CN116879979A - Snow depth forecasting method and system based on space-time neural network - Google Patents

Snow depth forecasting method and system based on space-time neural network Download PDF

Info

Publication number
CN116879979A
CN116879979A CN202310824139.7A CN202310824139A CN116879979A CN 116879979 A CN116879979 A CN 116879979A CN 202310824139 A CN202310824139 A CN 202310824139A CN 116879979 A CN116879979 A CN 116879979A
Authority
CN
China
Prior art keywords
snow
meteorological
meteorological factor
vector
factor vector
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310824139.7A
Other languages
Chinese (zh)
Inventor
陈圣劼
王禹
吕润清
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.)
Jiang Sushengqixiangtai
Original Assignee
Jiang Sushengqixiangtai
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 Jiang Sushengqixiangtai filed Critical Jiang Sushengqixiangtai
Priority to CN202310824139.7A priority Critical patent/CN116879979A/en
Publication of CN116879979A publication Critical patent/CN116879979A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/10—Devices for predicting weather conditions
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/14—Rainfall or precipitation gauges
    • 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
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • 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
    • G06N3/042—Knowledge-based neural networks; Logical representations of neural 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N3/048—Activation functions
    • 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)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Environmental & Geological Engineering (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Atmospheric Sciences (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Ecology (AREA)
  • Environmental Sciences (AREA)
  • Hydrology & Water Resources (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention providesThe snow depth forecasting method based on the space-time diagram neural network comprises the following steps: data preprocessing is carried out on meteorological factors influencing the depth of snow, and the meteorological factor vectors and the corresponding snow efficiencies of the meteorological factor vectors for N days are assumed to be recorded as follows:wherein x (n) represents a meteorological factor vector on the nth day, y (n) represents snow efficiency on the nth day, and y (n) ∈ [0,1]]The method comprises the steps of carrying out a first treatment on the surface of the And generating a weighted meteorological factor vector from the meteorological factor vector by utilizing a factor attention mechanism, mining multi-scale features in the weighted meteorological factor vector by utilizing convolution modules with different scales, fusing the multi-scale features by adopting a feature connection mode, and outputting snow accumulation efficiency by utilizing a full connection layer to form a complete prediction model. The snow depth forecasting method establishes an objective quantitative forecasting relation by excavating key factors influencing the snow depth.

Description

Snow depth forecasting method and system based on space-time neural network
Technical Field
The invention relates to the field of weather prediction, in particular to a snow depth prediction method and system based on a space-time neural network.
Background
Snow is a common phase of precipitation in winter. A weather process that typically defines a 24 hour day snowfall (melting into water) of 10 mm or more is defined as a snowstorm event. Although the critical value of the precipitation amount of the snow storm is 10 mm, the precipitation amount of 10 mm is totally condensed into solid snow, and when the snow on land is caused, the disaster causing degree is obviously increased. The snowfall process with larger field strength can not only cause direct economic losses such as traffic jams, damage to buildings and equipment, interruption of electric power and communication lines and the like, but also more likely to endanger the life safety of human beings.
The amount characterizing the degree of snowfall is, in addition to the amount of snow and the number of snowfall days, also the snow depth and snow efficiency. The prediction of snow depth is also a key content of snowfall services. Different snow depths have significant differences in the impact of urban operation and agriculture and corresponding disaster defense management measures. However, in the prior art, studies on snow depth and snow efficiency focus on qualitative analysis of single meteorological elements, and more specific quantitative analysis on physical processes affecting snow depth is rarely available; and most of the cases are studied, the conclusion is not necessarily universally representative, and more typical cases are required for supplementary verification.
Snow formation is related to intra-cloud, extra-cloud, and ground processes, etc., where intra-cloud microphysical processes are critical to snow formation. It can be seen that factors affecting snow depth need to consider not only snow fall and air temperature, but also water vapor pressure, ground temperature, etc., and that regional differences exist in various snow depth forecasting relations. The snow depth is more complex than the prediction of the snowfall amount, the snowfall water content of each snowfall may be different, and the temperature conditions may be different, which may cause different snow depths to appear in the same snowfall amount.
In view of this, the present invention has been made.
Disclosure of Invention
In view of the above, the invention discloses a prediction method of snow depth, which establishes an objective quantitative prediction relation by mining key factors influencing the snow depth and provides important reference information for weather first-line business snowfall prediction and disaster prevention and reduction emergency services.
Specifically, the invention is realized by the following technical scheme:
the invention provides a snow depth forecasting method based on a space-time diagram neural network, which comprises the following steps: data preprocessing is carried out on meteorological factors influencing the depth of snow, and the meteorological factor vectors and the corresponding snow efficiencies of the meteorological factor vectors for N days are assumed to be recorded as follows:
wherein x (n) represents a meteorological factor vector on the nth day, y (n) represents snow efficiency on the nth day, and y (n) ∈ [0,1];
and generating a weighted meteorological factor vector from the meteorological factor vector by utilizing a factor attention mechanism, mining multi-scale features in the weighted meteorological factor vector by utilizing convolution modules with different scales, fusing the multi-scale features by adopting a feature connection mode, and outputting snow accumulation efficiency by utilizing a full connection layer to form a complete prediction model.
In the prior art, weather prediction is directly related to the production and life of human society, and is one of the scientific fields of important human research. The traditional weather forecast adopts a very complex numerical weather model, and in recent years, more and more machine learning methods are applied to the field, and the traditional weather forecast abstracts weather forecast into a space-time forecast problem, and attempts are made to solve the problem by a multidimensional time sequence, a Graph Neural Network (GNN) and other methods. In the field of traffic flow prediction, the method of 'GNN+RNN' has achieved a lot of success in recent years, and the graph structure of static slice space is abstract, so that the graph dynamic problem in a space-time prediction scene is effectively solved from the accumulation of ebedding on dynamic time. Similar to traffic flow prediction, weather prediction can also be abstracted as a spatio-temporal prediction problem. However, a simple method of traffic flow prediction is often not effective.
The specificity in the weather forecast field is:
irregularities in data: the meteorological data captured by the various meteorological sensors distributed throughout the area is often irregular, making classical CNNs unsuitable.
High space-time dependence: different terrains exhibit completely different wind flow or temperature transfer models, while extreme climatic events tend to make the data non-compliant with the stationarity conditions.
The climate prediction task differs from the traffic flow prediction task in that each local point is affected by its four-sided eight-way factors, and the local under-space modes should be substantially similar, because in meteorology, heat and wind are freely spread. In the traffic flow prediction task, the mode between two adjacent traffic hubs can be greatly different due to the traffic flow. Snow is the most common scene in weather, and more research is done in the prior art.
In fact, the snow depth and the snow fall do not completely correspond, and the ratio of the snow increment to the snow fall, namely the snow accumulation efficiency, can be used for representing the influence condition of the snow depth under the same snow fall condition by other various factors (Yang, etc., 2013, yin Dongbing, etc., 2009, yang Chengfang, etc., 2015, bai Shuying, etc., 2014, li Dejun, etc., 2014). The research on the snow depth and the snow efficiency is still lacking at present, and the support of the newly added snow depth forecasting technology in China is relatively weak. The international snow depth forecasting technology mainly comprises four main categories: numerical model forecasting, climate, statistical forecasting models, and physical forecasting models (Alcott and Steenburgh, 2010). Middle term of EuropeFor example, a global forecasting mode (hereinafter referred to as ECMWF IFS) of the weather forecasting center is taken as an example, the snow depth forecasting is a part of a land physical process, and the snow depth is obtained through calculation of the surface snow water content and the snow density. The Roebber et al (2003; 2007) establishes a statistical prediction model of the snow-water ratio by a neural network method by analyzing statistical relationships between physical factors such as temperature, humidity, ground wind fields and the like and the snow-water ratio. Cobb and Waldstreicher (2005) by using the statistical relationship between the temperature and the snow-water ratio, a physical prediction model of the snow-water ratio is built by taking the vertical speed as a weight coefficient, and a new snow depth prediction technology is developed by combining the snow fall prediction, and the technology is well applied to the United states weather department. Yang and Xue Jianjun (2013) analyze snowfall versus snow depth using encrypted snowfall data and a linear fitting method: the ratio of the snow depth change value to the corresponding snow fall amount in winter in China is approximately 0.75cm & mm -1 The ratio has a significant trend of decreasing with rising air temperature and has significant regional differences, but does not show significant time variation characteristics. Wang Xiuqin et al (2013) statistically analyze the ground temperature (including the highest and lowest) and the snow temperature (including the highest and lowest) of 0cm and the cloud cover, the sunshine hours and the snow depth based on the actually measured ground temperature and the snow temperature of which the snow depth is more than or equal to 0cm in 2008-2010 of 5 national weather stations in Changji, xinjiang, and find out the relationship between the ground temperature and the snow temperature under different snow depths: when the snow depth is 0-50 cm, the difference between the ground temperature and the snow surface temperature is small and is obviously influenced by the snow depth and weather conditions, the snow depth is 6-400 cm and is mainly influenced by the snow depth, the snow depth exceeds 400cm, and the ground temperature tends to be constant. And (2017) of the auxiliary power and the like, through winter snow observation tests of 2015-2016, snow characteristic parameters such as snow depth, snow layer temperature, snow layer density, snow layer liquid water content and the like under natural conditions are measured, snow characteristic parameter influence factors and development change rules are analyzed, a radial basis neural network snow depth model is constructed, three main meteorological factors influencing the development of the snow parameters such as air temperature, ground surface temperature and water vapor pressure are obtained, and a snow depth forecast model is further built.Ginger mountain etc. (2017) establishes the relationship between the efficiency of the snow cover in south Beijing and the temperature of 2m on the ground: when the air temperature is more than or equal to 2 ℃, the precipitation phase is mainly rain, and the snow accumulation efficiency is almost 0; when the air temperature is more than or equal to 0 ℃ and less than 2 ℃, the precipitation phase state is wet snow, and the snow accumulation efficiency is parabolic along with the increase of the air temperature; when the air temperature is less than 0 ℃, the snowfall phase state is dry snow, and the snow accumulation efficiency is approximately constant 1. In the service forecast, the relation formula and forecast of the air temperature and the snowfall are utilized to forecast the snow depth. Previous studies on the factors affecting snow efficiency have found that low temperature and low humidity are two important conditions for improving snow efficiency. In certain areas, snow is easy to melt due to the conditions of high air humidity and precipitation or rain and snow, or the ground temperature is high, and the snow efficiency is low; the snow depth and the snow amount are in a positive correlation linear relation under the conditions of no liquid precipitation and drier ground (Yin Dongbing, 2009, bai Shuying, 2014, li Dejun, 2014, vivid, 2018, hu Shujuan, 2016). The inefficiency of warm and humid south compared to dry and cold north snow can also be explained by differences in warm and humid conditions (Yao Chen, et al, 2018). In addition, low wind speeds (Li Dejun, etc., 2014; vivid, etc., 2018), positive feedback between net radiation-ground temperature-snow efficiency (Liu Chaodeng, 2018) also favors the increase in snow efficiency. Hu Shujuan (2016) et al also ranked the contributions of factors affecting the change in snow depth in Xinjiang: soil surface temperature>Net radiation>Air temperature>Moisture content of soil volume>Wind speed>Relative humidity of air>Precipitation of water>The size of the water vapor pressure.
The research in the prior art is single, but the effect of the snow depth is various, so that the scheme of the invention exactly overcomes the defects in the prior art, discusses factors affecting the snow depth from various aspects and multiple dimensions, and predicts the snow depth more accurately.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments, wherein the accompanying drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention, and wherein like reference numerals represent like parts throughout the several views, and wherein:
FIG. 1 is a diagram of a specific architecture for constructing a weighted meteorological factor vector according to an embodiment of the present invention;
FIG. 2 is a diagram showing a specific structure for predicting snow efficiency according to an embodiment of the present invention;
fig. 3 is a schematic flow chart of a computer device according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings, wherein like reference numerals in the various drawings indicate identical or similar elements unless otherwise indicated, and wherein the implementations described in the following exemplary embodiments do not represent all implementations consistent with the present disclosure, but rather are merely examples of apparatuses and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims,
the terms used in this disclosure are used solely for the purpose of describing particular embodiments and are not intended to be limiting of the disclosure, as used in the present disclosure and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise, it being understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items,
it should be understood that although the terms first, second, third, etc. may be used in this disclosure to describe various information, these information should not be limited to these terms, which are merely used to distinguish one type of information from another, e.g., a first information may also be referred to as a second information, and similarly a second information may also be referred to as a first information, depending on context, as the term "if" as used herein may be interpreted as "at … …" or "at … …" or "in response to a determination",
the scheme of the invention has the following thought:
first, the snow depth data of the whole province in a distributed manner is abstracted into a plane data structure, coordinates are given, and the region to be predicted of the whole province is equally divided into a plurality of local spaces which are related to the actual geographic region and Euclidean distance (for example, each grid is defined by a maximum coordinate and a minimum coordinate in a grid form). In this way, the snow data in each cell area is studied. Each grid is then considered as a vertex of the graph for constructing a graph model of the snow depth prediction.
In terms of spatial modeling, in order to better simulate local consistency in meteorology, the influence of geographic locations on the weather is considered, and regions with similar properties have high similarity in snow depth even if separated by far distances. Each mesh is thus considered a node of a graph and based thereon a geographical structure is built to capture spatial features: the center point of each grid is regarded as the geographic position center of the grid, the distance of the center point is regarded as the edge weight of the geographic graph structure, the closer the distance is, the smaller the weight is, and the snow depth between the center point and the edge weight has certain similarity.
In the aspect of time modeling, historical snow features are considered, time characteristics are captured through LSTM and an attention mechanism, snow depth change in a time dimension is mastered, and data between each pair of grids can be predicted.
The network structure design adopts an improved MGSTGCN based on a local condition graph convolution kernel (location-characterized kernel) so as to meet three conditions of adjacent convolution kernels sharing under similar local characteristics and different geographic characteristics of the local condition convolution kernels. The time architecture part of MGSTGCN has the input gate, forget gate and output gate of LSTM as LSTM, but is derived from graph convolution operator and introduces a mechanism of attention, where the time sequence is input. The temporal structure in combination with the spatial structure forms the MGSTGCN network.
Specifically, the scheme of the invention provides a snow depth forecasting method based on a space-time neural network, which comprises the following steps:
data preprocessing is carried out on meteorological factors influencing the depth of snow, and the meteorological factor vectors and the corresponding snow efficiencies of the meteorological factor vectors for N days are assumed to be recorded as follows:
wherein x (n) represents a meteorological factor vector on the nth day, y (n) represents snow efficiency on the nth day, and y (n) ∈ [0,1];
and generating a weighted meteorological factor vector from the meteorological factor vector by utilizing a factor attention mechanism, mining multi-scale features in the weighted meteorological factor vector by utilizing convolution modules with different scales, fusing the multi-scale features by adopting a feature connection mode, and outputting snow accumulation efficiency by utilizing a full connection layer to form a complete prediction model.
Specifically, the method comprises the following steps:
step 1: data preprocessing
Recording meteorological factor vector as
x＝{x 1 ，x 2 ，…，x 20 }
Wherein x is 1， x 2 ，，x 20 The meteorological factors represented respectively are shown in table 1 below.
TABLE 1 Meteorological factors
Assuming that there are N days of meteorological factor vectors and their corresponding snow efficiencies, it can be noted that:
wherein x (n) represents a meteorological factor vector on the nth day, y (n) represents snow efficiency on the nth day, and y (n) ∈ [0,1]. In order to facilitate the following neural network to predict snow efficiency according to the meteorological factor vector, each factor in the meteorological factor vector needs to be normalized, and the normalization formula can be expressed as
Wherein x is i (n) represents the value of the ith meteorological factor on the nth day. Then, the normalized weather factor vector and its corresponding set of snow efficiencies may be expressed as
Step 2: construction of accumulated snow efficiency prediction model based on characteristic attention mechanism and multi-scale convolutional neural network
2-1: generating weighted meteorological factor vectors using factor attention mechanism
Because of the different degrees of association of different meteorological factors with snow efficiency, this embodiment adopts a factor attention mechanism to assign different weights to each meteorological factor, and constructs a weighted meteorological factor vector, as shown in fig. 1.
(1) Generating weight vectors
The weight vector is generated by adopting the method of the attention mechanism, and the process is as follows:
wherein lambda (n) represents a weight vector corresponding to the weather factor vector on the nth day; f (f) Att (W Att ) The model of the attention neural network comprises 3 fully connected layers, the number of the neurons of which is 20 in sequence,20 where r.epsilon. {2,4,5, 10} represents the factor compression rate and the layer 1 and layer 2 fully connected layers use ReLU as the activation function, while the layer 3 fully connected layer adopts the value of weight vector between 0-1 to be outputSigmoid is used as the activation function. The 3 full-connection layers can be expressed as:
wherein the method comprises the steps ofRespectively represent the weight and bias of the j-th full connection layer,>
(2) Generating weighted weather factors
The normalized meteorological factor vector is multiplied by the weight vector correspondingly to obtain a weighted meteorological factor vector, which is specifically shown as follows:
wherein, the ". If indicates Hadamard product.
2-2: snow accumulation efficiency prediction based on weighted meteorological factor vector and multi-scale convolutional neural network
The step adopts a multi-scale convolutional neural network to predict snow accumulation efficiency. Firstly, utilizing convolution modules with different scales to mine multi-scale features in meteorological factor vectors; fusing the multi-scale features by adopting a feature connection mode; finally, the snow collecting efficiency is outputted by using the full connection layer, and the snow collecting efficiency is shown in the figure 2.
(1) Constructing a multi-scale convolution module
The four groups of one-dimensional convolutions (convolution kernel scales are 3/5/7 respectively, the number of neurons is 128, and the activation functions are ReLU) are adopted to parallelly mine the characteristics in the meteorological factor vector. This process can be expressed as:
wherein the method comprises the steps ofIndicating a convolution kernel scale of S k Weights of one-dimensional convolution of v k Features representing its output, with dimensions 128 x (20-S k +1)。
(2) Fusing multiscale features
The patent adopts a characteristic splicing mode to fuse the characteristics of four groups of one-dimensional convolution outputs. First, v is pooled using global maximization k Turning to a one-dimensional feature vector, the process can be expressed as:
wherein, the one-dimensional feature vector is obtained by conversionIs 128 x 1. Then, the four sets of feature vectors are spliced to obtain a multi-scale fusion feature, which can be expressed as
(3) Efficiency of snow output
Based on the multi-scale fusion characteristics, 3 layers of full-connection layers (the number of neurons is 128/64/1 in sequence, the activation function of the two is ReLU, and the activation function of the two is Sigmoid) are utilized to output predicted snow accumulation efficiency.
Step 3: prediction model for snow accumulation efficiency of training characteristic attention mechanism and multi-scale convolutional neural network
In order to improve prediction accuracy, the method combines an optimization factor attention mechanism and a multi-scale convolutional neural network in an end-to-end mode, and does not separate and optimize alone.
3-1 cascade factor attention mechanism model and multi-scale convolutional neural network
Taking the output of the factor attention mechanism model as the input to the multi-scale convolutional neural network, the process can be expressed as:
wherein f MSCNN (W MSCNN ) Representing the multi-scale convolutional neural network described above, and f (W) representing a cascade of factor attention mechanisms and the multi-scale convolutional neural network.
3-2 division of training verification data, setting objective function
Will beRandomly dividing into training data set and verification data set according to the ratio of 9:1, wherein the training data set and the verification data set are respectively +.>And->And adopts cross entropy as an objective function, which can be expressed as:
3-3 selection optimization algorithm
The objective function is optimized by adopting random gradient descent, and the iterative formula is as follows:
wherein eta t Representing the learning rate of the t-th iteration, the present patent optimizes the above objective function with a dynamic learning rate aimed at obtaining more accurate predictive performance, which can be expressed as:
setting an initial learning rate eta 0 =0.001, a learning rate is reduced by 20% for each 10 iteration cycles;representing the gradient. Setting the total iteration number as 200, and taking the loss function value on the verification set within the total iteration numberT∈[1，200]Minimum weight W T Is the final model weight.
Step 4: predicting snow efficiency on day n+p
First, the meteorological factor vector on the n+p day is x (n+p) = { x 1 (N+P)，x 2 (N+P)，...，x 20 (n+p) } normalization as follows:
secondly, the weather factor vector on the n+P day after normalization is input to the weight W T In the snow efficiency prediction model of (2), the predicted snow efficiency on the n+p day is obtained as follows:
the invention provides a snow depth forecasting method and a snow depth forecasting system, which specifically comprises the following steps:
and a pretreatment module: the method is used for preprocessing data of meteorological factors affecting the depth of snow, and is characterized in that the meteorological factor vectors and the corresponding snow efficiencies of the meteorological factor vectors for N days are assumed to be:
wherein x (n) represents a meteorological factor vector on the nth day, y (n) represents snow efficiency on the nth day, and y (n) ∈ [0,1];
snow forecasting module: and generating a weighted meteorological factor vector from the meteorological factor vector by utilizing a factor attention mechanism, mining multi-scale features in the weighted meteorological factor vector by utilizing convolution modules with different scales, fusing the multi-scale features by adopting a feature connection mode, and outputting snow accumulation efficiency by utilizing a full connection layer.
In the implementation, each module may be implemented as an independent entity, or may be combined arbitrarily, and implemented as the same entity or several entities, and the implementation of each unit may be referred to the foregoing method embodiment, which is not described herein again.
Fig. 3 is a schematic structural diagram of a computer device according to the present disclosure. Referring to FIG. 3, the computer device 400 includes at least a memory 402 and a processor 401; the memory 402 is connected to the processor through the communication bus 403, and is configured to store computer instructions executable by the processor 401, where the processor 401 is configured to read the computer instructions from the memory 402 to implement the steps of the snow depth forecasting method according to any of the foregoing embodiments.
For the above-described device embodiments, reference is made to the description of the method embodiments for the relevant points, since they essentially correspond to the method embodiments. The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the objectives of the disclosed solution. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
Computer readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices including, for example, semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices), magnetic disks (e.g., internal magnetic disks or removable disks), magneto-optical disks, and CD-ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.
Finally, it should be noted that: while this specification contains many specific implementation details, these should not be construed as limitations on the scope of any invention or of what may be claimed, but rather as descriptions of features of specific embodiments of particular inventions. Certain features that are described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. On the other hand, the various features described in the individual embodiments may also be implemented separately in the various embodiments or in any suitable subcombination. Furthermore, although features may be acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.
Similarly, although operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In some cases, multitasking and parallel processing may be advantageous. Moreover, the separation of various system modules and components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.
Thus, particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve desirable results. Furthermore, the processes depicted in the accompanying drawings are not necessarily required to be in the particular order shown, or sequential order, to achieve desirable results. In some implementations, multitasking and parallel processing may be advantageous.
The foregoing description of the preferred embodiments of the present disclosure is not intended to limit the disclosure, but rather to cover all modifications, equivalents, improvements and alternatives falling within the spirit and principles of the present disclosure.

Claims (10)

1. A snow depth forecasting method based on a space-time neural network is characterized by comprising the following steps:
data preprocessing is carried out on meteorological factors influencing the depth of snow, and the meteorological factor vectors and the corresponding snow efficiencies of the meteorological factor vectors for N days are assumed to be recorded as follows:
wherein x (n) represents a meteorological factor vector on the nth day, y (n) represents snow efficiency on the nth day, and y (n) ∈ [0,1];
and generating a weighted meteorological factor vector from the meteorological factor vector by utilizing a factor attention mechanism, mining multi-scale features in the weighted meteorological factor vector by utilizing convolution modules with different scales, fusing the multi-scale features by adopting a feature connection mode, and outputting snow accumulation efficiency by utilizing a full connection layer to form a complete prediction model.
2. The snow depth forecasting method of claim 1, wherein each factor in the meteorological factor vector is normalized, and the normalization formula is expressed as:
wherein x is i (n) represents the value of the ith meteorological factor on the nth day, and the normalized meteorological factor vector and the corresponding aggregate of snow efficiencies are represented as
3. The snow depth forecasting method of claim 2, characterized in that the method of generating a weighted weather factor vector from the weather factor vector comprises the steps of:
wherein lambda (n) represents a weight vector corresponding to the weather factor vector on the nth day; f (f) Att (W Att ) The model of the attention neural network comprises 3 fully connected layers, the number of the neurons of which is 20 in sequence,20，r∈[2，4，5，10]Representing the factor compression rate, wherein the 1 st and 2 nd full-connection layers adopt a ReLU as an activation function, and the 3 rd full-connection layer adopts a Sigmoid as an activation function;
the 3 full-connection layers are expressed as follows:
wherein the method comprises the steps ofRespectively represent the weight and bias of the j-th full connection layer,>
correspondingly multiplying the normalized meteorological factor vector with the weight vector to obtain a weighted meteorological factor vector, which is specifically shown as follows:wherein, the ". If indicates Hadamard product.
4. A method of snow depth prediction according to claim 3, characterized in that the method of forming the multiscale features comprises:
the features in the meteorological factor vectors are parallelly mined by adopting four groups of one-dimensional convolutions, and the process is expressed as follows:
wherein the method comprises the steps ofIndicating a convolution kernel scale of S k Weights of one-dimensional convolution of v k Features representing its output, with dimensions 128 x (20-S k +1)。
5. The snow depth forecasting method according to claim 4, wherein the method for fusing the multi-scale features comprises:
utilizing global maximum poolsChemical transformation of v k Turning to a one-dimensional feature vector, the process is expressed as:
wherein, the one-dimensional feature vector is obtained by conversionIs 128 x 1;
then, the four sets of feature vectors are spliced to obtain the multi-scale fusion feature, which is expressed as
6. The snow depth prediction method according to claim 5, wherein the snow efficiency calculating method comprises:
based on multiscale fusion characteristics, the snow accumulation efficiency predicted by using 3 layers of full-connection layers is output:
7. the snow depth forecasting method of claim 6, further comprising a method of optimizing the predictive model, the steps comprising:
taking the output of the factor attention mechanism model as the input of the multi-scale convolutional neural network, the process is expressed as:
wherein f MSCNN (W MSCNN ) Representing the multiple rulerA degree convolutional neural network, and f (W) represents a cascade of factor attention mechanisms and a multi-scale convolutional neural network;
will beRandomly dividing the training data set and the verification data set into a training data set and a verification data set according to the proportion of 9:1, wherein the training data set and the verification data set are respectivelyAnd->And adopts cross entropy as an objective function, expressed as:
the objective function is optimized by adopting random gradient descent, and the iterative formula is as follows:
wherein eta t Representing the learning rate of the t-th iteration, the dynamic learning rate is expressed as:
setting an initial learning rate eta 0 =0.001, a learning rate is reduced by 20% for each 10 iteration cycles;representing the gradient, setting the total iteration number to be 200, and taking the loss function value on the verification set within the total iteration numberMinimum weight W T Weighting the final model;
the meteorological factor vector on the n+P day is x (n+P) = { x 1 (N+P)，x 2 (N+P)，...，x 20 (n+p) } normalization as follows:
secondly, the weather factor vector on the n+P day after normalization is input to the weight W T In the snow efficiency prediction model of (2), the predicted snow efficiency on the n+p day is obtained as follows:
8. the system of any one of claims 1-7, comprising:
and a pretreatment module: the method is used for preprocessing data of meteorological factors affecting the depth of snow, and is characterized in that the meteorological factor vectors and the corresponding snow efficiencies of the meteorological factor vectors for N days are assumed to be:
wherein x (n) represents a meteorological factor vector on the nth day, y (n) represents snow efficiency on the nth day, and y (n) ∈ [0,1];
snow forecasting module: and generating a weighted meteorological factor vector from the meteorological factor vector by utilizing a factor attention mechanism, mining multi-scale features in the weighted meteorological factor vector by utilizing convolution modules with different scales, fusing the multi-scale features by adopting a feature connection mode, and outputting snow accumulation efficiency by utilizing a full connection layer.
9. A computer readable storage medium having stored thereon a computer program, characterized in that the program when executed performs the steps of the snow depth forecasting method according to any one of claims 1-7.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the snow depth forecasting method according to any one of claims 1-7 when executing the program.
CN202310824139.7A 2023-07-05 2023-07-05 Snow depth forecasting method and system based on space-time neural network Pending CN116879979A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310824139.7A CN116879979A (en) 2023-07-05 2023-07-05 Snow depth forecasting method and system based on space-time neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310824139.7A CN116879979A (en) 2023-07-05 2023-07-05 Snow depth forecasting method and system based on space-time neural network

Publications (1)

Publication Number Publication Date
CN116879979A true CN116879979A (en) 2023-10-13

Family

ID=88265580

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310824139.7A Pending CN116879979A (en) 2023-07-05 2023-07-05 Snow depth forecasting method and system based on space-time neural network

Country Status (1)

Country Link
CN (1) CN116879979A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117574960A (en) * 2023-12-14 2024-02-20 江苏省气象台 Multi-factor plum rainfall prediction method based on self-adaptive graph structure and reinforcement integration

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117574960A (en) * 2023-12-14 2024-02-20 江苏省气象台 Multi-factor plum rainfall prediction method based on self-adaptive graph structure and reinforcement integration

Similar Documents

Publication Publication Date Title
Sharma et al. Forecasting daily global solar irradiance generation using machine learning
Yu et al. Prediction of the temperature in a Chinese solar greenhouse based on LSSVM optimized by improved PSO
Guermoui et al. A novel hybrid model for solar radiation forecasting using support vector machine and bee colony optimization algorithm: review and case study
CN103955606B (en) A kind of grassland based on remote sensing technology gradual Forecasting Methodology of the plague of locusts
Dong et al. Enhancing short-term forecasting of daily precipitation using numerical weather prediction bias correcting with XGBoost in different regions of China
CN112288164B (en) Wind power combined prediction method considering spatial correlation and correcting numerical weather forecast
Kisi et al. Estimation of dew point temperature using neuro-fuzzy and neural network techniques
Ma et al. Wheat powdery mildew forecasting in filling stage based on remote sensing and meteorological data
Awad et al. Predicting the energy production by solar photovoltaic systems in cold-climate regions
de Jesús et al. Solar pv power prediction using a new approach based on hybrid deep neural network
CN116879979A (en) Snow depth forecasting method and system based on space-time neural network
Dhillon et al. A solar energy forecast model using neural networks: Application for prediction of power for wireless sensor networks in precision agriculture
Vakitbilir et al. Hybrid deep learning models for multivariate forecasting of global horizontal irradiation
Park et al. Stacking deep transfer learning for short-term cross building energy prediction with different seasonality and occupant schedule
CN113553764A (en) Mountain fire prediction method based on deep learning network
Bushara et al. Computational intelligence in weather forecasting: A review
Hameed et al. Inflow forecasting using regularized extreme learning machine: Haditha reservoir chosen as case study
Bhawsar et al. A survey of weather forecasting based on machine learning and deep learning techniques
Gündoğdu et al. Application of feed forward and cascade forward neural network models for prediction of hourly ambient air temperature based on MERRA-2 reanalysis data in a coastal area of Turkey
Li et al. A stepwise clustered hydrological model for addressing the temporal autocorrelation of daily streamflows in irrigated watersheds
Nazeri-Tahroudi et al. Estimation of dew point temperature in different climates of Iran using support vector regression
Allawi et al. Monthly rainfall forecasting modelling based on advanced machine learning methods: Tropical region as case study
Mohd et al. Comparative study of rainfall prediction modeling techniques (A case study on Srinagar, J&K, India)
Gupta et al. Solar irradiation forecasting technologies: a review
Ren et al. Downscaling of daily extreme temperatures in the Yarlung Zangbo River Basin using machine learning techniques

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