CN117312832A - Depth sequence model-based medium-and-long-term cloud cover prediction method and system - Google Patents

Depth sequence model-based medium-and-long-term cloud cover prediction method and system Download PDF

Info

Publication number
CN117312832A
CN117312832A CN202311594604.9A CN202311594604A CN117312832A CN 117312832 A CN117312832 A CN 117312832A CN 202311594604 A CN202311594604 A CN 202311594604A CN 117312832 A CN117312832 A CN 117312832A
Authority
CN
China
Prior art keywords
input
module
prediction
trend
output
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202311594604.9A
Other languages
Chinese (zh)
Other versions
CN117312832B (en
Inventor
叶允明
陆沈源
李旭涛
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.)
Harbin Institute Of Technology shenzhen Shenzhen Institute Of Science And Technology Innovation Harbin Institute Of Technology
Original Assignee
Harbin Institute Of Technology shenzhen Shenzhen Institute Of Science And Technology Innovation Harbin Institute Of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Harbin Institute Of Technology shenzhen Shenzhen Institute Of Science And Technology Innovation Harbin Institute Of Technology filed Critical Harbin Institute Of Technology shenzhen Shenzhen Institute Of Science And Technology Innovation Harbin Institute Of Technology
Priority to CN202311594604.9A priority Critical patent/CN117312832B/en
Publication of CN117312832A publication Critical patent/CN117312832A/en
Application granted granted Critical
Publication of CN117312832B publication Critical patent/CN117312832B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/10—Devices for predicting weather conditions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G06F17/141—Discrete Fourier transforms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0499—Feedforward networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Databases & Information Systems (AREA)
  • Environmental & Geological Engineering (AREA)
  • Algebra (AREA)
  • Discrete Mathematics (AREA)
  • Ecology (AREA)
  • Atmospheric Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Operations Research (AREA)
  • Environmental Sciences (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a medium-long term cloud cover prediction method and a system based on a depth sequence model, which belong to the technical field of weather prediction, wherein historical weather data of a predicted position is input into a prediction branch module based on trend decomposition and period enhancement, and the historical weather data of an area near the predicted position is input into a weather system extraction module; stacking a plurality of basic blocks of the prediction branch module based on trend decomposition and period enhancement to obtain an output of the prediction branch module based on trend decomposition and period enhancement; and vector splicing is carried out on the output of the prediction branch module based on trend decomposition and period enhancement and the output of the weather system extraction module in the channel dimension, the vector is used as the input of the prediction projection module, and the prediction projection module projects the input vector into a specified shape, so that a final prediction result is obtained. The method utilizes the historical meteorological conditions of the area near the predicted position to assist cloud cover prediction, and is more in line with the actual situation and meteorological rules.

Description

Depth sequence model-based medium-and-long-term cloud cover prediction method and system
Technical Field
The invention belongs to the technical field of weather prediction, and particularly relates to a medium-long term cloud cover prediction method and system based on a depth sequence model.
Background
The cloud quantity comprises total cloud quantity, high cloud quantity, medium cloud quantity and low cloud quantity, and a reliable and accurate cloud parameter prediction method has important significance in many fields. For example, in the field of photovoltaic energy power generation, cloud cover is one of main factors influencing solar irradiation energy supply, and directly influences the power generation efficiency of a photovoltaic module, and prediction of the cloud cover can guide prediction of photovoltaic power generation power, so that more effective dispatching of electric power is performed, and greater economic benefit is brought. Currently, the main methods of cloud cover prediction are numerical mode prediction and statistical weather prediction. The numerical mode prediction obtains a predicted value by solving an atmospheric motion building equation, and the statistical weather prediction combines the numerical mode prediction with statistics to correct the output of the numerical mode prediction. The disadvantage of both methods is that the step of solving the atmospheric motion equation is extremely computationally demanding to handle.
With the emission of various meteorological satellites and the layout of meteorological monitoring stations, meteorological data are increasingly complete, so that weather forecast becomes possible by using a pure data driving method. In recent years, researchers have attempted to improve the accuracy of cloud prediction by using a method of machine learning. The traditional machine learning step generally consists of three parts, namely data preprocessing, feature extraction and classifier. Traditional machine learning is constructed through manual feature selection and feature engineering, and is difficult to be used for complex data structures. Deep learning can well solve the problem of difficult feature extraction in traditional machine learning. The deep learning is a data-oriented learning mode, does not depend on expert experience screening characteristics, can learn the characteristics of data through a large amount of input data, and has good generalization, universality and portability. With a large amount of data support, the effect of the deep learning model is often better than that of the traditional machine learning model.
Currently, in the field of weather forecast, research on cloud amount prediction by using deep learning is still less, and the existing cloud amount prediction technology mainly has the following defects: 1. the prediction precision is low, and most prediction technologies cannot realize hour-by-hour prediction; 2. the prediction accuracy is seriously reduced with time, and the cloud parameter prediction accuracy after 72 hours is poor.
Disclosure of Invention
In order to solve the problems, the invention provides a medium-long term cloud cover prediction method and a system based on a depth sequence model.
The invention is realized by the following technical scheme:
mid-long term cloud cover prediction system based on depth sequence model:
the cloud computing prediction system includes: the system comprises an input module, a prediction branch module based on trend decomposition and period enhancement, a weather system extraction module and a prediction projection module;
the input module is used for inputting historical meteorological data; comprising an input 1 and an input 2; input 1 to the trend decomposition and cycle enhancement module, input 2 to the weather system extraction module;
the prediction branch module based on trend decomposition and period enhancement consists of a plurality of basic blocks, wherein each basic block comprises a trend decomposition module, a period enhancement module and a feedforward network module;
and vector splicing is carried out on the output of the prediction branch module based on trend decomposition and period enhancement and the output of the weather system extraction module in the channel dimension, the vector is used as the input of the prediction projection module, and the prediction projection module projects the input vector into a specified shape, so that a final prediction result is obtained.
Further, input 1 and input 2 are used as initial data inputsThe cloud amount prediction system specifically comprises a step of inputting historical meteorological data of a predicted position represented by 1, wherein the dimension is as follows，/>Representing the length of the input sequence, < >>A weather data characteristic channel number representing the predicted location;
input 2 represents historical meteorological data for an area near the predicted location, with dimensions of，/>Representing the length of the input sequence, < >>Indicating the length of the region>Representing the width of the region>And the weather data characteristic channel number of the area near the predicted position is represented.
Further, the trend decomposition module extracts trend terms in the sequence by using a method of moving average, and the calculation formula of the moving average is as follows:
wherein,representing the first of trend itemsiValue of individual element->Representation ofTime series NojThe value of the time point at which the data is to be read,Nindicating the size of the sliding window;
usingnThe calculation formula for extracting the time sequence trend terms of the sliding windows with different sizes is as follows:
wherein,representing the trend term finally extracted, +.>Indicating use of the firstiTrend item extraction is carried out on the time sequence through a sliding window, and the time sequence is->Is a learnable weight for integrating trend terms extracted from different sliding windows.
Further, the period enhancement module first inputs a sequencePerforming Fourier transform to obtain a representation of the sequence in the frequency domain>Respectively calculateNThe magnitudes of the individual components are selectedMThe component with the largest amplitude is obtained>The rest ofN-MThe individual components are filled with 0 to give +.>Using a parameter which can be learnedOn the complex domain and +.>MultiplicationObtain->To enhance the main components in the sequence,
to filter the representation of the original input sequence in the frequency domain after the lower amplitude frequency domain components,
to replace the frequency domain component of the original sequence with 0 which has a lower amplitude in the frequency domain,
to use a learnable operator with +.>As a result of the multiplication the result of which,
wherein the method comprises the steps ofRIn the real number domain of the number,Sthe length of the data sequence is chosen to be,Das the number of characteristic channels of the data,Cis a plurality of fields, which are the fields,
finally toPerforming inverse Fourier transform to recover to time domain to obtain output +.>The method comprises the steps of carrying out a first treatment on the surface of the The calculation formula of the period enhancement module is as follows:
wherein the method comprises the steps ofRepresenting the fourier transform +.>Representing inverse fourier transform ++>Indicating the choice of the component with the greatest amplitude +.>Representing an extended vector with 0.
Further, the output of the trend decomposition module and the output of the period enhancement module are added to be used as the input of a feedforward network, and the input of the feedforward network and the original input are subjected to residual connection to obtain the output of a basic block;
the calculation formula of one basic block is as follows:
wherein,representing trend decomposition module->Representing a period enhancement module, feedForward->Representing a feed-forward network, T being a sequence trend, X representing an input of a basic block, S' representing an output of the basic block; multiple basic blocks are stacked to obtain the output of predictive branching module based on trend decomposition and period enhancement>Which is provided withMiddle->The number of characteristic channels output by the prediction branch module based on trend decomposition and period enhancement.
Furthermore, in the weather system extraction module, a three-dimensional convolution layer is used for carrying out convolution operation on input data; the three-dimensional convolution layer can simultaneously consider the information of the time dimension, the space dimension and the channel dimension;
for historical weather data, the time dimension represents data of different time steps, the space dimension represents data of different space positions, and the channel dimension represents data of different meteorological variables;
through three-dimensional convolution operations, the network utilizes correlations and patterns between these dimensions to extract features, including local patterns in space and trends in time;
reducing the spatial dimension of the input to 1 using an average pooling layer; under the condition of keeping the time and the channel dimension unchanged, dimension reduction is carried out on the space dimension;
through these convolution and pooling operations, the network is able to extract a feature representation with spatio-temporal information from historical weather data; these characteristic representations include weather patterns, spatial correlations, and temporal trend information within a certain region; the weather system extraction module has a receiving dimension ofIs output as +.>，
Wherein the method comprises the steps ofThe number of the characteristic channels output by the weather system extraction module is represented, the value is determined according to the number of weather attributes contained in the output of the weather system extraction module or the number of categories to be classified, and the weather system extraction module is defined according to actual task requirements.
Further, the output of the prediction branch module to be enhanced based on trend decomposition and periodAnd the output of the weather system extraction module>Vector stitching is carried out in the channel dimension; obtain->
Will beAs input to the predictive projection module, the predictive projection module projects the input vector into a specified shape to obtain a final predictive result +.>。
A medium-long term cloud cover prediction method based on a depth sequence model comprises the following steps: the method specifically comprises the following steps:
step 1, an input module divides historical meteorological data into an input 1 of historical meteorological data of a predicted position and an input 2 of historical meteorological data of an area near the predicted position;
step 2, obtaining trend items of the original sequence by the trend decomposition module through the input 1, obtaining season items and residual items of the time sequence by subtracting the original sequence from the trend items, enhancing main periodic components by the period enhancement module through the season items and the residual items of the time sequence,
step 3, adding the output of the trend decomposition module and the output of the period enhancement module to be used as the input of a feedforward network module, and carrying out residual connection on the input of the feedforward network and the original input to obtain the output of a basic block; stacking a plurality of basic blocks to obtain the output of a prediction branch module based on trend decomposition and period enhancement;
step 4, inputting the input 2 into a weather system extraction module;
and 5, vector splicing is carried out on the output of the prediction branch module based on trend decomposition and period enhancement and the output of the weather system extraction module in the channel dimension, the vector is used as the input of the prediction projection module, and the prediction projection module projects the input vector into a specified shape, so that a final prediction result is obtained.
An electronic device comprising a memory storing a computer program and a processor implementing the steps of the above method when the processor executes the computer program.
A computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the above method.
The invention has the beneficial effects that
The invention uses the trend decomposition module and the period enhancement module to enhance the predictable part in the time sequence and reduce the influence of random components on prediction.
The weather system extraction module is used, and the cloud cover prediction is assisted by utilizing the historical meteorological conditions of the area near the predicted position, so that the weather system extraction module is more in line with the actual situation and meteorological rules.
The projection module is used for projecting the extracted time sequence characteristics into a prediction result at one time, so that the accumulated error propagation in the prediction stage is avoided, and the prediction precision of an hour level can be achieved.
Drawings
FIG. 1 is a diagram showing the overall structure of a medium-long term cloud cover prediction model based on a depth sequence model;
FIG. 2 is a trend decomposition module of the present invention;
FIG. 3 is a periodic enhancement module of the present invention;
fig. 4 is a 3 DCNN-based weather system extraction module according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
With reference to fig. 1 to 4.
Mid-long term cloud cover prediction system based on depth sequence model:
the cloud computing prediction system includes: the system comprises an input module, a prediction branch module based on trend decomposition and period enhancement, a weather system extraction module based on 3DCNN (three-dimensional convolutional neural network) and a prediction projection module;
the input module is used for inputting historical meteorological data; comprising an input 1 and an input 2; input 1 to the trend decomposition and cycle enhancement module, input 2 to the weather system extraction module;
the prediction branch module based on trend decomposition and period enhancement consists of a plurality of basic blocks, wherein each basic block comprises a trend decomposition module, a period enhancement module and a feedforward network module;
and vector splicing is carried out on the output of the prediction branch module based on trend decomposition and period enhancement and the output of the weather system extraction module in the channel dimension, the vector is used as the input of the prediction projection module, and the prediction projection module projects the input vector into a specified shape, so that a final prediction result is obtained.
Input 1 and input 2 are used as initial data to input the cloud amount prediction system, specifically, input 1 represents historical meteorological data of a predicted position, and the dimension is，/>Representing the length of the input sequence, < >>A weather data characteristic channel number representing the predicted location;
input 2 represents historical meteorological data for an area near the predicted location, with dimensions of，/>Representing the length of an input sequenceDegree (f)>Indicating the length of the region>Representing the width of the region>And the weather data characteristic channel number of the area near the predicted position is represented.
The trend decomposition module extracts trend items in the sequence by using a method of moving average, and the calculation formula of the moving average is as follows:
wherein,representing the first of trend itemsiValue of individual element->Representing time series NojThe value of the time point at which the data is to be read,Nindicating the size of the sliding window;
usingnThe calculation formula for extracting the time sequence trend terms of the sliding windows with different sizes is as follows:
wherein,representing the trend term finally extracted, +.>Indicating use of the firstiTrend item extraction is carried out on the time sequence through a sliding window, and the time sequence is->Is a weight which can be learned byAnd integrating the trend items extracted from different sliding windows.
The method comprises the steps that a seasonal item and a residual item of a time sequence are obtained through subtraction of an original sequence and a trend item, the seasonal item and the residual item obtained through trend decomposition are used as inputs of a period enhancement module, and the seasonal item and the residual item are not separated but are fused together into a vector;
seasonal items reflect periodic or seasonal patterns in the time series data; the residual term represents a part of the time series data that cannot be interpreted by the trend term and the season term, i.e., remaining random fluctuations or noise; it contains random fluctuations in time series, outliers or other variations that cannot be interpreted by trend and seasonal terms; because the input vector contains residual error items, the randomness is larger, if the randomness can be reduced, the components with main periodicity can be enhanced, and the prediction accuracy can be effectively improved; the overall structure of the period enhancement module is shown in fig. 3;
the period enhancement module first pairs an input sequencePerforming Fourier transform to obtain a representation of the sequence in the frequency domain>To screen out the major components in the sequence, the calculation was performed separatelyNThe magnitudes of the individual components are selectedMThe component with the largest amplitude is obtained>The rest ofN-MThe individual components are filled with 0 to give +.>Using a learnable parameter +.>On the complex domain and +.>Multiplication to obtain->To enhance the main components in the sequence,
to filter the representation of the original input sequence in the frequency domain after the lower amplitude frequency domain components,
to replace the frequency domain component of the original sequence with 0 which has a lower amplitude in the frequency domain,
to use a learnable operator with +.>As a result of the multiplication the result of which,
wherein the method comprises the steps ofRIn the real number domain of the number,Sthe length of the data sequence is chosen to be,Das the number of characteristic channels of the data,Cis a plurality of fields, which are the fields,
finally toPerforming inverse Fourier transform to recover to time domain to obtain output +.>The method comprises the steps of carrying out a first treatment on the surface of the The calculation formula of the period enhancement module is as follows:
wherein the method comprises the steps ofRepresenting the fourier transform +.>Representing inverse fourier transform ++>Indicating the choice of the component with the greatest amplitude +.>Representing an extended vector with 0.
The output of the trend decomposition module and the output of the period enhancement module are added to be used as the input of a feedforward network, and the input of the feedforward network and the original input are subjected to residual connection to obtain the output of a basic block;
the calculation formula of one basic block is as follows:
wherein,representing trend decomposition module->Representing a period enhancement module, feedForward->Representing a feed-forward network, T being a sequence trend, X representing an input of a basic block, S' representing an output of the basic block; multiple basic blocks are stacked to obtain the output of predictive branching module based on trend decomposition and period enhancement>Wherein->The number of characteristic channels output by the prediction branch module based on trend decomposition and period enhancement.
When cloud amount prediction is performed on a designated area, certain limitation exists only by using historical meteorological data of the area as a predictor, and accurate prediction cannot be formed on the change of local weather; weather and cloud cover generally have a certain variability in space, and the historical meteorological data of the area is only relied on to capture the difference and the characteristics in space, so that the data of the nearby area can be considered to better reflect the changes and the modes in space; the weather data of the nearby region may provide information about weather patterns and trends that are helpful in predicting cloud cover for the target region;
by considering data in nearby regions, shared weather features and trends can be utilized to refine the predictive model; the historical meteorological data of the area nearby the target prediction area is added into the cloud cover prediction model, so that factors such as variability, geographic conditions, meteorological modes and trends in space can be comprehensively considered, and the prediction accuracy and reliability are improved; in order to use the weather data in the area range in the prediction model, a weather system extraction module based on 3DCNN is adopted, and the structure of the weather system extraction module is shown in FIG. 4;
in the weather system extraction module, a three-dimensional convolution layer is used for carrying out convolution operation on input data; the three-dimensional convolution layer can simultaneously consider the information of the time dimension, the space dimension and the channel dimension;
for historical weather data, the time dimension represents data of different time steps, the space dimension represents data of different space positions, and the channel dimension represents data of different meteorological variables;
through three-dimensional convolution operations, the network utilizes correlations and patterns between these dimensions to extract features, including local patterns in space and trends in time;
reducing the spatial dimension of the input to 1 using an average pooling layer; the pooling operation can reduce the dimension of the space under the condition of keeping the time and the dimension of the channel unchanged; the purpose of this is to capture the overall characteristics of the input data in a spatial range, not just the details of each specific location;
through these convolution and pooling operations, the network is able to extract a feature representation with spatio-temporal information from historical weather data; these characteristic representations may include information such as weather patterns, spatial correlations, and temporal trends within a region; this feature extraction process enables the network to better understand and utilize historical weather data, thereby improving the accuracy and reliability of weather predictions or analyses over this range;
as shown in fig. 4, the 3 DCNN-based weather system extraction module receives a dimension asIs a weather data sequence within a certain range, H and W are the number of rows and columns of the range, S is the length of the data sequence, C is the number of characteristic channels of the data, and the output of the module is a +.>Is a vector of (2); the output vector contains space-time characteristics abstracted from an original weather data sequence, and can be directly spliced with historical weather data of a target prediction area to be used as input of a prediction model;
in the invention, the weather system extraction module has a receiving dimension ofIs output as +.>，
Wherein the method comprises the steps ofThe number of the characteristic channels output by the weather system extraction module is represented, the value is determined according to the number of weather attributes contained in the output of the weather system extraction module or the number of categories to be classified, and the weather system extraction module is defined according to actual task requirements.
Output of predictive branching module to be based on trend decomposition and period enhancementAnd the output of the weather system extraction module>Vector stitching is carried out in the channel dimension; obtain->
Will beAs input to the predictive projection module, the predictive projection module projects the input vector into a specified shape to obtain a final predictive result +.>。
A medium-long term cloud cover prediction method based on a depth sequence model comprises the following steps: the method specifically comprises the following steps:
step 1, an input module divides historical meteorological data into an input 1 of historical meteorological data of a predicted position and an input 2 of historical meteorological data of an area near the predicted position;
step 2, obtaining trend items of the original sequence by the trend decomposition module through the input 1, obtaining season items and residual items of the time sequence by subtracting the original sequence from the trend items, enhancing main periodic components by the period enhancement module through the season items and the residual items of the time sequence,
step 3, adding the output of the trend decomposition module and the output of the period enhancement module to be used as the input of a feedforward network module, and carrying out residual connection on the input of the feedforward network and the original input to obtain the output of a basic block; stacking a plurality of basic blocks to obtain the output of a prediction branch module based on trend decomposition and period enhancement;
step 4, inputting the input 2 into a weather system extraction module; the network can better understand and utilize the historical weather data of the area near the predicted position, so that the accuracy and the reliability of weather prediction or analysis in the range are improved;
and 5, vector splicing is carried out on the output of the prediction branch module based on trend decomposition and period enhancement and the output of the weather system extraction module in the channel dimension, the vector is used as the input of the prediction projection module, and the prediction projection module projects the input vector into a specified shape, so that a final prediction result is obtained.
An electronic device comprising a memory storing a computer program and a processor implementing the steps of the above method when the processor executes the computer program.
A computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the above method.
The memory in embodiments of the present application may be either volatile memory or nonvolatile memory, or may include both volatile and nonvolatile memory. The nonvolatile memory may be read only memory, ROM, programmable ROM, PROM, erasable PROM, EPROM, electrically erasable EPROM, EEPROM, or flash memory. The volatile memory may be random access memory random access memory, RAM, which acts as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM, SRAM, dynamic RAM, DRAM, synchronous DRAM, SDRAM, double data rate synchronous DRAM double data rate SDRAM, DDR SDRAM, enhanced SDRAM, ESDRAM, synchronous link DRAM, SLDRAM and direct memory bus RAM, DR RAM. It should be noted that the memory of the methods described herein is intended to comprise, without being limited to, these and any other suitable types of memory.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the processes or functions described in accordance with embodiments of the present application are produced in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by means of a wired, such as coaxial cable, optical fiber, digital subscriber line digital subscriber line, DSL, or wireless, such as infrared, wireless, microwave, or the like. The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a magnetic medium such as a floppy disk, a hard disk, a magnetic tape, an optical medium such as a high-density digital video disk digital video disc, a DVD, or a semiconductor medium such as a solid state disk, an SSD, or the like.
In implementation, each step of the method may be implemented by an integrated logic circuit of hardware in a processor or an instruction in a form of a software component. The steps of a method disclosed in connection with the embodiments of the present application may be embodied directly in a hardware processor for execution, or in a combination of hardware and software modules in the processor for execution. The software modules may be located in a random access memory, flash memory, read only memory, programmable read only memory, or electrically erasable programmable memory, registers, etc. as well known in the art. The storage medium is located in a memory, and the processor reads the information in the memory and, in combination with its hardware, performs the steps of the above method. To avoid repetition, a detailed description is not provided herein.
It should be noted that the processor in the embodiments of the present application may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method embodiments may be implemented by integrated logic circuits of hardware in a processor or instructions in software form. The processor may be a general purpose processor, a digital signal processor DSP, an application specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The disclosed methods, steps, and logic blocks in the embodiments of the present application may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the embodiments of the present application may be embodied directly in hardware, in a decoded processor, or in a combination of hardware and software modules in a decoded processor. The software modules may be located in a random access memory, flash memory, read only memory, programmable read only memory, or electrically erasable programmable memory, registers, etc. as well known in the art. The storage medium is located in a memory, and the processor reads the information in the memory and, in combination with its hardware, performs the steps of the above method.
The invention provides a medium-long term cloud cover prediction method and a system based on a depth sequence model, which are described in detail, and the principle and the implementation mode of the invention are described, and the description of the embodiment is only used for helping to understand the method and the core idea of the invention; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present invention, the present description should not be construed as limiting the present invention in view of the above.

Claims (10)

1. The medium-long term cloud amount prediction system based on the depth sequence model is characterized in that:
the cloud computing prediction system includes: the system comprises an input module, a prediction branch module based on trend decomposition and period enhancement, a weather system extraction module and a prediction projection module;
the input module is used for inputting historical meteorological data; comprising an input 1 and an input 2; input 1 to the trend decomposition and cycle enhancement module, input 2 to the weather system extraction module;
the prediction branch module based on trend decomposition and period enhancement consists of a plurality of basic blocks, wherein each basic block comprises a trend decomposition module, a period enhancement module and a feedforward network module;
and vector splicing is carried out on the output of the prediction branch module based on trend decomposition and period enhancement and the output of the weather system extraction module in the channel dimension, the vector is used as the input of the prediction projection module, and the prediction projection module projects the input vector into a specified shape, so that a final prediction result is obtained.
2. The cloud computing system of claim 1, wherein:
input 1 and input 2 are used as initial data to input the cloud amount prediction system, specifically, input 1 represents historical meteorological data of a predicted position, and the dimension is，/>Representing the length of the input sequence, < >>A weather data characteristic channel number representing the predicted location;
input 2 represents historical meteorological data for an area near the predicted location, with dimensions of，/>Representing the length of the input sequence, < >>Representing a regionLength of (L)>Representing the width of the region>And the weather data characteristic channel number of the area near the predicted position is represented.
3. The cloud computing system of claim 2, wherein:
the trend decomposition module extracts trend items in the sequence by using a method of moving average, and the calculation formula of the moving average is as follows:
wherein,representing the first of trend itemsiValue of individual element->Representing time series NojThe value of the time point at which the data is to be read,Nindicating the size of the sliding window;
usingnThe calculation formula for extracting the time sequence trend terms of the sliding windows with different sizes is as follows:
wherein,representing the trend term finally extracted, +.>Indicating use of the firstiTrend item extraction is carried out on the time sequence through a sliding window, and the time sequence is->Is a learnable weight for integrating trend terms extracted from different sliding windows.
4. A cloud computing prediction system as recited in claim 3, wherein:
the period enhancement module first pairs an input sequencePerforming Fourier transform to obtain a representation of the sequence in the frequency domain>Respectively calculateNThe magnitudes of the individual components are selectedMThe component with the largest amplitude is obtained>The rest ofN-MThe individual components are filled with 0 to give +.>Using a learnable parameter +.>On complex domain andmultiplication to obtain->To enhance the main components in the sequence,
to filter the representation of the original input sequence in the frequency domain after the lower amplitude frequency domain components,
to replace the frequency domain component of the original sequence with 0 which has a lower amplitude in the frequency domain,
to use a learnable operator with +.>As a result of the multiplication the result of which,
wherein the method comprises the steps ofRIn the real number domain of the number,Sthe length of the data sequence is chosen to be,Das the number of characteristic channels of the data,Cis a plurality of fields, which are the fields,
finally toPerforming inverse Fourier transform to recover to time domain to obtain output +.>The method comprises the steps of carrying out a first treatment on the surface of the The calculation formula of the period enhancement module is as follows:
wherein the method comprises the steps ofRepresenting the fourier transform +.>Representing inverse fourier transform ++>Indicating the choice of the component with the greatest amplitude +.>Representing an extended vector with 0.
5. The cloud computing system of claim 4, wherein:
the output of the trend decomposition module and the output of the period enhancement module are added to be used as the input of a feedforward network, and the input of the feedforward network and the original input are subjected to residual connection to obtain the output of a basic block;
the calculation formula of one basic block is as follows:
wherein,representing trend decomposition module->Representing a period enhancement module, feedForward->Representing a feed-forward network, T being a sequence trend, X representing an input of a basic block, S' representing an output of the basic block; multiple basic blocks are stacked to obtain the output of predictive branching module based on trend decomposition and period enhancement>Wherein->The number of characteristic channels output by the prediction branch module based on trend decomposition and period enhancement.
6. The cloud computing system of claim 5, wherein:
in the weather system extraction module, a three-dimensional convolution layer is used for carrying out convolution operation on input data; the three-dimensional convolution layer simultaneously considers the information of the time dimension, the space dimension and the channel dimension;
for historical weather data, the time dimension represents data of different time steps, the space dimension represents data of different space positions, and the channel dimension represents data of different meteorological variables;
through three-dimensional convolution operations, the network utilizes correlations and patterns between these dimensions to extract features, including local patterns in space and trends in time;
reducing the spatial dimension of the input to 1 using an average pooling layer; under the condition of keeping the time and the channel dimension unchanged, dimension reduction is carried out on the space dimension;
through these convolution and pooling operations, the network is able to extract a feature representation with spatio-temporal information from historical weather data; these characteristic representations include weather patterns, spatial correlations, and temporal trend information within a certain region; the weather system extraction module has a receiving dimension ofIs output as +.>，
Wherein the method comprises the steps ofThe number of characteristic channels output by the weather system extraction module is represented, and the value depends on the number of weather attributes contained in the output by the weather system extraction module or the number of weather attributes to be classifiedAnd the category number is defined according to the actual task demand.
7. The cloud computing system of claim 6, wherein:
output of predictive branching module to be based on trend decomposition and period enhancementAnd the output of the weather system extraction module>Vector stitching is carried out in the channel dimension; obtain->，
Will beAs input to the predictive projection module, the predictive projection module projects the input vector into a specified shape to obtain a final predictive result +.>。
8. A medium-long term cloud cover prediction method based on a depth sequence model is characterized by comprising the following steps of:
the method specifically comprises the following steps:
step 1, an input module divides historical meteorological data into an input 1 of historical meteorological data of a predicted position and an input 2 of historical meteorological data of an area near the predicted position;
step 2, obtaining trend items of the original sequence by the trend decomposition module through the input 1, obtaining season items and residual items of the time sequence by subtracting the original sequence from the trend items, enhancing main periodic components by the period enhancement module through the season items and the residual items of the time sequence,
step 3, adding the output of the trend decomposition module and the output of the period enhancement module to be used as the input of a feedforward network module, and carrying out residual connection on the input of the feedforward network and the original input to obtain the output of a basic block; stacking a plurality of basic blocks to obtain the output of a prediction branch module based on trend decomposition and period enhancement;
step 4, inputting the input 2 into a weather system extraction module;
and 5, vector splicing is carried out on the output of the prediction branch module based on trend decomposition and period enhancement and the output of the weather system extraction module in the channel dimension, the vector is used as the input of the prediction projection module, and the prediction projection module projects the input vector into a specified shape, so that a final prediction result is obtained.
9. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of claim 8 when executing the computer program.
10. A computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the method of claim 8.
CN202311594604.9A 2023-11-28 2023-11-28 Depth sequence model-based medium-and-long-term cloud cover prediction method and system Active CN117312832B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311594604.9A CN117312832B (en) 2023-11-28 2023-11-28 Depth sequence model-based medium-and-long-term cloud cover prediction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311594604.9A CN117312832B (en) 2023-11-28 2023-11-28 Depth sequence model-based medium-and-long-term cloud cover prediction method and system

Publications (2)

Publication Number Publication Date
CN117312832A true CN117312832A (en) 2023-12-29
CN117312832B CN117312832B (en) 2024-07-12

Family

ID=89281374

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311594604.9A Active CN117312832B (en) 2023-11-28 2023-11-28 Depth sequence model-based medium-and-long-term cloud cover prediction method and system

Country Status (1)

Country Link
CN (1) CN117312832B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118735082A (en) * 2024-09-03 2024-10-01 南京信息工程大学 3D-TimesNet-based sub-season air temperature prediction correction method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA3146167A1 (en) * 2019-07-08 2021-01-14 Indigo Ag, Inc. Crop yield forecasting models
US20220172130A1 (en) * 2017-12-14 2022-06-02 Business Objects Software Ltd Multi-step time series forecasting with residual learning
US20220172038A1 (en) * 2020-11-30 2022-06-02 International Business Machines Corporation Automated deep learning architecture selection for time series prediction with user interaction
CN115456314A (en) * 2022-11-11 2022-12-09 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) Atmospheric pollutant space-time distribution prediction system and method
CN116933124A (en) * 2023-06-16 2023-10-24 新奥新智科技有限公司 Time series data prediction method, device, equipment and storage medium

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220172130A1 (en) * 2017-12-14 2022-06-02 Business Objects Software Ltd Multi-step time series forecasting with residual learning
CA3146167A1 (en) * 2019-07-08 2021-01-14 Indigo Ag, Inc. Crop yield forecasting models
US20220172038A1 (en) * 2020-11-30 2022-06-02 International Business Machines Corporation Automated deep learning architecture selection for time series prediction with user interaction
CN115456314A (en) * 2022-11-11 2022-12-09 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) Atmospheric pollutant space-time distribution prediction system and method
CN116933124A (en) * 2023-06-16 2023-10-24 新奥新智科技有限公司 Time series data prediction method, device, equipment and storage medium

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
LI KUANG等: "Predicting Taxi Demand Based on 3D Convolutional Neural Network and Multi-task Learning", 《REMOTE SENSING》, pages 1 - 18 *
叶允明等: "基于卷积门控循环单元神经网络的 临近预报方法研究", 《高原气象》, pages 411 - 423 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118735082A (en) * 2024-09-03 2024-10-01 南京信息工程大学 3D-TimesNet-based sub-season air temperature prediction correction method

Also Published As

Publication number Publication date
CN117312832B (en) 2024-07-12

Similar Documents

Publication Publication Date Title
CN111784041B (en) Wind power prediction method and system based on graph convolution neural network
CN111091233B (en) Short-term wind power prediction modeling method for wind power plant
CN117312832B (en) Depth sequence model-based medium-and-long-term cloud cover prediction method and system
CN112651290A (en) Hydroelectric generating set vibration trend prediction method and system
CN109800898A (en) A kind of intelligence short-term load forecasting method and system
CN113988477A (en) Photovoltaic power short-term prediction method and device based on machine learning and storage medium
CN111210446A (en) Video target segmentation method, device and equipment
Liao et al. Short‐term wind speed multistep combined forecasting model based on two‐stage decomposition and LSTM
CN113361803A (en) Ultra-short-term photovoltaic power prediction method based on generation countermeasure network
CN116307291B (en) Distributed photovoltaic power generation prediction method and prediction terminal based on wavelet decomposition
CN117154680A (en) Wind power prediction method based on non-stationary transducer model
CN116468166A (en) Deep learning runoff prediction model and method
CN110516792A (en) Non-stable time series forecasting method based on wavelet decomposition and shallow-layer neural network
CN114118401A (en) Neural network-based power distribution network flow prediction method, system, device and storage medium
CN117764617A (en) Property running cost prediction method and device based on big data analysis
CN116307049A (en) Photovoltaic power generation power prediction method, system, terminal and medium
CN111476408B (en) Power communication equipment state prediction method and system
CN116341759A (en) Method, system, equipment and storage medium for predicting short-term power load
CN118069717B (en) Time sequence data characteristic prediction method based on cyclic neural network
CN116805782B (en) New energy output scene generation method and device and computer equipment
CN115964785A (en) High-speed rail bridge Liang Kekao degree calculation important sampling method based on depth generation network
CN117010079A (en) Spacecraft operation state short-time prediction method based on online learning
CN117578428A (en) Fan power recursion prediction method, device, computer equipment and storage medium
CN115936230A (en) Solar output prediction method and device and terminal
CN118052160A (en) Tropical large-scale vertical motion decomposition and simulation system, method, terminal and medium

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