CN107703564B - Rainfall prediction method and system and electronic equipment - Google Patents

Rainfall prediction method and system and electronic equipment Download PDF

Info

Publication number
CN107703564B
CN107703564B CN201710958075.4A CN201710958075A CN107703564B CN 107703564 B CN107703564 B CN 107703564B CN 201710958075 A CN201710958075 A CN 201710958075A CN 107703564 B CN107703564 B CN 107703564B
Authority
CN
China
Prior art keywords
radar
rainfall prediction
time
neural network
rainfall
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201710958075.4A
Other languages
Chinese (zh)
Other versions
CN107703564A (en
Inventor
王永灿
王书强
申妍燕
邓黎明
胡明辉
曹松
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.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
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 Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Priority to CN201710958075.4A priority Critical patent/CN107703564B/en
Publication of CN107703564A publication Critical patent/CN107703564A/en
Application granted granted Critical
Publication of CN107703564B publication Critical patent/CN107703564B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/10—Devices for predicting weather conditions
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/95—Radar or analogous systems specially adapted for specific applications for meteorological use
    • 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
    • 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)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • General Physics & Mathematics (AREA)
  • Environmental & Geological Engineering (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Electromagnetism (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Health & Medical Sciences (AREA)
  • Atmospheric Sciences (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Ecology (AREA)
  • Environmental Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application relates to the technical field of meteorological service, in particular to a rainfall prediction method, a rainfall prediction system and electronic equipment. The rainfall prediction method comprises the following steps: step a: extracting radar maps with different heights in different time periods and meteorological information sequences in corresponding time periods; step b: constructing a rainfall prediction model based on a deep neural network; step c: inputting the extracted radar map and weather information sequence into a rainfall prediction model, carrying out time sequence analysis on the input radar map and weather information sequence through the rainfall prediction model to obtain radar map space-time characteristics and weather information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the weather information time sequence characteristics. The method and the device effectively utilize the long-term and short-term time sequence dependency relationship in the radar chart along with the time variation trend, integrate other meteorological information, and integrate various data information sources, thereby accurately predicting the future rainfall in a short term.

Description

Rainfall prediction method and system and electronic equipment
Technical Field
The application relates to the technical field of meteorological service, in particular to a rainfall prediction method, a rainfall prediction system and electronic equipment.
Background
Precipitation forecasting has been a very important issue in the field of weather services. Accurate weather forecast service can help people plan outdoor activities and can even provide early warning of accidents such as flood or traffic. Short-term precipitation prediction is a real-life significant and challenging task, and in order to predict short-term precipitation, it usually involves analysis of the following information: and finding some modes of precipitation evolution according to historical data by using the relation between the current precipitation and the radar refractive index and the precipitation relation between the target site and the surrounding area.
Doppler radar detection of precipitation is a common method used in weather. In radar weather, the strength of a weather target to the backscattering capability of radar waves is generally called weather target strength, and commonly used parameters for representing the weather target strength include reflectivity and a reflectivity factor. The sum of the backscattering cross sections of the heavy cloud rain particles per unit volume is called the reflectivity of the meteorological target. The sum of the 6 th power of the diameter of precipitation particles per unit volume of a precipitation target, denoted as the radar reflectivity factor, is commonly expressed in units of mm6/m 3. The magnitude of the reflectivity factor Z reflects the size and the number density of precipitation particles inside the meteorological target, and is commonly used for representing the strength of the meteorological target.
In recent years, weather stations have collected a large amount of radar map data and other weather data, and have begun to use radar map data for short-term rainfall predictions. However, the existing short-term rainfall prediction method mainly uses radar map data information with single height, utilizes the traditional optical flow method to extrapolate a radar map along with time, and then carries out manual estimation, which is a relatively conventional short-term rainfall prediction method and also is a basic component of a weather radar system and strong weather early warning service in current weather forecast. However, the prediction efficiency is low, the radar map data information is not sufficiently mined, the prediction accuracy is low, the situation of predicting one hour in the future is still good, but the prediction is very poor for the next second hour and the third hour, and the current prediction requirement is seriously difficult to meet.
In recent years, deep learning subverts algorithm design ideas in many fields such as image recognition, text understanding, voice recognition and the like, and a new mode of directly outputting a final result by training an end-to-end deep neural network model from training data is gradually formed. Each layer in the deep learning can adjust itself for the final task, and finally the universal cooperation among the layers is realized, so that the accuracy of the task can be greatly improved. With the advent of the big data age and the development of various more powerful computing devices such as a GPU (Graphics Processing Unit), deep learning can fully and automatically learn abstract knowledge expression by fully utilizing various massive data, i.e., concentrating original data into certain knowledge.
The Chinese patent application (application number CN 201610752369) provides a rainfall prediction method and a rainfall prediction system, which utilize wind speed, temperature, air pressure and humidity data of each meteorological automatic station to set time scale intervals and prediction delay time, and establish a multi-time scale SVM rainfall dislocation prediction model based on the meteorological data corresponding to the time scale intervals; calculating a model parameter gamma value and a parameter C value, inputting the model into an SVM rainfall dislocation prediction model to obtain a rainfall prediction result corresponding to the prediction delay time, and calculating the rainfall probability according to the rainfall prediction result. However, the technology mainly aims at wind speed, temperature, air pressure and humidity data, and is lack of mining and utilization of effective information of radar maps, meanwhile, because the SVM solves the support vector by means of quadratic programming, and the solving of the quadratic programming involves the calculation of an m-order matrix (m is the number of samples), and when the number of m is large, the storage and calculation of the matrix consume a large amount of machine memory and operation time.
Guozotan, Shoda, Yuanyuan and the like of the computer school of Beijing postal university provide a prediction model based on a neural network in a short-time rainfall prediction method based on neural network and model integration, and the rainfall probability in a certain area within 36min can be predicted through a Doppler radar image sequence. Through the comparative analysis of the neural network and the traditional optical flow method, an integrated prediction model combining the advantages of the two methods is also provided. The method uses a radar chart sequence with single height to train a convolutional neural network so as to predict whether a target pixel point rains within 36min in the future, namely a binary classification problem. Although radar map information and a convolutional neural network are utilized, the method has limited mining on the time-space information, only uses a single-height radar map without considering the mutual motion change relationship of cloud layers with different heights, only extracts the spatial characteristic information insufficiently in the planar convolution, and extracts the timing information insufficiently when the radar maps with different timing sequences are used as the multi-channel input of the convolutional neural network. And the final prediction is only whether rainfall occurs within 36 minutes in the future, the prediction time is short, and the prediction granularity is coarse.
Disclosure of Invention
The application provides a rainfall prediction method, a rainfall prediction system and electronic equipment, and aims to solve at least one of the technical problems in the prior art to a certain extent.
In order to solve the above problems, the present application provides the following technical solutions:
a rainfall prediction method comprising:
step a: extracting radar maps with different heights in different time periods and meteorological information sequences in corresponding time periods;
step b: constructing a rainfall prediction model based on a deep neural network;
step c: inputting the extracted radar map and weather information sequence into a rainfall prediction model, carrying out time sequence analysis on the input radar map and weather information sequence through the rainfall prediction model to obtain radar map space-time characteristics and weather information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the weather information time sequence characteristics.
The technical scheme adopted by the embodiment of the application further comprises the following steps: in the step a, the extracting radar maps with different heights in different time periods and the meteorological information sequence in the corresponding time period specifically includes: extracting radar maps with different heights in different time periods, overlapping the radar maps with the different heights in each time period respectively to obtain radar pattern data corresponding to each time period, and setting a sample label of each radar pattern data according to rainfall of each time period; and acquiring a weather information sequence in a time period corresponding to each radar pattern sample data, and processing the extracted weather information sequence into a vector form.
The technical scheme adopted by the embodiment of the application further comprises the following steps: in the step b, the constructing a rainfall prediction model based on a deep neural network further comprises: inputting the radar pattern data and the meteorological information sequence into a rainfall prediction model based on a deep neural network, and training the rainfall prediction model based on the deep neural network through a back propagation algorithm.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the rainfall prediction model comprises a 3D convolutional neural network and a long-short term memory cyclic neural network, and the training of the rainfall prediction model based on the deep neural network through a back propagation algorithm specifically comprises the following steps:
step b 1: initializing parameters of the rainfall prediction model based on the deep neural network;
step b 2: inputting radar pattern sample data with sample labels and a meteorological information sequence into a rainfall prediction model based on a deep neural network;
step b 3: carrying out convolution pooling operation on the radar pattern data of each time period through a 3D convolution neural network, extracting radar map spatial characteristic information of each time period, and inputting the extracted radar map spatial characteristic information of each time period and a weather information sequence into a long-term and short-term memory recurrent neural network;
step b 4: and extracting radar map space-time characteristics and meteorological information time sequence characteristics through a long-short term memory cyclic neural network, and outputting rainfall prediction values of all time periods according to the radar map space-time characteristics and the meteorological information time sequence characteristics.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the training of the rainfall prediction model based on the deep neural network through the back propagation algorithm further comprises:
step b 5: calculating a loss function according to the output rainfall prediction value and the rainfall error in the sample label;
step b 6: judging whether the loss function meets the minimum value or not, and if not, executing the step b 7; if the minimum value is met, performing step b 8;
step b 7: adjusting network parameters according to the loss function by using a back propagation algorithm;
step b 8: and saving the network parameters, and finishing the model training.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the rainfall prediction model based on the deep neural network comprises an input layer, a first convolution layer, a correction layer, a pooling layer, a second convolution layer, a full-connection layer and an output layer; the method comprises the steps that radar pattern data and meteorological information sequences are input through an input layer, convolution operation is conducted on the input radar pattern data through a first convolution layer, correction and pooling processing are conducted through a correction layer and a pooling layer, radar pattern space characteristic information corresponding to the radar pattern data is output through a second convolution layer, the second convolution layer is fully connected to a long-short term memory cyclic neural network through a full connection layer, radar pattern space-time characteristics and meteorological information time sequence characteristics are extracted through the long-short term memory cyclic neural network according to the radar pattern space characteristic information and the meteorological information sequences, and rainfall prediction values are output through an output layer.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the long and short term memory cyclic neural network comprises a forgetting gate, an input gate and an output gate; the forgetting gate is used for selectively memorizing and forgetting long-term memory of radar map spatial characteristic information and meteorological information sequences; the input gate is used for storing effective information in the radar chart and weather information sequence input at the current moment and adding the effective information into the long-term memory of the radar chart and the weather information sequence; the output gate is used for acquiring information which can have instant influence on the next moment from the accumulated radar chart and meteorological information sequence.
Another technical scheme adopted by the embodiment of the application is as follows: a rainfall prediction system comprising:
a radar map extraction module: the method comprises the steps of extracting radar maps with different heights in different time periods;
weather information extraction module: the weather information sequence is used for extracting the weather information sequence in the time period corresponding to the radar chart;
a network model construction module: the rainfall prediction model is used for constructing a rainfall prediction model based on a deep neural network;
a rainfall prediction module: the rainfall prediction model is used for inputting the extracted radar map and meteorological information sequence into the rainfall prediction model, performing time sequence analysis on the input radar map and meteorological information sequence through the rainfall prediction model to obtain radar map space-time characteristics and meteorological information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the meteorological information time sequence characteristics.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the radar map extraction module extracts radar maps with different heights in different time periods, and specifically comprises the following steps: extracting radar maps with different heights in different time periods, overlapping the radar maps with the different heights in each time period respectively to obtain radar pattern data corresponding to each time period, and setting a sample label of each radar pattern data according to rainfall of each time period; the meteorological information extraction module extracts a meteorological information sequence in a time period corresponding to the radar chart, and the meteorological information sequence is specifically as follows: and acquiring a weather information sequence in a time period corresponding to each radar pattern sample data, and processing the extracted weather information sequence into a vector form.
The technical scheme adopted by the embodiment of the application further comprises a network model training module, wherein the network model training module is used for inputting the radar pattern data and the meteorological information sequence into a rainfall prediction model based on the deep neural network, and the rainfall prediction model based on the deep neural network is trained through a back propagation algorithm.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the rainfall prediction model comprises a 3D convolutional neural network and a long-short term memory cycle neural network, and the network model training module comprises:
an initialization unit: parameters for initializing the deep neural network-based rainfall prediction model;
a sample input unit: the rainfall prediction model is used for inputting the radar pattern data with the sample labels and the meteorological information sequence into the rainfall prediction model based on the deep neural network;
a first feature extraction unit: the device is used for respectively carrying out convolution pooling operation on the radar pattern data of each time period through a 3D convolution neural network, extracting radar map spatial characteristic information of each time period, and inputting the extracted radar map spatial characteristic information and meteorological information sequence of each time period into a long-short term memory recurrent neural network;
a second feature extraction unit: the method is used for extracting the radar map time-space characteristics and the meteorological information time sequence characteristics through the long-short term memory cyclic neural network and outputting rainfall prediction values of all time periods according to the radar map time-space characteristics and the meteorological information time sequence characteristics.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the network model training module further comprises:
a loss function calculation unit: the loss function is calculated according to the output rainfall prediction value and the rainfall error in the sample label;
loss function value judging means: the device is used for judging whether the loss function meets the minimum value or not, and optimizing network parameters through a network optimization unit if the loss function does not meet the minimum value; if the minimum value is met, storing the network parameters through a parameter storage unit;
a network optimization unit: the system is used for adjusting network parameters according to the loss function by applying a back propagation algorithm;
a parameter storage unit: and storing the network parameters, and finishing the model training.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the rainfall prediction model based on the deep neural network comprises an input layer, a first convolution layer, a correction layer, a pooling layer, a second convolution layer, a full-connection layer and an output layer; the method comprises the steps that radar pattern data and meteorological information sequences are input through an input layer, convolution operation is conducted on the input radar pattern data through a first convolution layer, correction and pooling processing are conducted through a correction layer and a pooling layer, radar pattern space characteristic information corresponding to the radar pattern data is output through a second convolution layer, the second convolution layer is fully connected to a long-short term memory cyclic neural network through a full connection layer, radar pattern space-time characteristics and meteorological information time sequence characteristics are extracted through the long-short term memory cyclic neural network according to the radar pattern space characteristic information and the meteorological information sequences, and rainfall prediction values are output through an output layer.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the long and short term memory cyclic neural network comprises a forgetting gate, an input gate and an output gate; the forgetting gate is used for selectively memorizing and forgetting long-term memory of radar map spatial characteristic information and meteorological information sequences; the input gate is used for storing effective information in the radar chart and weather information sequence input at the current moment and adding the effective information into the long-term memory of the radar chart and the weather information sequence; the output gate is used for acquiring information which can have instant influence on the next moment from the accumulated radar chart and meteorological information sequence.
The embodiment of the application adopts another technical scheme that: an electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the one processor to cause the at least one processor to perform the following operations of the rainfall prediction method described above:
step a: extracting radar maps with different heights in different time periods and meteorological information sequences in corresponding time periods;
step b: constructing a rainfall prediction model based on a deep neural network;
step c: inputting the extracted radar map and weather information sequence into a rainfall prediction model, carrying out time sequence analysis on the input radar map and weather information sequence through the rainfall prediction model to obtain radar map space-time characteristics and weather information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the weather information time sequence characteristics.
Compared with the prior art, the embodiment of the application has the advantages that: according to the rainfall prediction method, the rainfall prediction system and the electronic equipment, radar maps with different heights are used, the cloud layer information of a target area and a surrounding area is extracted on a radar map plane with a single height by using the 3D convolutional neural network, the relation between the radar maps with adjacent heights is learned, and the spatial characteristic information of the radar maps is automatically extracted; then, the time sequence analysis is carried out on the change trend of the radar map spatial characteristic information, the weather information sequence and the historical rainfall information through a long-term and short-term memory cyclic neural network, the long-term and short-term time sequence dependency relationship in the radar map change trend along with time is effectively utilized, other weather information is integrated, various data information sources are integrated, the purpose of advantage complementation is achieved, the model can be better established on the change trend along with time and the relation between the radar map and the rainfall, the error is further reduced, and therefore the future rainfall in the short term is accurately predicted; the method is time-saving, labor-saving, efficient and quick, and overcomes the defect that the prediction error is increased due to information loss caused by the utilization of a single-height radar map in the traditional method.
Drawings
FIG. 1 is a flow chart of a rainfall prediction method of an embodiment of the present application;
FIG. 2 is a radar plot a time spans before time t;
FIG. 3 is a schematic diagram of a destination site;
FIG. 4 is a diagram of a network architecture of a first implementation;
FIG. 5 is a diagram of a network architecture of a second implementation;
FIG. 6 is a schematic diagram of a 3-gate structure of an embodiment of the present application;
FIG. 7 is a schematic diagram of a rainfall prediction system according to an embodiment of the present application;
fig. 8 is a schematic structural diagram of hardware devices of a rainfall prediction method according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application 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 present application and are not intended to limit the present application.
In the prior art, the traditional short-term rainfall prediction method based on radar maps only selects radar maps with a single height for analysis, but because the radar maps with different heights in a certain time period have a plurality of radar maps with different heights and the radar maps with different heights have a mutual variation relationship, the radar maps with the single height cause a great amount of information loss and cause large prediction errors. In order to solve the problems in the prior art, the rainfall prediction method provided by the embodiment of the application superposes a plurality of radar map data of different heights in the same time period, uses a 3D convolutional neural network to extract cloud layer information of a target area and a peripheral area on a radar map plane of a single height, simultaneously learns the relation between radar maps of adjacent heights, and automatically extracts radar map spatial feature information; and extracting the time-space characteristics and meteorological information time sequence characteristics of the radar chart through a long-term short-term memory cyclic neural network, so that the future rainfall in a short term is accurately predicted.
Specifically, please refer to fig. 1, which is a flowchart illustrating a rainfall prediction method according to an embodiment of the present application. The rainfall prediction method comprises the following steps:
step 100: extracting radar maps with different heights in different time periods from meteorological data, superposing a plurality of radar maps with different heights in each time period respectively to obtain radar pattern data corresponding to each time period, and setting a sample label of each radar pattern data according to rainfall of each time period;
in step 100, the embodiment of the present application predicts the total ground rainfall of the target site in a certain future time period by selecting a series of radar maps of the same target site and the surrounding area at different heights in different time periods and processing the radar maps into a designated size and dimension. According to the method, the radar maps distributed at different heights are fully utilized, not only are convolution carried out on the radar map plane at a single height, cloud layer information of a target area and a peripheral area is extracted, meanwhile, convolution is carried out on the radar map height dimension, the relation of the radar maps between adjacent heights is learned, radar map data are fully utilized, information loss caused by the fact that the radar maps at the single height are utilized in the traditional method is avoided, and therefore prediction accuracy is improved. Specifically, the radar pattern data includes the following dimensions:
1. each radar pattern data comprises a target station (the target station is positioned in the center of the radar pattern data);
2. each radar pattern data comprises the total rainfall of the target station in a certain time period;
3. each radar pattern datum is formed by superposing a radar patterns with a time spans, the interval time of the radar patterns under different time spans is 6 minutes, and the radar pattern datum can be set according to actual application; radar maps at different heights, and the total height is H; as shown in fig. 2, the single radar pattern sample data is a radar map of a time spans a before time t, sample labels of the radar pattern sample data are total rainfall amounts from (t + i) to (t + i +1), and i is a time period after time t.
4. According to the latitude and longitude of a target station, the occupied area of each radar pattern data is m × m square kilometers, and the area is marked as m × m grids; specifically, as shown in fig. 3, it is a schematic diagram of a target station, and a point at the center of the diagram is the target station.
Step 200: acquiring a meteorological information sequence in a time period corresponding to each radar pattern book data to obtain a training sample data set consisting of the radar pattern book data and the corresponding meteorological information sequence;
in step 200, in order to keep consistent with the radar pattern data and facilitate processing, a meteorological information sequence such as wind speed, temperature, air pressure, humidity and the like corresponding to the time period of the radar pattern data is extracted from meteorological information collected by a meteorological station, and the extracted meteorological information sequence is processed into a vector form.
Step 300: constructing a rainfall prediction model based on a deep neural network;
in step 300, the rainfall prediction model based on the deep neural network according to the embodiment of the present application includes two parts, namely, a 3D convolutional neural network and a long-short term memory cycle neural network, and the rainfall prediction model based on the deep neural network includes two implementation manners, as shown in fig. 4, which is a network architecture diagram of the first implementation manner. The first realization mode is that radar pattern sample data and meteorological information sequence are completely separated: the method comprises the steps of firstly extracting radar map space-time characteristics in radar pattern data through a 3D convolutional neural network and a long-short term memory neural network, and predicting first rainfall according to the extracted radar map space-time characteristics; then extracting the time sequence characteristics of the meteorological information sequence through the long short-term memory neural network, and predicting second rainfall according to the meteorological information time sequence characteristics; and finally, performing weight fusion on the two predicted rainfall quantities to obtain a final rainfall quantity predicted value.
Fig. 5 is a diagram of a network architecture according to a second implementation manner. The second implementation manner is as follows: and performing convolution operation on the radar pattern data through a 3D convolution neural network to obtain radar map spatial characteristic information of each time period, inputting the radar map spatial characteristic information and a weather information sequence into a long-short term memory cyclic neural network for time sequence analysis to obtain radar map space-time characteristics and weather information time sequence characteristics, and obtaining a final rainfall prediction value by integrating the cloud layer distribution and the variation trend of the target area and the surrounding area and the variation trend of the weather information. Since the two implementations are substantially similar in implementation details, the following embodiments of the present application are specifically described only by taking the second implementation as an example, and the specific implementation can be selected according to actual situations.
Specifically, please refer to fig. 4, which is a general architecture diagram of a rainfall prediction model based on a deep neural network according to an embodiment of the present application. The rainfall prediction model based on the deep neural network comprises the following steps: the device comprises an input layer, a first convolution layer, a correction layer, a pooling layer, a second convolution layer, a full-connection layer and an output layer; the method comprises the steps of inputting radar pattern data and meteorological information sequences through an input layer, performing convolution operation on the radar pattern data of each time period through a first convolution layer, performing correction and pooling processing through a correction layer and a pooling layer, outputting radar pattern spatial characteristic information of each time period through a second convolution layer, fully connecting to a long-short term memory cyclic neural network through a full connection layer, learning radar pattern spatial characteristic information and meteorological information sequence transformation trend along with time through the long-short term memory cyclic neural network, obtaining radar pattern space-time characteristics and meteorological information time sequence characteristics of each time period, and finally outputting rainfall prediction values through an output layer.
Further, in the embodiment of the present application, the long-short term memory recurrent neural network includes a forgetting gate (forgetgate), an input gate (input gate), and an output gate (output gate). The gate is a method for selectively passing information, and the radar map spatial characteristic information and the meteorological information sequence can be removed or added to the cell state of the long-short term memory neural network through 3 gate structures, namely the radar map spatial characteristic information and the meteorological information sequence can be selectively forgotten and stored through the 3 gate structures. The method comprises the steps of selecting effective information in radar map spatial characteristic information and weather information sequences in real time through a forgetting gate, discarding ineffective information, adding new radar map spatial characteristic information and weather information sequences through an input gate, and repeating in a cyclic mode, so that a long-term dependence information and a short-term dependence information in the radar map spatial characteristic information and the weather information sequences can be learned through a long-term and short-term memory cyclic neural network, the change trend of the radar map spatial characteristic information and the weather information sequences is mined, and prediction errors are reduced.
Specifically, please refer to fig. 6, which is a schematic diagram of a 3-gate structure according to an embodiment of the present application. The functions of the forgetting gate, the input gate and the output gate are as follows:
1. forget the door: selective memory and forgetting for long-term memory of radar map spatial characteristic information and meteorological information sequences; because each radar pattern data and weather information sequence comprises a radar map and a weather information sequence at the current time and a plurality of past times, but some information extracted from the radar map and the weather information sequence at the plurality of past times can become invalid information along with the development of time, the invalid information needs to be discarded through a forgetting gate, and the information still valid in the radar map and the weather information sequence at the plurality of past times is kept. The forgetting gate inputs and outputs the previous moment according to the radar chart and the weather information sequence at the current moment (namely, the information which can generate instant influence on the current moment in the radar chart and the weather information sequence accumulated in the past)
Figure BDA0001432926340000161
) Outputting a forgetting ginseng between 0 and 1Number of
Figure BDA0001432926340000162
Wherein 0 represents that the radar map and weather information sequence accumulated in the past is completely forgotten, and 1 represents that the radar map and weather information sequence accumulated in the past is completely memorized.
2. An input gate: used for storing effective information in the radar chart and meteorological information sequence input at the current moment and adding the effective information into the long-term memory C of the radar chart and meteorological information sequencetIn (1). In the embodiment of the application, the input gate inputs the radar chart and the weather information sequence at the current moment and outputs the radar chart and the weather information sequence at the last moment (namely, information which can generate instant influence on the current moment in the radar chart and the weather information sequence accumulated in the past)
Figure BDA0001432926340000163
) Determining the influence degree i of the radar map and the meteorological information sequence at the current moment on the whole radar map and the meteorological information sequencetThen forgetting parameters obtained from the forgetting gate
Figure BDA0001432926340000164
Extraction of long-term memory CtThe effective information of radar chart and weather information sequence in (1), the influence degree i obtained by the input gatetExtracting useful information for long-term memory of radar chart and meteorological information sequence in radar chart and meteorological information sequence input at current moment, and updating long-term memory C of radar chart and meteorological information sequencet。
3. An output gate for obtaining information from the accumulated radar map and weather information sequence that will have an immediate effect on the next time instant
Figure BDA0001432926340000171
Long-term memory C of output gate for radar chart and meteorological information sequence based on current momenttAnd the radar chart and the weather information sequence input at the current moment, namely the long-term memory C of the radar chart and the weather information sequencetReflects the whole variation trend of the radar chart and the meteorological information sequence extracted according to the previous information and the current momentThe input radar chart and weather information sequence can reflect the expectation of the radar chart and the weather information sequence at the next moment to a certain extent.
Step 400: inputting radar pattern sample data and meteorological information sequences in the training sample data set into a rainfall prediction model based on a deep neural network, and training the rainfall prediction model based on the deep neural network through a back propagation algorithm;
in step 400, the training method for training the rainfall prediction model based on the deep neural network includes the following steps:
step 410: initializing parameters of a rainfall prediction model based on a deep neural network;
step 420: inputting radar pattern data with rainfall tags and meteorological information sequences into a rainfall prediction model based on a deep neural network;
step 430: continuously convolving the radar pattern data of each time period through a 3D convolutional neural network, and extracting radar pattern space characteristic information of each time period;
step 440: inputting the extracted radar map space characteristic information and meteorological information sequence of each time period into a long-short term memory cyclic neural network, extracting radar map space-time characteristics and meteorological information time sequence characteristics through the long-short term memory cyclic neural network, and outputting rainfall prediction values of each time period according to the radar map space-time characteristics and the meteorological information time sequence characteristics;
step 450; calculating a loss function L according to the output rainfall prediction value and the rainfall error in the sample label;
step 460: judging whether the loss function L meets the minimum value, if not, executing step 470; if the minimum is met, go to step 480;
in step 460, the application uses Root Mean Square Error (RMSE) to evaluate performance as a loss function, since the objective of the application is to predict the exact amount of rainfall to reduce the prediction error; RMSE is defined as follows:
Figure BDA0001432926340000181
in the above formula, X is a prediction vector, Y is a vector of an observed value, and n is an observed data size.
Step 470: adjusting network parameters according to the loss function L by using a back propagation algorithm until the loss function L meets the minimum value;
step 480: and saving the network parameters, and finishing the model training.
Step 500: and performing short-term rainfall prediction by using a trained rainfall prediction network model based on the deep neural network.
In step 500, to predict the rainfall in a certain time period (i +1 th time) in the future of the current time of a certain target area, only a radar map and a weather information sequence of a time span within m x m range before the current time are extracted by taking the target area as the center, the radar map and the weather information sequence are processed into the same form as the training sample data set, then a rainfall prediction model based on a deep neural network is input, the extraction of the space-time characteristics of the radar map, the modeling of the transformation trend of the radar map on the space and time domain, the modeling of the relation between the radar map and the corresponding rainfall, the modeling of the change trend of the weather information along with time and the modeling process of the relation between the weather information and the corresponding rainfall are automatically completed through the rainfall prediction model based on the deep neural network, and the rainfall prediction value in the time period from the i to the i +1 th time after the current time is directly output, the manual estimation process is omitted, time and labor are saved, and the method is convenient and easy to use.
Please refer to fig. 7, which is a schematic structural diagram of a rainfall prediction system according to an embodiment of the present application. The rainfall prediction system comprises a radar map extraction module, a meteorological information extraction module, a network model construction module, a network model training module and a rainfall prediction module.
A radar map extraction module: the radar pattern data acquisition system is used for extracting radar patterns with different heights in different time periods from meteorological data, and overlapping a plurality of radar patterns with different heights in each time period to obtain radar pattern data corresponding to each time period; the method and the device for predicting the total ground rainfall of the target station in a certain time period in the future predict the total ground rainfall of the target station by selecting a series of radar maps of the same target station and surrounding areas at different heights in different time periods and processing the radar maps into specified sizes and dimensions. Specifically, the radar pattern data includes the following dimensions:
1. each radar pattern data comprises a target station (the target station is positioned in the center of the radar pattern data);
2. each radar pattern data comprises the total rainfall of the target station in a certain time period;
3. each radar pattern datum is formed by superposing a radar patterns with a time spans, the interval time of the radar patterns under different time spans is 6 minutes, and the radar pattern datum can be set according to actual application; radar maps at different heights, and the total height is H; as shown in fig. 2, the single radar pattern sample data is a radar map of a time spans a before time t, sample labels of the radar pattern sample data are total rainfall amounts from (t + i) to (t + i +1), and i is a time period after time t.
4. According to the latitude and longitude of a target station, the occupied area of each radar pattern data is m × m square kilometers, and the area is marked as m × m grids; specifically, as shown in fig. 3, it is a schematic diagram of a target station, and a point at the center of the diagram is the target station.
Weather information extraction module: the system comprises a time period acquisition module, a data acquisition module and a data processing module, wherein the time period acquisition module is used for acquiring a weather information sequence in a time period corresponding to each radar pattern data; in order to keep consistent with the radar pattern data and facilitate processing, a weather information sequence such as wind speed, temperature, air pressure, humidity and the like corresponding to the radar pattern data time period is extracted from weather information collected by a weather station, and the extracted weather information sequence is processed into a vector form.
A network model construction module: the rainfall prediction model is used for constructing a rainfall prediction model based on a deep neural network; the rainfall prediction model based on the deep neural network comprises a 3D convolutional neural network and a long-short term memory cyclic neural network, and comprises two implementation modes, wherein the first implementation mode is that radar pattern data is completely separated from a meteorological information sequence: the method comprises the steps of firstly extracting radar map space-time characteristics in radar pattern data through a 3D convolutional neural network and a long-short term memory neural network, and predicting first rainfall according to the extracted radar map space-time characteristics; then extracting the time sequence characteristics of the meteorological information sequence through the long short-term memory neural network, and predicting second rainfall according to the meteorological information time sequence characteristics; and finally, performing weight fusion on the two predicted rainfall quantities to obtain a final rainfall quantity predicted value.
The second implementation manner is as follows: and performing convolution operation on the radar pattern data through a 3D convolution neural network to obtain radar map spatial characteristic information of each time period, inputting the radar map spatial characteristic information and a weather information sequence into a long-short term memory cyclic neural network for time sequence analysis to obtain radar map space-time characteristics and weather information time sequence characteristics, and obtaining a final rainfall prediction value by integrating the cloud layer distribution and the variation trend of the target area and the surrounding area and the variation trend of the weather information. Since the two implementations are substantially similar in implementation details, the following embodiments of the present application are specifically described only by taking the second implementation as an example.
Specifically, the invention provides an overall architecture diagram of a rainfall prediction model based on a deep neural network according to an embodiment of the application. The rainfall prediction model based on the deep neural network comprises the following steps: the device comprises an input layer, a first convolution layer, a correction layer, a pooling layer, a second convolution layer, a full-connection layer and an output layer; the method comprises the steps of inputting radar pattern data and meteorological information sequences through an input layer, performing convolution operation on the radar pattern data of each time period through a first convolution layer, performing correction and pooling processing through a correction layer and a pooling layer, outputting radar pattern space characteristic information of each time period through a second convolution layer, fully connecting to a long-short term memory cyclic neural network through a full connecting layer, extracting radar pattern space-time characteristics and meteorological information time sequence characteristics of each time period through the long-short term memory cyclic neural network, and finally outputting rainfall prediction values through an output layer.
Further, in the embodiment of the present application, the long-short term memory recurrent neural network includes a forgetting gate (forgetgate), an input gate (input gate), and an output gate (output gate). The gate is a method for selectively passing information, and the radar map spatial characteristic information and the meteorological information sequence can be removed or added to the cell state of the long-short term memory neural network through 3 gate structures, namely the radar map spatial characteristic information and the meteorological information sequence can be selectively forgotten and stored through the 3 gate structures. The method comprises the steps of selecting effective information in radar map spatial characteristic information and weather information sequences in real time through a forgetting gate, discarding ineffective information, adding new radar map spatial characteristic information and weather information sequences through an input gate, and repeating in a cyclic mode, so that a long-term dependence information and a short-term dependence information in the radar map spatial characteristic information and the weather information sequences can be learned through a long-term and short-term memory cyclic neural network, the change trend of the radar map spatial characteristic information and the weather information sequences is mined, and prediction errors are reduced. Specifically, the forgetting gate, the input gate, and the output gate function as follows:
1. forget the door: selective memory and forgetting for long-term memory of radar map spatial characteristic information and meteorological information sequences; because each radar pattern data and weather information sequence comprises a radar map and a weather information sequence at the current time and a plurality of past times, but some information extracted from the radar map and the weather information sequence at the plurality of past times can become invalid information along with the development of time, the invalid information needs to be discarded through a forgetting gate, and the information still valid in the radar map and the weather information sequence at the plurality of past times is kept. The forgetting gate outputs a forgetting parameter between 0 and 1 according to the input of the radar chart and the weather information sequence at the current moment and the output at the previous moment (namely the information which can generate instant influence on the current moment in the past accumulated radar chart and the weather information sequence)
Figure BDA0001432926340000221
Wherein 0 represents that the radar map and weather information sequence accumulated in the past is completely forgotten, and 1 represents that the radar map and weather information sequence accumulated in the past is completely memorized.
2. An input gate: radar for saving input at current timeAdding effective information in the chart and weather information sequence into long-term memory C of the radar chart and weather information sequencetIn (1). In the embodiment of the application, the input gate determines the influence degree i of the radar map and the weather information sequence at the current moment on the whole radar map and the weather information sequence through the input of the radar map and the weather information sequence at the current moment and the output of the previous moment (namely, the information which can generate instant influence on the current moment in the radar map and the weather information sequence accumulated in the past)tThen forgetting parameters obtained from the forgetting gate
Figure BDA0001432926340000231
Extraction of long-term memory CtThe effective information of radar chart and weather information sequence in (1), the influence degree i obtained by the input gatetExtracting useful information for long-term memory of radar chart and meteorological information sequence in radar chart and meteorological information sequence input at current moment, and updating long-term memory C of radar chart and meteorological information sequencet。
3. An output gate for obtaining information from the accumulated radar map and weather information sequence that will have an immediate effect on the next time instant
Figure BDA0001432926340000232
Long-term memory C of output gate for radar chart and meteorological information sequence based on current momenttAnd the radar chart and the weather information sequence input at the current moment, namely the long-term memory C of the radar chart and the weather information sequencetThe overall change trend of the radar chart and the weather information sequence extracted according to the previous information is reflected, and the expectation of the radar chart and the weather information sequence at the next moment can be reflected to a certain extent by adding the radar chart and the weather information sequence input at the current moment.
A network model training module: the rainfall prediction model based on the deep neural network is trained through a back propagation algorithm; specifically, the network model training module comprises:
an initialization unit: parameters for initializing a deep neural network-based rainfall prediction model;
a sample input unit: the rainfall prediction model is used for inputting radar pattern data with rainfall tags and meteorological information sequences into the rainfall prediction model based on the deep neural network;
a first feature extraction unit: the device is used for respectively carrying out continuous convolution pooling operation on the radar pattern data of each time period through a 3D convolution neural network, and extracting radar pattern space characteristic information of each time period;
a second feature extraction unit: the long and short term memory cyclic neural network is used for inputting the extracted radar map space characteristic information and meteorological information sequence of each time period into the long and short term memory cyclic neural network, extracting radar map space-time characteristics and meteorological information time sequence characteristics through the long and short term memory cyclic neural network, and outputting rainfall prediction values of each time period according to the radar map space-time characteristics and the meteorological information time sequence characteristics;
a loss function calculation unit: the loss function L is calculated according to the output rainfall prediction value and the rainfall error in the sample label;
loss function value judging means: the device is used for judging whether the loss function L meets the minimum value or not, and optimizing network parameters through a network optimization unit if the loss function L does not meet the minimum value; if the minimum value is met, storing the network parameters through a parameter storage unit; wherein, since the objective of the present application is to predict the exact amount of rainfall to reduce the prediction error, the present application uses the Root Mean Square Error (RMSE) as a loss function to evaluate the performance; RMSE is defined as follows:
Figure BDA0001432926340000241
in the above formula, X is a prediction vector, Y is a vector of an observed value, and n is an observed data size.
A network optimization unit: the method is used for adjusting network parameters according to the loss function L by applying a back propagation algorithm until the loss function L meets the minimum value;
a parameter storage unit: and storing the network parameters, and finishing the model training.
A rainfall prediction module: the method is used for performing short-term rainfall prediction by using a trained rainfall prediction network model based on the deep neural network. Wherein, to predict the rainfall of a certain time period (i time to i +1 time) in the future of the current time of a certain target area, only a radar map and a meteorological information sequence of a time span within a range of a few hours m x m before the current time are extracted by taking the target area as the center, the radar map and the meteorological information sequence are processed into the same form as the training sample data set, a rainfall prediction model based on a deep neural network is input, the extraction of radar map space-time characteristics, the modeling of the transformation trend of the radar map on the space and the time domain, the modeling of the relation between the radar map and the corresponding rainfall, the modeling of the change trend of the meteorological information along with time and the modeling process of the relation between the meteorological information and the corresponding rainfall are automatically completed through the rainfall prediction model based on the deep neural network, and the rainfall prediction value of the i time period to the i +1 time period after the current time is directly output, the manual estimation process is omitted, time and labor are saved, and the method is convenient and easy to use.
Fig. 8 is a schematic structural diagram of a hardware device of a rainfall prediction method according to an embodiment of the present invention, and as shown in fig. 8, the device includes one or more processors and a memory. Taking a processor as an example, the apparatus may further include: an input system and an output system.
The processor, memory, input system, and output system may be connected by a bus or other means, as exemplified by the bus connection in fig. 8.
The memory, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs, non-transitory computer executable programs, and modules. The processor executes various functional applications and data processing of the electronic device, i.e., implements the processing method of the above-described method embodiment, by executing the non-transitory software program, instructions and modules stored in the memory.
The memory may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data and the like. Further, the memory may include high speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory optionally includes memory located remotely from the processor, and these remote memories may be connected to the processing system over a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input system may receive input numeric or character information and generate a signal input. The output system may include a display device such as a display screen.
The one or more modules are stored in the memory and, when executed by the one or more processors, perform the following for any of the above method embodiments:
step a: extracting radar maps with different heights in different time periods and meteorological information sequences in corresponding time periods;
step b: constructing a rainfall prediction model based on a deep neural network;
step c: inputting the extracted radar map and weather information sequence into a rainfall prediction model, carrying out time sequence analysis on the input radar map and weather information sequence through the rainfall prediction model to obtain radar map space-time characteristics and weather information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the weather information time sequence characteristics.
The product can execute the method provided by the embodiment of the invention, and has corresponding functional modules and beneficial effects of the execution method. For technical details that are not described in detail in this embodiment, reference may be made to the method provided by the embodiment of the present invention.
An embodiment of the present invention provides a non-transitory (non-volatile) computer storage medium storing computer-executable instructions that may perform the following operations:
step a: extracting radar maps with different heights in different time periods and meteorological information sequences in corresponding time periods;
step b: constructing a rainfall prediction model based on a deep neural network;
step c: inputting the extracted radar map and weather information sequence into a rainfall prediction model, carrying out time sequence analysis on the input radar map and weather information sequence through the rainfall prediction model to obtain radar map space-time characteristics and weather information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the weather information time sequence characteristics.
An embodiment of the present invention provides a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions that, when executed by a computer, cause the computer to perform the following:
step a: extracting radar maps with different heights in different time periods and meteorological information sequences in corresponding time periods;
step b: constructing a rainfall prediction model based on a deep neural network;
step c: inputting the extracted radar map and weather information sequence into a rainfall prediction model, carrying out time sequence analysis on the input radar map and weather information sequence through the rainfall prediction model to obtain radar map space-time characteristics and weather information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the weather information time sequence characteristics.
According to the rainfall prediction method, the rainfall prediction system and the electronic equipment, radar maps with different heights are used, the cloud layer information of a target area and a surrounding area is extracted on a radar map plane with a single height by using the 3D convolutional neural network, the relation between the radar maps with adjacent heights is learned, and the spatial characteristic information of the radar maps is automatically extracted; then, the time sequence analysis is carried out on the change trend of the radar map spatial characteristic information, the weather information sequence and the historical rainfall information through a long-term and short-term memory cyclic neural network, the long-term and short-term time sequence dependency relationship in the radar map change trend along with time is effectively utilized, other weather information is integrated, various data information sources are integrated, the purpose of advantage complementation is achieved, the model can be better established on the change trend along with time and the relation between the radar map and the rainfall, the error is further reduced, and therefore the future rainfall in the short term is accurately predicted; the method is time-saving, labor-saving, efficient and quick, and overcomes the defect that the prediction error is increased due to information loss caused by the utilization of a single-height radar map in the traditional method.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (9)

1. A rainfall prediction method, comprising:
step a: extracting radar maps with different heights in different time periods and meteorological information sequences in corresponding time periods;
step b: constructing a rainfall prediction model based on a deep neural network;
step c: inputting the extracted radar map and weather information sequence into a rainfall prediction model, carrying out time sequence analysis on the input radar map and weather information sequence through the rainfall prediction model to obtain radar map space-time characteristics and weather information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the weather information time sequence characteristics;
in the step a, the extracting radar maps with different heights in different time periods and the meteorological information sequence in the corresponding time period specifically includes: extracting radar maps with different heights in different time periods, overlapping the radar maps with the different heights in each time period respectively to obtain radar pattern data corresponding to each time period, and setting a sample label of each radar pattern data according to rainfall of each time period; acquiring a weather information sequence in a time period corresponding to each radar pattern sample data, and processing the extracted weather information sequence into a vector form;
in the step b, the constructing a rainfall prediction model based on a deep neural network further comprises: inputting the radar pattern data and the meteorological information sequence into a rainfall prediction model based on a deep neural network, and training the rainfall prediction model based on the deep neural network through a back propagation algorithm;
the rainfall prediction model comprises a 3D convolutional neural network and a long-short term memory cycle neural network, and the training of the rainfall prediction model based on the deep neural network through a back propagation algorithm specifically comprises the following steps:
step b 1: initializing parameters of the rainfall prediction model based on the deep neural network;
step b 2: inputting radar pattern sample data with sample labels and a meteorological information sequence into a rainfall prediction model based on a deep neural network;
step b 3: carrying out convolution pooling operation on the radar pattern data of each time period through a 3D convolution neural network, extracting radar map spatial characteristic information of each time period, and inputting the extracted radar map spatial characteristic information of each time period and a weather information sequence into a long-term and short-term memory recurrent neural network;
step b 4: and extracting radar map space-time characteristics and meteorological information time sequence characteristics through a long-short term memory cyclic neural network, and outputting rainfall prediction values of all time periods according to the radar map space-time characteristics and the meteorological information time sequence characteristics.
2. The rainfall prediction method of claim 1 wherein said training of said deep neural network based rainfall prediction model by a back propagation algorithm further comprises:
step b 5: calculating a loss function according to the output rainfall prediction value and the rainfall error in the sample label;
step b 6: judging whether the loss function meets the minimum value or not, and if not, executing the step b 7; if the minimum value is met, performing step b 8;
step b 7: adjusting network parameters according to the loss function by using a back propagation algorithm;
step b 8: and saving the network parameters, and finishing the model training.
3. The rainfall prediction method according to claim 1 or 2, wherein the deep neural network-based rainfall prediction model comprises an input layer, a first convolutional layer, a correction layer, a pooling layer, a second convolutional layer, a fully-connected layer, and an output layer; the method comprises the steps that radar pattern data and meteorological information sequences are input through an input layer, convolution operation is conducted on the input radar pattern data through a first convolution layer, correction and pooling processing are conducted through a correction layer and a pooling layer, radar pattern space characteristic information corresponding to the radar pattern data is output through a second convolution layer, the second convolution layer is fully connected to a long-short term memory cyclic neural network through a full connection layer, radar pattern space-time characteristics and meteorological information time sequence characteristics are extracted through the long-short term memory cyclic neural network according to the radar pattern space characteristic information and the meteorological information sequences, and rainfall prediction values are output through an output layer.
4. The rainfall prediction method of claim 3 wherein said long and short term memory cycling neural network comprises a forgetting gate, an input gate and an output gate; the forgetting gate is used for selectively memorizing and forgetting long-term memory of radar map spatial characteristic information and meteorological information sequences; the input gate is used for storing effective information in the radar chart and weather information sequence input at the current moment and adding the effective information into the long-term memory of the radar chart and the weather information sequence; the output gate is used for acquiring information which can have instant influence on the next moment from the accumulated radar chart and meteorological information sequence.
5. A rainfall prediction system comprising:
a radar map extraction module: the method comprises the steps of extracting radar maps with different heights in different time periods;
weather information extraction module: the weather information sequence is used for extracting the weather information sequence in the time period corresponding to the radar chart;
a network model construction module: the rainfall prediction model is used for constructing a rainfall prediction model based on a deep neural network;
a rainfall prediction module: the rainfall prediction model is used for inputting the extracted radar map and meteorological information sequence into the rainfall prediction model, performing time sequence analysis on the input radar map and meteorological information sequence through the rainfall prediction model to obtain radar map space-time characteristics and meteorological information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the meteorological information time sequence characteristics;
the radar map extraction module extracts radar maps with different heights in different time periods, and specifically comprises the following steps: extracting radar maps with different heights in different time periods, overlapping the radar maps with the different heights in each time period respectively to obtain radar pattern data corresponding to each time period, and setting a sample label of each radar pattern data according to rainfall of each time period; the meteorological information extraction module extracts a meteorological information sequence in a time period corresponding to the radar chart, and the meteorological information sequence is specifically as follows: acquiring a weather information sequence in a time period corresponding to each radar pattern sample data, and processing the extracted weather information sequence into a vector form;
the network model training module is used for inputting the radar pattern data and the meteorological information sequence into a rainfall prediction model based on a deep neural network and training the rainfall prediction model based on the deep neural network through a back propagation algorithm;
the rainfall prediction model comprises a 3D convolutional neural network and a long-short term memory cycle neural network, and the network model training module comprises:
an initialization unit: parameters for initializing the deep neural network-based rainfall prediction model;
a sample input unit: the rainfall prediction model is used for inputting the radar pattern data with the sample labels and the meteorological information sequence into the rainfall prediction model based on the deep neural network;
a first feature extraction unit: the device is used for respectively carrying out convolution pooling operation on the radar pattern data of each time period through a 3D convolution neural network, extracting radar map spatial characteristic information of each time period, and inputting the extracted radar map spatial characteristic information and meteorological information sequence of each time period into a long-short term memory recurrent neural network;
a second feature extraction unit: the method is used for extracting the radar map time-space characteristics and the meteorological information time sequence characteristics through the long-short term memory cyclic neural network and outputting rainfall prediction values of all time periods according to the radar map time-space characteristics and the meteorological information time sequence characteristics.
6. The rainfall prediction system of claim 5, wherein the network model training module further comprises:
a loss function calculation unit: the loss function is calculated according to the output rainfall prediction value and the rainfall error in the sample label;
loss function value judging means: the device is used for judging whether the loss function meets the minimum value or not, and optimizing network parameters through a network optimization unit if the loss function does not meet the minimum value; if the minimum value is met, storing the network parameters through a parameter storage unit;
a network optimization unit: the system is used for adjusting network parameters according to the loss function by applying a back propagation algorithm;
a parameter storage unit: and storing the network parameters, and finishing the model training.
7. The rainfall prediction system of claim 5 or 6 wherein the deep neural network based rainfall prediction model comprises an input layer, a first convolutional layer, a correction layer, a pooling layer, a second convolutional layer, a fully-connected layer, and an output layer; the method comprises the steps that radar pattern data and meteorological information sequences are input through an input layer, convolution operation is conducted on the input radar pattern data through a first convolution layer, correction and pooling processing are conducted through a correction layer and a pooling layer, radar pattern space characteristic information corresponding to the radar pattern data is output through a second convolution layer, the second convolution layer is fully connected to a long-short term memory cyclic neural network through a full connection layer, radar pattern space-time characteristics and meteorological information time sequence characteristics are extracted through the long-short term memory cyclic neural network according to the radar pattern space characteristic information and the meteorological information sequences, and rainfall prediction values are output through an output layer.
8. The rainfall prediction system of claim 7 wherein the long and short term memory cycling neural network comprises a forgetting gate, an input gate, and an output gate; the forgetting gate is used for selectively memorizing and forgetting long-term memory of radar map spatial characteristic information and meteorological information sequences; the input gate is used for storing effective information in the radar chart and weather information sequence input at the current moment and adding the effective information into the long-term memory of the radar chart and the weather information sequence; the output gate is used for acquiring information which can have instant influence on the next moment from the accumulated radar chart and meteorological information sequence.
9. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the following operations of the rainfall prediction method of any of the above claims 1 to 4:
step a: extracting radar maps with different heights in different time periods and meteorological information sequences in corresponding time periods;
step b: constructing a rainfall prediction model based on a deep neural network;
step c: inputting the extracted radar map and weather information sequence into a rainfall prediction model, carrying out time sequence analysis on the input radar map and weather information sequence through the rainfall prediction model to obtain radar map space-time characteristics and weather information time sequence characteristics corresponding to each time period, and outputting rainfall prediction values corresponding to each time period according to the radar map space-time characteristics and the weather information time sequence characteristics.
CN201710958075.4A 2017-10-13 2017-10-13 Rainfall prediction method and system and electronic equipment Active CN107703564B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710958075.4A CN107703564B (en) 2017-10-13 2017-10-13 Rainfall prediction method and system and electronic equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710958075.4A CN107703564B (en) 2017-10-13 2017-10-13 Rainfall prediction method and system and electronic equipment

Publications (2)

Publication Number Publication Date
CN107703564A CN107703564A (en) 2018-02-16
CN107703564B true CN107703564B (en) 2020-04-14

Family

ID=61183831

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710958075.4A Active CN107703564B (en) 2017-10-13 2017-10-13 Rainfall prediction method and system and electronic equipment

Country Status (1)

Country Link
CN (1) CN107703564B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102576427B1 (en) 2021-04-19 2023-09-08 중앙대학교 산학협력단 Real-time Rainfall Prediction Device using Cloud Images, and Rainfall Prediction Method using the same, and a computer-readable storage medium

Families Citing this family (69)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108446798A (en) * 2018-03-08 2018-08-24 重庆邮电大学 Urban population flow prediction method based on dual path space-time residual error network
CN108680911B (en) * 2018-05-17 2022-06-03 电子科技大学 Radar target direction finding method based on neural network
CN108761461B (en) * 2018-05-29 2022-02-18 南京信息工程大学 Rainfall forecasting method based on weather radar echo time sequence image
JP7438655B2 (en) * 2018-06-05 2024-02-27 株式会社東芝 Weather analysis equipment, weather analysis methods, and programs
CN108897073A (en) * 2018-06-20 2018-11-27 西安电子科技大学 Weather prediction method based on Beidou signal and terrestrial wireless signal
CN110659641B (en) * 2018-06-28 2023-05-26 杭州海康威视数字技术股份有限公司 Text recognition method and device and electronic equipment
CN109239807A (en) * 2018-06-29 2019-01-18 平安科技（深圳）有限公司 Rainfall appraisal procedure and system and terminal
CN109064460B (en) * 2018-08-01 2021-09-28 中国科学院合肥物质科学研究院 Wheat severe disease prediction method based on multiple time sequence attribute element depth characteristics
CN109190810B (en) * 2018-08-16 2021-05-04 天津大学 TDNN-based prediction method for NDVI (normalized difference vegetation index) of northern grassland area of China
CN109272144B (en) * 2018-08-16 2021-05-04 天津大学 BPNN-based prediction method for NDVI (normalized difference of variance) in northern grassland area of China
CN109344992B (en) * 2018-08-21 2021-10-19 广东工业大学 Modeling method for user control behavior habits of smart home integrating time-space factors
CN109409561B (en) * 2018-08-22 2022-04-19 国网天津市电力公司 Construction method of multi-time scale time sequence collaborative prediction model
CN110929543B (en) * 2018-08-30 2023-08-25 阿里巴巴集团控股有限公司 Meteorological data processing method and device
CN109614744B (en) * 2018-12-28 2022-11-22 浙江理工大学 Big data-based precipitation detection method and system
CN109976153B (en) * 2019-03-01 2021-03-26 北京三快在线科技有限公司 Method and device for controlling unmanned equipment and model training and electronic equipment
CN109799550B (en) * 2019-03-20 2022-02-18 北京百度网讯科技有限公司 Method and device for predicting rainfall intensity
RU2757591C1 (en) * 2019-03-25 2021-10-19 Общество С Ограниченной Ответственностью «Яндекс» Method and system for generating weather forecast
CN110096795B (en) * 2019-04-29 2023-07-18 国网湖南省电力有限公司 Correction forecasting method and system for atmospheric haze numerical forecasting
NL2023188B1 (en) * 2019-05-24 2020-12-02 Studio34 B V Method and apparatus for generating a weather forecast
CN110135654B (en) * 2019-05-24 2021-08-27 北京百度网讯科技有限公司 Method and apparatus for predicting strong convection weather
CN110363327B (en) * 2019-06-04 2021-09-03 东南大学 ConvLSTM and 3D-CNN-based short rainfall prediction method
CN110633626A (en) * 2019-07-31 2019-12-31 浙江大学 Convolutional neural network rainfall intensity online quantification method for rainy pictures
CN110610190A (en) * 2019-07-31 2019-12-24 浙江大学 Convolutional neural network rainfall intensity classification method for rainy pictures
CN110457374B (en) * 2019-08-20 2023-04-18 贵州东方世纪科技股份有限公司 Method for identifying typical rainstorm process in time period
CN110568441B (en) * 2019-09-09 2023-04-07 大连海事大学 Marine radar rain detection identification method based on convolutional neural network
CN110728411B (en) * 2019-10-18 2022-04-12 河海大学 High-low altitude area combined rainfall prediction method based on convolutional neural network
CN110824481A (en) * 2019-10-28 2020-02-21 兰州大方电子有限责任公司 Quantitative precipitation prediction method based on radar reflectivity extrapolation
CN110766333B (en) * 2019-10-29 2024-01-26 北京依派伟业数码科技有限公司 Intelligent weather phenomenon information processing method and system
CN111062410B (en) * 2019-11-05 2023-05-30 复旦大学 Star information bridge weather prediction method based on deep learning
CN110826810B (en) * 2019-11-13 2022-07-15 吉林农业大学 Regional rainfall prediction method combining spatial reasoning and machine learning
CN111008604A (en) * 2019-12-09 2020-04-14 上海眼控科技股份有限公司 Prediction image acquisition method and device, computer equipment and storage medium
CN110929963A (en) * 2019-12-16 2020-03-27 深圳智润新能源电力勘测设计院有限公司 Wind speed prediction method, wind speed prediction device, and storage medium
CN110934061A (en) * 2019-12-26 2020-03-31 裕华生态环境股份有限公司 Garden irrigation water-saving system
CN111123410A (en) * 2019-12-26 2020-05-08 国网北京市电力公司 Precipitation monitoring system and method, storage medium and processor
CN111145245B (en) * 2019-12-31 2023-07-21 哈尔滨工业大学（深圳） Short-time adjacent rainfall forecasting method, system and computer-readable storage medium
CN111239739A (en) * 2020-01-10 2020-06-05 上海眼控科技股份有限公司 Weather radar echo map prediction method and device, computer equipment and storage medium
CN111291903A (en) * 2020-01-20 2020-06-16 上海眼控科技股份有限公司 Precipitation amount estimation method and device, computer equipment and readable storage medium
CN111489525A (en) * 2020-03-30 2020-08-04 南京信息工程大学 Multi-data fusion meteorological prediction early warning method
CN111427101B (en) * 2020-04-07 2022-04-26 南京气象科技创新研究院 Thunderstorm strong wind grading early warning method, system and storage medium
CN111507929A (en) * 2020-04-15 2020-08-07 上海眼控科技股份有限公司 Meteorological cloud picture prediction method and device, computer equipment and storage medium
CN111487624A (en) * 2020-04-23 2020-08-04 上海眼控科技股份有限公司 Method and equipment for predicting rainfall capacity
CN111487695A (en) * 2020-04-28 2020-08-04 国网江苏省电力有限公司电力科学研究院 Prediction method and prediction system for squall line system
CN111830595A (en) * 2020-06-09 2020-10-27 上海眼控科技股份有限公司 Meteorological element prediction method and equipment
CN111856618A (en) * 2020-06-11 2020-10-30 上海眼控科技股份有限公司 Meteorological element prediction method and equipment
CN111983732B (en) * 2020-07-27 2022-08-05 南京信息工程大学 Rainfall intensity estimation method based on deep learning
CN111882128B (en) * 2020-07-28 2021-09-28 中原工学院 TCN-based flood season climate trend prediction method
CN111932519A (en) * 2020-08-12 2020-11-13 上海眼控科技股份有限公司 Weather prediction method and device, computer equipment and storage medium
CN111983729A (en) * 2020-08-20 2020-11-24 上海眼控科技股份有限公司 Weather phenomenon determination method, weather phenomenon determination device, computer equipment and medium
CN111736157B (en) * 2020-08-26 2021-01-05 蔻斯科技(上海)有限公司 PPI data-based prediction method and device for nowcasting
CN111950812B (en) * 2020-08-31 2022-05-24 福建农林大学 Method and device for automatically identifying and predicting rainfall
CN112183886B (en) * 2020-10-22 2024-03-15 南京信息工程大学 Short-time adjacent rainfall prediction method based on convolution network and attention mechanism
CN112470888A (en) * 2020-11-02 2021-03-12 广东职业技术学院 Automatic watering method and system for smart community
CN112415635B (en) * 2020-11-19 2022-03-29 中山大学 Gamma and Gaussian distribution coupled monthly scale rainfall forecast correction method
CN112698427B (en) * 2020-12-09 2023-01-31 最美天气(上海)科技有限公司 Short-term forecasting method and system based on space-time forecasting model
CN112711083B (en) * 2021-01-07 2022-07-05 国网福建省电力有限公司 Multi-source precipitation data dynamic fusion method and system based on adaptive weight characteristics
CN112764129B (en) * 2021-01-22 2022-08-26 易天气(北京)科技有限公司 Method, system and terminal for thunderstorm short-term forecasting
CN112800691B (en) * 2021-04-15 2021-07-30 中国气象局公共气象服务中心(国家预警信息发布中心) Method and device for constructing precipitation level prediction model
CN112988853B (en) * 2021-05-20 2021-09-10 北京市水利自动化研究所 Rainfall station network layout method based on geostatistical and optimization algorithm
CN114118544A (en) * 2021-11-11 2022-03-01 深圳市赑玄阁科技有限公司 Urban waterlogging detection method and device
CN114091765A (en) * 2021-11-25 2022-02-25 山西勇利信息科技有限公司 Future rainfall prediction method based on space-time bidirectional multi-granularity dynamic integration
CN114492952B (en) * 2022-01-06 2023-08-01 清华大学 Short-term precipitation prediction method and device based on deep learning
CN114399235B (en) * 2022-03-25 2022-07-08 北京慧图科技(集团)股份有限公司 Method and system for judging disaster risk level based on rain condition data
CN115792847B (en) * 2022-11-08 2023-07-18 江西师范大学 Quantitative precipitation estimation method based on neural network and echo vertical information
CN115629431B (en) * 2022-12-22 2023-03-14 成都数之联科技股份有限公司 Water vapor content prediction method, device, equipment and medium
CN116047633B (en) * 2023-01-28 2023-10-03 广州爱浦路网络技术有限公司 Rainfall monitoring method based on 5GS
CN116740935B (en) * 2023-06-26 2024-04-30 河北高速公路集团有限公司 Expressway environment prediction method, device, equipment and storage medium
CN117434624B (en) * 2023-09-28 2024-04-16 兰州中心气象台(兰州干旱生态环境监测预测中心) Strong convection weather identification and development prejudgment method based on semantic segmentation
CN117492113B (en) * 2023-12-29 2024-04-09 江西飞尚科技有限公司 Rainfall monitoring regulation and control method, system, electronic equipment and storage medium
CN117991412B (en) * 2024-04-07 2024-06-04 无锡九方科技有限公司 Extreme precipitation prediction method and system based on multi-mode data

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08211161A (en) * 1995-02-06 1996-08-20 Tokyo Electric Power Co Inc:The Prediction system of weather disaster
JP2006208195A (en) * 2005-01-28 2006-08-10 National Research Institute For Earth Science & Disaster Provention Device and method for estimating three-dimensional distribution of rainfall intensity and quantity
JP2010054266A (en) * 2008-08-27 2010-03-11 Kajima Corp System for predicting inundation
KR101352568B1 (en) * 2012-12-28 2014-01-24 경북대학교 산학협력단 Apparatus and method for precipitation estimation
JP2015175834A (en) * 2014-03-18 2015-10-05 日本電気株式会社 Information acquisition system, information provision terminal, service management server, information acquisition method, information provision terminal control program, and service management server control program
CN104977584A (en) * 2015-06-29 2015-10-14 深圳市气象台 Convective weather approach prediction method and system
CN106127725A (en) * 2016-05-16 2016-11-16 北京工业大学 A kind of millimetre-wave radar cloud atlas dividing method based on multiresolution CNN
CN106405682A (en) * 2016-08-29 2017-02-15 深圳先进技术研究院 Rainfall prediction method and rainfall prediction device
CN106782602A (en) * 2016-12-01 2017-05-31 南京邮电大学 Speech-emotion recognition method based on length time memory network and convolutional neural networks
RU2015154086A (en) * 2015-12-16 2017-06-21 Федеральное государственное бюджетное учреждение "Арктический и антарктический научно-исследовательский институт" METHOD FOR ESTIMATING RAIN INTENSITY BY MEASUREMENT DATA OF THE SATELLITE MICROWAVE RADIOMETER AMSR2
CN107230351A (en) * 2017-07-18 2017-10-03 福州大学 A kind of Short-time Traffic Flow Forecasting Methods based on deep learning

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08211161A (en) * 1995-02-06 1996-08-20 Tokyo Electric Power Co Inc:The Prediction system of weather disaster
JP2006208195A (en) * 2005-01-28 2006-08-10 National Research Institute For Earth Science & Disaster Provention Device and method for estimating three-dimensional distribution of rainfall intensity and quantity
JP2010054266A (en) * 2008-08-27 2010-03-11 Kajima Corp System for predicting inundation
KR101352568B1 (en) * 2012-12-28 2014-01-24 경북대학교 산학협력단 Apparatus and method for precipitation estimation
JP2015175834A (en) * 2014-03-18 2015-10-05 日本電気株式会社 Information acquisition system, information provision terminal, service management server, information acquisition method, information provision terminal control program, and service management server control program
CN104977584A (en) * 2015-06-29 2015-10-14 深圳市气象台 Convective weather approach prediction method and system
RU2015154086A (en) * 2015-12-16 2017-06-21 Федеральное государственное бюджетное учреждение "Арктический и антарктический научно-исследовательский институт" METHOD FOR ESTIMATING RAIN INTENSITY BY MEASUREMENT DATA OF THE SATELLITE MICROWAVE RADIOMETER AMSR2
CN106127725A (en) * 2016-05-16 2016-11-16 北京工业大学 A kind of millimetre-wave radar cloud atlas dividing method based on multiresolution CNN
CN106405682A (en) * 2016-08-29 2017-02-15 深圳先进技术研究院 Rainfall prediction method and rainfall prediction device
CN106782602A (en) * 2016-12-01 2017-05-31 南京邮电大学 Speech-emotion recognition method based on length time memory network and convolutional neural networks
CN107230351A (en) * 2017-07-18 2017-10-03 福州大学 A kind of Short-time Traffic Flow Forecasting Methods based on deep learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
基于BP神经网络的降雨量预测研究;葛彩莲;《节水灌溉》;20160516;正文107-109页 *
基于神经网络和模型集成的段时降雨预测方法;郭尚瓒;《气象科技进展》;20170215;正文第8-9页 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102576427B1 (en) 2021-04-19 2023-09-08 중앙대학교 산학협력단 Real-time Rainfall Prediction Device using Cloud Images, and Rainfall Prediction Method using the same, and a computer-readable storage medium

Also Published As

Publication number Publication date
CN107703564A (en) 2018-02-16

Similar Documents

Publication Publication Date Title
CN107703564B (en) Rainfall prediction method and system and electronic equipment
US20220043182A1 (en) Spatial autocorrelation machine learning-based downscaling method and system of satellite precipitation data
CN110363327B (en) ConvLSTM and 3D-CNN-based short rainfall prediction method
Gao et al. Hourly forecasting of solar irradiance based on CEEMDAN and multi-strategy CNN-LSTM neural networks
CN112016052B (en) Near-surface daily maximum air temperature estimation method, system and terminal based on multi-source data
CN109359166B (en) Space growth dynamic simulation and driving force factor contribution degree synchronous calculation method
CN113505521B (en) Urban waterlogging rapid forecasting method based on neural network-numerical simulation
CN110134907B (en) Rainfall missing data filling method and system and electronic equipment
CN114067019A (en) Urban waterlogging risk map rapid prefabricating method coupling deep learning and numerical simulation
CN114330935B (en) New energy power prediction method and system based on multiple combination strategies integrated learning
CN113420939A (en) Cloud picture forecasting method, equipment and storage medium
Sampath et al. Estimation of rooftop solar energy generation using satellite image segmentation
CN116720156A (en) Weather element forecasting method based on graph neural network multi-mode weather data fusion
CN111811465B (en) Method for predicting sea wave effective wave height based on multi-sine function decomposition neural network
CN115882454B (en) Source load power prediction method and system based on multi-element information fusion
CN117374956A (en) Short-term prediction method for photovoltaic power generation of comprehensive energy station
Ye et al. Multi-year ENSO forecasts using parallel convolutional neural networks with heterogeneous architecture
CN113159434A (en) Radar echo prediction method, device, equipment and storage medium
CN116955450A (en) Hydrologic forecasting method and system for fusion of drainage basin production confluence space-time process
CN116090596A (en) Typhoon intensity prediction method, typhoon intensity prediction device, typhoon intensity prediction equipment and storage medium
CN115099318A (en) Training method and application method of wind speed space-time prediction network and electronic equipment
CN114861530A (en) ENSO intelligent prediction method, device, equipment and storage medium
CN116108761B (en) Regional climate simulation method and system for coupling deep learning and HASM
CN116976227A (en) Storm water increasing forecasting method and system based on LSTM machine learning
CN116522648A (en) Lake algae state prediction method and application

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