transfer learning of claim 6, wherein the specific process of step S4 is as follows:
s31, obtaining the optimal source LSTM neural network prediction model of the Bay B constructed in the steps S21 to S24;
s32, adopting the water temperature data and the meteorological data of the A bay to carry out normalization processing on the data of the A bay, and dividing the data into a training set and a testing set according to the proportion of 0.5: 0.5;
s33, freezing a partial network layer of the optimal source LSTM neural network prediction model, and finely adjusting the network based on the training set part of the A bay so as to realize transfer learning of the optimal source LSTM neural network prediction model;
s34, obtaining a daily frequency water temperature prediction result of 7 days in the future by reverse normalization based on the test set part of the gulf A;
s35, comparing the predicted result with the measured data, and determining the accuracy, Root Mean Square Error (RMSE), average absolute error (MAE) and coefficient R of the water temperature early warning level 2 Counting the test result of the model as a test standard, and selecting an optimal migration LSTM neural network prediction model;
s36, setting the initial weight ratio of the optimal migration LSTM neural network prediction model and the pure differential regression prediction model and the mixed differential regression prediction model of the A gulf as 1:1:1, taking root mean square error RMSE as a statistical index, carrying out statistics on the prediction accuracy of the three models every day, enabling the weight to be larger when the accuracy is larger, and dynamically updating the weight of the combined model every day.
8. The migration learning-based combined bay water temperature prediction method of claim 7, wherein the calculation formula of the normalization process in steps S21 and S32 is:
wherein X is the original measured data, X normal Is normalized data, X mean And X std Mean and variance, respectively, for bay B.
9. The combined gulf water temperature prediction method of claim 7, wherein the denormalization formula in steps S23 and 34 is:
X final =X pre *X std +X mean
wherein, X pre Is the original predicted value, X final Is the final predicted value of the inverse normalization, X mean And X std Mean and variance, respectively, for bay B;
the calculation formula of the water temperature early warning grade accuracy in the steps S24 and S35 is as follows:
wherein, the division of the water temperature grades refers to the suitable growth temperature of the large yellow croaker, and the total number of the grades is 7;
and testing all the trained models by taking the root mean square error RMSE as the most important index, and selecting the model with the highest precision in the verification set and the test set as the optimal LSTM neural network prediction model or the migration LSTM neural network prediction model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210556998.8A CN114862035B (en) | 2022-05-20 | 2022-05-20 | Combined bay water temperature prediction method based on transfer learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210556998.8A CN114862035B (en) | 2022-05-20 | 2022-05-20 | Combined bay water temperature prediction method based on transfer learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114862035A true CN114862035A (en) | 2022-08-05 |
CN114862035B CN114862035B (en) | 2024-05-03 |
Family
ID=82639481
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210556998.8A Active CN114862035B (en) | 2022-05-20 | 2022-05-20 | Combined bay water temperature prediction method based on transfer learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114862035B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115757561A (en) * | 2022-12-09 | 2023-03-07 | 国网湖北省电力有限公司信息通信公司 | Data center equipment operation data time sequence combination prediction method |
CN116128049A (en) * | 2023-04-04 | 2023-05-16 | 厦门大学 | XGBoost model-based migration condition selection method for water quality prediction model |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20190057013A (en) * | 2017-11-17 | 2019-05-27 | 한국외국어대학교 연구산학협력단 | System for managing detailed weather prediction information in real time and method to use for impact forecasting of heat-wave and tropical night using the system |
US20190385055A1 (en) * | 2018-06-14 | 2019-12-19 | Electronics And Telecommunications Research Institute | Method and apparatus for artificial neural network learning for data prediction |
KR20200017604A (en) * | 2018-08-02 | 2020-02-19 | 한국에너지기술연구원 | Energy performance prediction method and energy operation management system by using optimized physical learning model and machine learning methods |
CN112116147A (en) * | 2020-09-16 | 2020-12-22 | 南京大学 | River water temperature prediction method based on LSTM deep learning |
KR102220748B1 (en) * | 2019-12-17 | 2021-02-26 | (주)지오시스템리서치 | Ocean Prediction System Using Future Learning of Water Temperature Time Series Prediction Data Using Deep Learning |
CN113033618A (en) * | 2021-03-03 | 2021-06-25 | 四川大学 | Layered reservoir water taking and discharging water temperature prediction model and prediction method based on support vector regression |
CN114036838A (en) * | 2021-11-09 | 2022-02-11 | 云南大学 | Vertical water temperature simulation method based on multilayer LSTM neural network |
CN114242156A (en) * | 2021-12-17 | 2022-03-25 | 厦门大学 | Real-time prediction method and system for relative abundance of pathogenic vibrios on marine micro-plastic |
CN114372631A (en) * | 2022-01-11 | 2022-04-19 | 电子科技大学 | Data-lacking area runoff prediction method based on small sample learning and LSTM |
-
2022
- 2022-05-20 CN CN202210556998.8A patent/CN114862035B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20190057013A (en) * | 2017-11-17 | 2019-05-27 | 한국외국어대학교 연구산학협력단 | System for managing detailed weather prediction information in real time and method to use for impact forecasting of heat-wave and tropical night using the system |
US20190385055A1 (en) * | 2018-06-14 | 2019-12-19 | Electronics And Telecommunications Research Institute | Method and apparatus for artificial neural network learning for data prediction |
KR20200017604A (en) * | 2018-08-02 | 2020-02-19 | 한국에너지기술연구원 | Energy performance prediction method and energy operation management system by using optimized physical learning model and machine learning methods |
KR102220748B1 (en) * | 2019-12-17 | 2021-02-26 | (주)지오시스템리서치 | Ocean Prediction System Using Future Learning of Water Temperature Time Series Prediction Data Using Deep Learning |
CN112116147A (en) * | 2020-09-16 | 2020-12-22 | 南京大学 | River water temperature prediction method based on LSTM deep learning |
CN113033618A (en) * | 2021-03-03 | 2021-06-25 | 四川大学 | Layered reservoir water taking and discharging water temperature prediction model and prediction method based on support vector regression |
CN114036838A (en) * | 2021-11-09 | 2022-02-11 | 云南大学 | Vertical water temperature simulation method based on multilayer LSTM neural network |
CN114242156A (en) * | 2021-12-17 | 2022-03-25 | 厦门大学 | Real-time prediction method and system for relative abundance of pathogenic vibrios on marine micro-plastic |
CN114372631A (en) * | 2022-01-11 | 2022-04-19 | 电子科技大学 | Data-lacking area runoff prediction method based on small sample learning and LSTM |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115757561A (en) * | 2022-12-09 | 2023-03-07 | 国网湖北省电力有限公司信息通信公司 | Data center equipment operation data time sequence combination prediction method |
CN116128049A (en) * | 2023-04-04 | 2023-05-16 | 厦门大学 | XGBoost model-based migration condition selection method for water quality prediction model |
Also Published As
Publication number | Publication date |
---|---|
CN114862035B (en) | 2024-05-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108491970B (en) | Atmospheric pollutant concentration prediction method based on RBF neural network | |
CN110705743B (en) | New energy consumption electric quantity prediction method based on long-term and short-term memory neural network | |
CN114862035B (en) | Combined bay water temperature prediction method based on transfer learning | |
CN104899135B (en) | Software Defects Predict Methods and system | |
CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
CN111767517B (en) | BiGRU multi-step prediction method, system and storage medium applied to flood prediction | |
CN111753893A (en) | Wind turbine generator power cluster prediction method based on clustering and deep learning | |
CN109919356B (en) | BP neural network-based interval water demand prediction method | |
CN108549770A (en) | The adaptive inversion method of Parameters for Rockfill Dams based on QGA-MMRVM | |
CN112116162A (en) | Power transmission line icing thickness prediction method based on CEEMDAN-QFAOA-LSTM | |
CN114912077B (en) | Sea wave forecasting method integrating random search and mixed decomposition error correction | |
CN112966891A (en) | River water environment quality prediction method | |
CN113537469B (en) | Urban water demand prediction method based on LSTM network and Attention mechanism | |
Li et al. | A k-nearest neighbor locally weighted regression method for short-term traffic flow forecasting | |
CN112307410A (en) | Seawater temperature and salinity information time sequence prediction method based on shipborne CTD measurement data | |
CN111242270A (en) | Time series prediction model based on improved multi-target difference optimization echo state network | |
CN107729988B (en) | Blue algae bloom prediction method based on dynamic deep belief network | |
CN114971005B (en) | Bay water temperature combination prediction method based on LSTM and differential regression model dynamic weighting | |
CN116842358A (en) | Soft measurement modeling method based on multi-scale convolution and self-adaptive feature fusion | |
CN116960962A (en) | Mid-long term area load prediction method for cross-area data fusion | |
Xia et al. | Environmental factor assisted chlorophyll-a prediction and water quality eutrophication grade classification: A comparative analysis of multiple hybrid models based on a SVM | |
CN114781606A (en) | Method for predicting medium and long term electricity consumption in region based on federal learning | |
CN116341705A (en) | Long-period memory network water quality parameter prediction method based on sparse label | |
CN115906669A (en) | Dense residual error network landslide susceptibility evaluation method considering negative sample selection strategy | |
Zhou et al. | Predicting and explaining karst spring dissolved oxygen using interpretable deep learning approach |
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 |