CN111784023A - Short-time adjacent fog visibility prediction method - Google Patents
Short-time adjacent fog visibility prediction method Download PDFInfo
- Publication number
- CN111784023A CN111784023A CN202010409474.7A CN202010409474A CN111784023A CN 111784023 A CN111784023 A CN 111784023A CN 202010409474 A CN202010409474 A CN 202010409474A CN 111784023 A CN111784023 A CN 111784023A
- Authority
- CN
- China
- Prior art keywords
- fog
- data
- prediction
- time
- visibility
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 145
- 238000007781 pre-processing Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims description 91
- 238000003066 decision tree Methods 0.000 claims description 36
- 230000006870 function Effects 0.000 claims description 21
- 239000000654 additive Substances 0.000 claims description 17
- 230000000996 additive effect Effects 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 11
- 238000004140 cleaning Methods 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 8
- 238000013507 mapping Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000002349 favourable effect Effects 0.000 description 3
- 238000012417 linear regression Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 238000003672 processing method Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Human Resources & Organizations (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Economics (AREA)
- Artificial Intelligence (AREA)
- Game Theory and Decision Science (AREA)
- General Business, Economics & Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A short-time adjacent fog visibility prediction method comprises the following steps: the method comprises the steps of firstly, acquiring real-time data related to fog visibility prediction; secondly, preprocessing the real-time data in the first step; and thirdly, inputting the data preprocessed in the second step into a model for predicting fog visibility at different early warning times, which is trained by an XGboost method, to obtain a predicted fog visibility value of each piece of real-time data at different prediction times. The invention well solves the problem of the forecast of the fog. The difficulty of the fog forecast is reduced and the fog forecast is more accurate.
Description
Technical Field
The invention relates to the field of meteorology, in particular to a short-time adjacent fog visibility prediction method.
Background
Fog is a common type of disastrous weather, and is caused by a weather phenomenon in which a large number of minute water droplets in the air float in the air, resulting in a fog visibility of less than 1 km. At present, the forecast of the fog is issued by the meteorological bureau based on real-time data of a live field, and subsequent early warning measures are not effectively prevented, so that the forecast of the fog becomes a problem to be solved. However, the difficulty of forecasting is not only that the complicated factors considered by fog forecasting are numerous, that is, many influencing factors change in real time, but also that a large amount of computing power and an accurate computing model are required.
Disclosure of Invention
The purpose of the invention is as follows:
the invention aims to solve the technical problem that the technical defect that the prediction of the visibility of fog at different short-time adjacent moments is difficult to complete in the prior art, and provides the short-time adjacent fog visibility prediction of XGboost and an improved linear regression method.
The technical scheme is as follows:
a short-time adjacent fog visibility prediction method is characterized by comprising the following steps: comprises the following steps:
the method comprises the steps of firstly, acquiring real-time data related to fog visibility prediction;
secondly, preprocessing the real-time data in the first step;
and thirdly, inputting the data preprocessed in the second step into a model for predicting fog visibility at different early warning times, which is trained by an XGboost method, to obtain a predicted fog visibility value of each piece of real-time data at different prediction times.
In a first step, real-time data relating to the prediction of fog visibility includes actually collected data of air pressure, temperature and relative humidity for the predicted time period.
And the second step of preprocessing the real-time data comprises the steps of sequentially carrying out data cleaning, data merging on all stations and minimizing the similar fog visibility prediction attribute data according to the actual fog visibility prediction attribute threshold.
The model construction method for predicting fog visibility in the third step is as follows:
s1, acquiring and processing an original data set related to fog visibility prediction;
s2, preprocessing the original data in the step S1 to obtain all characteristic attribute data related to fog visibility prediction;
s3, marking to obtain all characteristic attribute data related to fog visibility prediction, and establishing fog visibility prediction sample training data at different early warning times according to the marks corresponding to the prediction range;
and S4, training the fog visibility prediction samples at different moments of the training data in the step S3 by using an XGboost method, obtaining the relation between each predicted fog visibility value and a meteorological attribute feature space at different early warning times, and establishing fog visibility prediction models at different early warning times according to the relation.
S2, the method for obtaining the new fog visibility prediction attribute in the step comprises the following steps: pretreatment: the real-time data method comprises the following steps of sequentially carrying out data cleaning, station data merging and similar fog visibility prediction attribute data minimization according to an actual fog visibility prediction attribute threshold value:
data addition characteristics: adding different time attributes to each fog visibility prediction data attribute, and calculating a new attribute by using the relationship among the attributes of the fog visibility prediction data;
the labeling method in the step S3 is as follows: determining the fog moment according to a specific threshold value; and marking the fog visibility value of the fog visibility prediction data according to the obtained fog-existence moments and different early warning times, and obtaining fog visibility prediction training samples at different early warning times.
Actual data attributes relevant for fog visibility prediction include barometric pressure, temperature, and relative humidity.
The method for obtaining all characteristic attribute data related to the fog visibility prediction comprises the following steps:
the method for establishing the time window adds different time attributes to each fog visibility prediction data attribute, the method for establishing the time window obtains a new time by subtracting the time corresponding to the current data from the corresponding threshold range according to each set time window model threshold range, finds out the line number of the data corresponding to the data set according to the obtained new time, and adds some fog visibility prediction attributes in the line data into the current time as the new attributes of the current time;
the time window model threshold range is set to one hour and two hours; the method for calculating the relationship among the attributes comprises the steps of subtracting the attributes at the same moment in each piece of data and calculating the ratio; the new attributes obtained from each of the fog visibility prediction data include temperature dew point difference, temperature dew point ratio, and span temperature variation.
In the step S3, the method for establishing different early warning time fog prediction training samples includes:
firstly, determining the time with or without fog, wherein the determining method is to meet the requirement of a certain time period, set a threshold value for the characteristics of fog and fog, and judge the initial time of the time period as the time with or without fog if the data of the time period meet the requirement of the threshold value;
respectively establishing fog visibility prediction training samples at different moments at the moment of judging whether fog exists and the early warning moment corresponding to the moment of judging whether fog exists, subtracting the range of the early warning time threshold from the moment corresponding to the fog data of each station to obtain a new moment for each early warning time threshold, marking the data of each station corresponding to the obtained new moment, and forming a fog prediction training sample data set of appointed early warning time by all characteristic data of the marking and the corresponding moment; the label of the fog prediction training sample represents the fog visibility value condition of the attribute data at the moment in the future early warning time, and the data set is used for the next XGboost method and model training;
in the step S4, the XGBoost method is a method for predicting visibility in fog in a future time period by training sample data, that is, the fog visibility prediction sample data at different early warning moments are respectively input to the XGBoost for training, a threshold segmentation point related to the division of each meteorological attribute feature value and a leaf node related to the visibility value are obtained by using each round of additive training fitting residual error and greedy algorithm, and each sub-tree is formed, each obtained sub-tree corresponds to a feature space, and the integrated learning method model based on trees is used for analyzing each meteorological attribute while predicting the visibility in fog.
The expression of the XGboost is shown as formula (1); the XGboost is used for fitting data in each round of training through an addition model, an objective function of the XGboost adopts an expression mode of Taylor second-order expansion, and the Taylor expansion of the objective function is shown as the following formula (2); optimizing an objective function and dividing a feature space by adopting a greedy algorithm to obtain an optimal residual error of each round; the specific formula of the optimal subtree of each XGboost formed in each round is shown as (3);
(1) (2) and (3) respectively representing a general expression of the XGboost, an objective function based on Taylor second-order expansion and an optimal decision tree in each round, wherein k represents the number of the decision trees forming the XGboost integrated learning method, F represents the decision tree space forming the XGboost, X represents training sample data, T represents the number of leaf nodes in the XGboost, w represents the number of the leaf nodes in the XGboostjRepresenting the score of the leaf j node. Where λ is the regularization coefficient and γ is the controlNumber coefficient of leaf nodes, prevention of overfitting, giFor the first derivative, h, of the new decision tree under XGboost additive trainingiFor the second derivative of the iterative decision tree under XGboost additive training, IjFor the sample where the jth leaf node in the kth leaf is located, fkRepresenting the kth sub-tree constituting the method of the integration tree.
The advantages and effects are as follows:
the invention constructs a short-time adjacent fog visibility prediction method of an XGboost method and an improved linear regression method, which comprises the following steps:
s1, acquiring and processing an original data set related to fog visibility prediction; the original data come from data sets of different years, different times and different sites; the processing method comprises the steps of data cleaning, and merging data according to sites;
s2, marking and processing the original data set to obtain all characteristic attribute data related to fog visibility prediction, and forming fog visibility prediction training data at different early warning times; the method for obtaining the new fog visibility prediction attribute comprises the following steps: adding different time attributes to each fog visibility prediction data attribute, and calculating a new attribute by using the relationship among the attributes of the fog visibility prediction data; determining the fog moment according to a specific threshold value; marking the fog visibility value of the fog visibility prediction data according to the obtained moment and different early warning times, and obtaining fog visibility prediction training samples of different early warning times;
the method comprises the steps of S3, XGboost, training by using fog visibility prediction samples at different moments to obtain the relation between each predicted fog visibility value and meteorological attribute feature space at different early warning time, and establishing fog visibility prediction models at different early warning time;
s4, during data prediction, inputting each piece of preprocessed real-time data into trained models for predicting fog visibility at different early warning times respectively to obtain a fog visibility predicted value of each piece of real-time data at different early warning times;
furthermore, in the method for predicting the visibility of the nearby fog in the short time, the original data of the visibility prediction of the fog is obtained by downloading the meteorological attributes of all stations related to the visibility prediction of the fog in different years from ground observation data of a meteorological data unified service interface (CIMISS) and obtaining the original data of the visibility prediction of the fog;
the obtained original data attributes of the fog visibility prediction comprise air pressure, temperature, relative humidity and the like;
further, in the method for predicting the visibility of the nearby fog in the short time, the method for processing the original data comprises the steps of cleaning the data according to a specific original visibility prediction attribute threshold value of the fog, merging the data of all stations and minimizing the similar visibility prediction attribute data;
furthermore, in the method for predicting the visibility of the nearby fog in the short time of the present invention, the method for adding the attributes of different moments to the attributes of each predicted data of the visibility of the fog is to obtain a new moment by establishing a time window model and subtracting the corresponding time of the current data from the corresponding threshold range according to each set time threshold range, find out the number of rows of the data corresponding to the data set according to the obtained new moment, and add some predicted attributes of the visibility of the fog in the row as the new attributes of the current moment to the current moment;
the time window model threshold range is set to 1 hour and 2 hours;
further, in the method for predicting the visibility of the nearby fog in the short time, the calculation of the relation among the attributes of the predicted data of the visibility of the fog comprises subtraction calculation and ratio calculation of the attributes at the same moment in each piece of data;
subtraction of attributes at different times for each piece of data (for calculating the amount of change);
the new attributes obtained from each fog visibility prediction data comprise a plurality of characteristics such as temperature dew point difference, temperature dew point ratio, span temperature change and the like;
the original attribute features and the new attribute features jointly represent all meteorological elements which can influence the prediction of the visibility of the fog;
further, in the method for predicting visibility of nearby fog for a short time of the present invention, the method for determining whether fog exists or not is to satisfy the requirement of a certain time period, set a threshold for the fog exists or not, and determine that the starting time of the time period is the fog existing or not if the data of the time period satisfies the threshold requirement;
furthermore, in the method for predicting the visibility of the nearby fog in the short time, the method for establishing the training set of the fog visibility prediction samples at different moments subtracts the range of the early warning time threshold from the moment corresponding to the fog data of each station to obtain a new moment, labels the data of each station corresponding to the obtained new moment, and the labels and the corresponding data form a training sample data set of the fog visibility prediction samples at the appointed early warning time;
the label of the fog visibility prediction training sample represents the fog visibility value condition of the attribute data at the moment in the future early warning time, and the data set is used for the next training of a fog visibility prediction model;
the prediction of different early warning time thresholds considered by the patent comprises the prediction of half an hour, 1 hour, 2 hours and 3 hours in the future of the fog, and fog visibility prediction training samples with different early warning time thresholds are respectively established;
furthermore, in the method for predicting visibility of the short-term neighboring fog, the XGboost method is a method for predicting visibility of the fog in a future time period through training sample data, namely the fog visibility prediction sample data at different early warning moments are respectively input into the XGboost for training, each round of additive training fitting residual error and greedy algorithm are utilized to obtain threshold segmentation points related to the division of the characteristic values of the meteorological attributes and leaf nodes related to the visibility values, each sub-tree is formed, each obtained sub-tree corresponds to one characteristic space, and the fog visibility prediction is realized while each meteorological attribute is analyzed;
the XGboost method is an integrated learning method, a plurality of decision trees are integrated to fit the residual error of each decision tree in each round, leaf nodes of each feature space corresponding to a specific fog visibility value are obtained and are divided into feature spaces, and the expression of the XGboost is shown as (1);
the XGboost is used for fitting data in each round of training through an addition model, an objective function of the XGboost adopts an expression mode of Taylor second-order expansion, so that the XGboost is more favorable for training fitting errors, and a Taylor expansion formula of the objective function is shown as the following formula (2);
optimizing an objective function and dividing a feature space by adopting a greedy algorithm to obtain an optimal residual error of each round; the specific formula of the optimal subtree of each XGboost formed in each round is shown as (3);
(1) and (2) and (3) respectively represent a general expression of the XGboost, an objective function based on Taylor second-order expansion and an optimal decision tree in each round, wherein k represents the number of the decision trees forming the XGboost integrated learning method, F represents the decision tree space forming the XGboost, X represents training sample data, T represents the number of leaf nodes in the XGboost, and w represents the fraction of the leaf nodes. Where λ is the regularization coefficient, γ is the coefficient controlling the number of leaf nodes, preventing overfitting, giFor the first derivative, h, of the new decision tree under XGboost additive trainingiFor the second derivative of the iterative decision tree under XGboost additive training, IjIs the sample of the jth leaf node in the kth leaf; (ii) a
Furthermore, in the method for predicting the visibility of the short-term adjacent fog, the process of analyzing and predicting the visibility of the fog and the meteorological characteristic space by the XGboost method is to establish good fog visibility prediction training samples aiming at different early warning times and respectively obtain a mapping model of the predicted fog visibility and the meteorological characteristic space at each early warning time by using the XGboost method;
further, in the method for predicting the visibility of the short-time adjacent fog, the specific prediction method includes preprocessing each piece of real-time data of each station, and inputting the preprocessed real-time data into trained models for predicting the visibility of the fog at different early warning times to obtain predicted values of the visibility of each piece of real-time data at different early warning times;
further, in the method for predicting the visibility of the short-time adjacent fog, the specific prediction method includes preprocessing each piece of real-time data of each station, and inputting the preprocessed real-time data into trained models for predicting the visibility of the fog at different early warning times to obtain predicted values of the visibility of each piece of real-time data at different early warning times;
further comprising the steps of:
and (4) acquiring data for testing, inputting the testing data into the model by using the XGboost method model obtained in the steps S1-S4, and comparing the obtained fog visibility prediction result with the label to obtain the fitting degree of the fog visibility prediction.
The invention well solves the problem of forecasting the fog, so that the difficulty of forecasting the fog is reduced. By predicting the visibility, the fog trend at the future moment can be accurately analyzed from the angle of the value, and the weather personnel can make effective judgment conveniently.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a topological structure diagram of the combined model of the present invention in real-time data prediction;
FIG. 2 is a flow chart of a method of establishing fog visibility prediction data;
FIG. 3 is a flow diagram of the XGboost method;
FIG. 4 is a block diagram of the XGboost method;
Detailed Description
A short-time adjacent fog visibility prediction method comprises the following steps:
the method comprises the steps of firstly, acquiring real-time data related to fog visibility prediction;
secondly, preprocessing the real-time data in the first step;
and thirdly, inputting the data preprocessed in the second step into a model for predicting fog visibility at different early warning times, which is trained by an XGboost method, to obtain a predicted fog visibility value of each piece of real-time data at different prediction times.
In a first step, real-time data relating to the prediction of fog visibility includes actually collected data of air pressure, temperature and relative humidity for the predicted time period.
The method for preprocessing the real-time data in the second step comprises the steps of sequentially cleaning the data, merging the data of all the stations and minimizing the similar fog visibility prediction attribute data according to the actual fog visibility prediction attribute threshold (the threshold and the related minimization are based on expert knowledge of meteorology).
The model construction method for predicting fog visibility in the third step is as follows:
s1, acquiring and processing an original data set related to fog visibility prediction; the original data sets come from different years, different times and different observation sites; the method for processing the original data comprises the steps of clearing error weather attribute values, merging data according to sites and the like;
s2, preprocessing the original data in the step S1 to obtain all characteristic attribute data related to fog visibility prediction;
s3, marking to obtain all characteristic attribute data related to fog visibility prediction, and establishing fog visibility prediction sample training data at different early warning times according to the marks corresponding to the prediction range;
and S4, training the fog visibility prediction samples at different moments of the training data in the step S3 by using an XGboost method, obtaining the relation between each predicted fog visibility value and a meteorological attribute feature space at different early warning times, and establishing fog visibility prediction models at different early warning times according to the relation.
S2, the method for obtaining the new fog visibility prediction attribute in the step comprises the following steps: pretreatment: the real-time data method comprises the following steps of sequentially carrying out data cleaning, station data merging and similar fog visibility prediction attribute data minimization according to an actual fog visibility prediction attribute threshold value:
data addition characteristics: adding different time attributes to each fog visibility prediction data attribute, and calculating a new attribute by using the relationship among the attributes of the fog visibility prediction data;
the labeling method in the step S3 is as follows: determining the fog moment according to a specific threshold value; and marking the fog visibility value of the fog visibility prediction data (data after preprocessing) according to the obtained fog-existence moments and different early warning times, and obtaining fog visibility prediction training samples at different early warning times. (according to different early warning time, the specific visibility value in the foggy state and the fogless state is marked in the sample at the moment)
Actual data attributes relevant for fog visibility prediction include barometric pressure, temperature, and relative humidity.
The method for obtaining all characteristic attribute data related to the fog visibility prediction comprises the following steps:
the method for establishing the time window adds different time attributes to each fog visibility prediction data attribute, the method for establishing the time window obtains a new time by subtracting the time corresponding to the current data from the corresponding threshold range according to each set time window model threshold range, finds out the line number of the data corresponding to the data set according to the obtained new time, and adds some fog visibility prediction attributes (the original attributes are the original data before preprocessing) in the line data as the new attributes of the current time;
the time window model threshold range is set to one hour and two hours; the method for calculating the relationship among the attributes comprises the steps of subtraction calculation and ratio calculation (which can be carried out in sequence) among the attributes at the same moment in each piece of data; subtraction calculation (for calculating variation) between attributes at different times of each piece of data; the new attributes obtained from each piece of fog visibility prediction data comprise temperature dew point difference, temperature dew point ratio, span temperature change and the like; the attribute features of the pre-and post-pre-processed data collectively represent all meteorological elements that may influence the prediction of fog visibility.
In the step S3, the method for establishing different early warning time fog prediction training samples includes:
firstly, determining the fog-existing time (according to expert knowledge in meteorology), wherein the determining method is that the requirement of a certain time period is met, a threshold value (a value determined by the expert in meteorology) is set for the fog-existing and fog-free characteristics, and if the data of the time period meets the threshold value requirement, the initial time of the time period is judged to be the fog-existing time;
respectively establishing fog visibility prediction training samples at different moments at the moment of judging whether fog exists and the early warning moment corresponding to the moment of judging whether fog exists, subtracting the range of the early warning time threshold from the moment corresponding to the fog data of each station to obtain a new moment for each early warning time threshold, marking the data of each station corresponding to the obtained new moment, and forming a fog prediction training sample data set of appointed early warning time by all characteristic data of the marking and the corresponding moment; the label of the fog prediction training sample represents the fog visibility value condition of the attribute data at the moment in the future early warning time, and the data set is used for the next XGboost method and model training; the prediction of different early warning time thresholds considered by the patent comprises the prediction of half an hour, 1 hour, 2 hours and 3 hours in the future of the fog, and fog visibility prediction training samples with different early warning time thresholds are respectively established;
description of the related Art
1. Predicted (classified) results: 0/1 fog free/foggy (negative/positive sample)
(1) Positive sample (hazy) definition: visibility <1000 m and relative humidity > 80%.
The fogging state is a case where a time point at which fogging occurs for the first time between 17:00 (previous day) and 10:00 (subsequent day) is regarded as a data sample, and a subsequent fogging duration is not regarded as a data sample. For example, 17:00 is fogged for the first time, 17:10,17:20, 17:30 … is fogged for the last time, then data samples are created based on the data at time 17:00, 17:10,17:20, 17:30, etc. are not created for it.
(2) Negative sample (no fog) definition: if the visibility is greater than 1000 meters for three consecutive days, a negative sample is considered. Select 20 on the middle day: 00, 23: 00, 02: 00,5: the data at time 00 is taken as a negative sample. For example, No. 24 to No. 26 have no fog, and No. 25 and No. 2 are selected.
(3) Long sea site data not used (weather bureau provides site name and corresponding number)
2. Inputting an attribute: instantaneous and partial variations (the main basis for prediction). For example, if the prediction at 19 o 'clock is taken as an example, the first-time prediction needs to refer to the attribute values of the first 1 h and the second 2h as prediction factors, and if the prediction at 19 o' clock is taken as an example, the first-time prediction needs to refer to the attribute values of 18 o 'clock 30 minutes and 17 o' clock 30 minutes, and other times are similar.
The following takes the issue of a big fog warning at 19 hours as an example:
(1) instantaneous amount (19, 18, 17): the method comprises the following steps: air pressure, sea level air pressure, 2m temperature, 2m dew point temperature, relative humidity, wind direction, wind speed, rainfall, actual water air pressure, saturated water vapor pressure, actual specific humidity, and saturated specific humidity. And meteorological elements calculated using the above elements in the following table.
(2) Relative change amount:
1 hour (19 hours and 18 hours) variance: (saturation specific humidity-actual specific humidity) variation, temperature T variation, dew point temperature TD variation, temperature dew point difference variation T-TD and relative humidity RHW variation.
2 hour (19 hours and 17 hours) variance: (saturation specific humidity-actual specific humidity) variation, temperature T variation, dew point temperature TD variation, temperature dew point difference variation T-TD and relative humidity RHW variation.
3. Forecasting the demand: it was predicted whether or not fogging occurred at 0.5 hour, 1 hour, 2 hours, and 3 hours.
4. Establishing a data sample: respectively establishing the time of 0.5 hour before the fog appearsData samples, 1 hour, 2 hours, 3 hours data samples. For example, whether fog appears at 20:00 is predicted (fog appears at 20 in the data), a data sample at 19:30 is a predicted 0.5 hour data sample, a data sample at 19:00 is a predicted 1 hour sample, a data sample at 18:00 is a predicted 2 hour sample, and a data sample at 17:00 is a predicted 3 hour sample. For the predicted 0.5 hour data sample (19:30), the data includes instantaneous data at the time of the portion 19:30, and the variance. The variance is 19:30 difference of the instantaneous data from the previous 1 hour (Dt-h), from the previous 2 hours (Dt-2 h), etc. E.g. calculating the amount of temperature change, T19:30–T18:30，T19:30–T17:30
5. If some attribute data is missing, interpolation padding can be performed by using data of adjacent 20 minutes.
6. Raw data, data samples were collected 10 minutes.
Data processing and training sample library establishment steps:
1. merging same site data
Data items in the original data having the same site (attribute number) are merged into a data subset.
2. Data quality control
Invalid data in the data set is deleted according to the following conditions.
Conditions are as follows:
(1) attribute value of 9999
(2) Judging according to the visibility. When "relative humidity" is less than 80%, and min (1 minute visibility, 10 minute visibility) <1000, the visibility data is considered erroneous.
3. Secondary physical quantities (refer to the above-described amount of change and relative amount of change) are calculated.
4. The positive and negative samples (0 and 1) are labeled. Reference is made to positive and negative sample definitions.
5. A set of positive samples is established.
(1) Determining a predicted time (first foggy);
(2) establishing a sample set 0.5 hour before the prediction time;
(3) establishing a sample set 1 hour before a prediction moment;
(4) establishing a sample set 2 hours before a prediction moment;
(5) establishing a sample set 3 hours before the prediction time;
6. a set of negative examples is established.
In the step S4, the XGBoost method is a method for predicting visibility in fog in a future time period by training sample data, that is, the fog visibility prediction sample data at different early warning moments are respectively input to the XGBoost for training, a threshold segmentation point related to the division of each meteorological attribute feature value and a leaf node related to the visibility value are obtained by using each round of additive training fitting residual error and greedy algorithm, and each sub-tree is formed, each obtained sub-tree corresponds to a feature space, and the integrated learning method model based on trees is used for analyzing each meteorological attribute while predicting the visibility in fog.
The XGboost method is an integrated learning method, a plurality of decision trees are integrated to fit the residual error of each decision tree in each round, leaf nodes of each feature space corresponding to a specific fog visibility value are obtained and are divided into feature spaces, and the expression of the XGboost is shown as a formula (1); the XGboost is used for fitting data in each round of training through an addition model, an objective function of the XGboost adopts an expression mode of Taylor second-order expansion, so that the XGboost is more favorable for training fitting errors, and a Taylor expansion formula of the objective function is shown as the following formula (2); optimizing an objective function and dividing a feature space by adopting a greedy algorithm to obtain an optimal residual error of each round; the specific formula of the optimal subtree of each XGboost formed in each round is shown as (3);
(1) (2) and (3) respectively representing a general expression of the XGboost, an objective function based on Taylor second-order expansion and an optimal decision tree in each round, wherein k represents the number of the decision trees forming the XGboost integrated learning method, F represents the decision tree space forming the XGboost, X represents training sample data, T represents the number of leaf nodes in the XGboost, w represents the number of the leaf nodes in the XGboostjRepresenting the score of the leaf j node. Where λ is the regularization coefficient, γ is the coefficient controlling the number of leaf nodes, preventing overfitting, giFor the first derivative, h, of the new decision tree under XGboost additive trainingiFor the second derivative of the iterative decision tree under XGboost additive training, IjFor the sample where the jth leaf node in the kth leaf is located, fkRepresenting the kth sub-tree constituting the method of the integration tree.
The process of analyzing and predicting the fog visibility and the meteorological characteristic space by the XGboost method is to establish good fog visibility prediction training samples aiming at different early warning times, and obtain a mapping model of the predicted fog visibility and the meteorological characteristic space at each early warning time by the XGboost method;
preprocessing each piece of real-time data of each station, and inputting the preprocessed real-time data into trained models for predicting fog visibility at different early warning times to obtain predicted fog visibility values of each piece of real-time data at different early warning times;
further comprising the steps of:
and (4) acquiring data for testing, inputting the testing data into the model by using the XGboost method model obtained in the steps S1-S4, and comparing the obtained fog visibility prediction result with the label to obtain the fitting degree of the fog visibility prediction.
For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
1. First, raw data and real-time data need to be acquired
2. Preprocessing to obtain new attributes of original data and real-time data (the attributes of the two are the same)
3. Marking the preprocessed original data, and respectively establishing different short-time near visibility prediction samples according to different short-time near early warning times (the preprocessed real-time data does not need to be subjected to the step
4. Different short-time adjacent visibility prediction training models are built for different training samples (a chain related to the original data in the third step), the training is to find the corresponding relation, finally predict the preprocessed real-time data (because the data is input), and input the data into the prediction models built in the different short-time adjacent prediction models (the 4 th step (the last step)), so that the visibility conditions in different short-time adjacent time at the moment in the future are obtained.
The short-time near fog visibility prediction model not only can analyze main attribute characteristics influencing fog visibility prediction according to different early warning time, but also can train according to data corresponding to the analyzed fog visibility prediction characteristics, so that the trained model can learn the change of regularity, and the model has strong generalization capability, and the short-time fog visibility condition can be accurately predicted in extremely short time. Based on the above description, the invention provides a combined prediction method of the XGboost method; the XGboost method aims at obtaining a feature space corresponding to each datum through analysis of a training sample; the improved linear regression method improves the generalization capability of the model by adding an adaptive moment estimation and regularization method; the two methods are effectively combined, so that the fog visibility can be accurately predicted in a short time, and main attributes influencing the predicted fog visibility are analyzed.
In order to obtain a data source related to the fog visibility prediction method, the data of the embodiment is derived from a national integrated weather information sharing platform (hereinafter, referred to as CIMISS) system, which is a weather data unified service interface with unified standard, rich resources, and fast and efficient invocation. And the method is convenient for people to acquire relevant fog visibility prediction data.
In order to obtain the original data of the predicted fog visibility, the attribute data of the predicted fog visibility sample in this embodiment are all from a ground radar observation data interface of the CIMISS, the data year is 10 minutes of data of all stations in nearly five years, and the downloaded original data of the predicted fog visibility include air pressure, temperature, relative humidity and the like.
In order to process original data of fog visibility prediction, the data source of the embodiment needs to be cleaned according to a specific fog visibility prediction attribute threshold value, and part of the data with the attribute not meeting the meteorological range is cleaned; and merging the data according to each site, wherein the downloaded attributes comprise some similar attributes, such as 1-minute fog visibility and 10-minute fog visibility, the two similar fog visibility prediction attributes are minimized, and the two attributes are merged. (ii) a
In order to add different time attributes as new attributes to each fog visibility prediction data attribute in each website, a time window model method needs to be adopted to obtain the different time attributes for the embodiment, according to each set time threshold range, the method performs difference between the time corresponding to the current data and the corresponding threshold range to obtain new time, according to the obtained new time, the line number of the data corresponding to the data set is found out, and certain fog visibility prediction attributes in the line are taken as the new attributes of the current time and added to the current time, and the threshold range of the fog visibility prediction time window model adopted by the method is 1 hour and 2 hours.
In order to obtain possible new fog visibility prediction attributes, different methods need to be performed on the fog visibility prediction data of the embodiment, wherein the calculation of the relation among the attributes of the fog visibility prediction data comprises subtraction calculation of the attributes of each piece of data at the same moment, for example, the data at the same moment are subtracted to obtain new change attributes in consideration of the difference between the attribute visible temperature and the dew point temperature at the same moment; calculating the ratio of the attributes at the same moment, and acquiring a related change coefficient, for example, dividing the change of the temperature and the dew point temperature by the total temperature difference to obtain a new relative quantity attribute; meanwhile, the attribute values of the same data at different moments are subtracted to obtain a time variation serving as a new attribute, for example, the temperature value at the moment is subtracted from the temperature attribute value before one hour in the data attribute to obtain a temperature variation after one hour; the obtained new attributes comprise a plurality of characteristics such as temperature dew point difference, temperature dew point ratio, span temperature change and the like; the original attribute features and the new attribute features jointly represent all meteorological elements which can influence the fog visibility prediction, and meanwhile, the attribute features can comprehensively express fog in a more diversified manner;
in order to determine the time of fog existence at each station, firstly, defining fog-free characteristics for the fog visibility prediction data of the embodiment, namely meeting the requirement of a certain time period, respectively setting threshold values of a certain interval for the fog-free characteristics and the fog-free characteristics, and if the time period data meets the requirement of the threshold values of the fog-free characteristics or the fog-free characteristics in the time period data, judging that the initial time of the time period is the fog-free time or the fog-free time;
in order to establish sample data sets at different moments, the method of the embodiment subtracts a moment corresponding to the existence of fog data of each station from a threshold range of the early warning time to obtain a new moment, labels the data of each station corresponding to the obtained new moment, and the labels and the corresponding data form a fog visibility prediction training sample data set at a specified early warning time; the label of the fog visibility prediction training sample represents the fog visibility value condition of the attribute data at the moment in the future early warning time, and the data set is used for the next training of a fog visibility prediction model; the different early warning time thresholds considered by the method comprise predictions of half an hour, 1 hour, 2 hours and 3 hours in the future of the fog, and fog visibility prediction training samples with different early warning time thresholds are respectively established;
in order to analyze the relation between fog visibility and meteorological feature space at different early warning times, the XGboost method is adopted in the embodiment, the method is characterized in that fog visibility prediction sample data at different early warning times are respectively input into the XGboost for training, threshold segmentation points related to the division of each meteorological attribute feature value and leaf nodes related to the visibility value are obtained by utilizing each round of additive training fitting residual error and a greedy algorithm, each sub-tree obtained by the method corresponds to one feature space, and the integrated learning method model based on the tree is used for analyzing each meteorological attribute while the fog visibility prediction is realized;
the XGboost method is an integrated learning method, a plurality of decision trees are integrated to fit the residual error of each decision tree in each round, leaf nodes of each feature space corresponding to a specific fog visibility value are obtained and are divided into feature spaces, and the expression of the XGboost is shown as (1);
the XGboost is used for fitting data in each round of training through an addition model, an objective function of the XGboost adopts an expression mode of Taylor second-order expansion, so that the XGboost is more favorable for training fitting errors, and a Taylor expansion formula of the objective function is shown as the following formula (2); the specific flow is shown in fig. 3; the specific structure is shown in FIG. 4;
optimizing an objective function and dividing a feature space by adopting a greedy algorithm to obtain an optimal residual error of each round; the specific formula of the optimal subtree of each XGboost formed in each round is shown as (3);
(1) and (2) and (3) respectively represent a general expression of the XGboost, an objective function based on Taylor second-order expansion and an optimal decision tree in each round, wherein k represents the number of the decision trees forming the XGboost integrated learning method, F represents the decision tree space forming the XGboost, X represents training sample data, T represents the number of leaf nodes in the XGboost, and w represents the fraction of the leaf nodes. Where λ is the regularization coefficient, γ is the coefficient controlling the number of leaf nodes, preventing overfitting, giFor the first derivative, h, of the new decision tree under XGboost additive trainingiFor the second derivative of the iterative decision tree under XGboost additive training, IjIs the sample of the jth leaf node in the kth leaf;
in order to obtain the mapping between the predicted fog visibility values at different early warning times and the meteorological feature space, the XGboost method is used for establishing good fog visibility prediction training samples aiming at different early warning times and respectively obtaining a mapping model between the predicted fog visibility values at each early warning time and the meteorological feature space by using the XGboost method;
in order to predict by using the trained XGboost method model, each piece of real-time data of each station is preprocessed and then input into the XGboost method trained at the corresponding short-time adjacent moment, the real-time data of each piece of fog visibility is predicted, and the output value is the predicted value of the fog visibility of each station at different early warning moments; the specific real-time data prediction process is shown in fig. 1.
The specific prediction method and steps of the combination method of this embodiment are shown in fig. 2. The short-time adjacent fog classified prediction establishing method specifically comprises the following steps:
s1, acquiring and processing an original data set related to fog visibility prediction; the original data come from data sets of different years, different times and different sites; the processing method comprises the steps of data cleaning, and merging data according to sites;
s2, marking and processing the original data set to obtain all characteristic attribute data related to fog visibility prediction, and forming fog visibility prediction training data at different early warning times; the method for obtaining the new fog visibility prediction attribute comprises the following steps: adding different time attributes to each fog visibility prediction data attribute, and calculating a new attribute by using the relationship among the attributes of the fog visibility prediction data; determining the fog moment according to a specific threshold value; marking the fog visibility value of the fog visibility prediction data according to the obtained moment and different early warning times, and obtaining fog visibility prediction training samples of different early warning times;
the method comprises the steps of S3, XGboost, training by using fog visibility prediction samples at different moments to obtain the relation between each predicted fog visibility value and meteorological attribute feature space at different early warning time, and establishing fog visibility prediction models at different early warning time;
and S4, during data prediction, inputting each piece of preprocessed real-time data into trained models for predicting fog visibility at different early warning times respectively to obtain the predicted fog visibility values of each piece of real-time data at different early warning times.
(1) And (2) and (3) respectively represent a general expression of the XGboost, an objective function based on Taylor second-order expansion and an optimal decision tree in each round, wherein k represents the number of the decision trees forming the XGboost integrated learning method, F represents the decision tree space forming the XGboost, X represents training sample data, T represents the number of leaf nodes in the XGboost, and w represents the fraction of the leaf nodes. Where λ is the regularization coefficient, γ is the coefficient controlling the number of leaf nodes, preventing overfitting, giFor the first derivative, h, of the new decision tree under XGboost additive trainingiFor the second derivative of the iterative decision tree under XGboost additive training, IjIs the sample of the jth leaf node in the kth leaf;
and acquiring test data, processing the test data by utilizing the steps S1-S4 to obtain a predicted test result, comparing the test result with an actual result to obtain the degree of fitting of the fog visibility prediction, and judging the prediction effect of the model.
The method for predicting the visibility of the short-term neighboring fog corresponds to the method, and the method can be referred to specifically.
Claims (10)
1. A short-time adjacent fog visibility prediction method is characterized by comprising the following steps: comprises the following steps:
the method comprises the steps of firstly, acquiring real-time data related to fog visibility prediction;
secondly, preprocessing the real-time data in the first step;
and thirdly, inputting the data preprocessed in the second step into a model for predicting fog visibility at different early warning times, which is trained by an XGboost method, to obtain a predicted fog visibility value of each piece of real-time data at different prediction times.
2. The short-term adjacent fog visibility prediction method as claimed in claim 1, wherein: in a first step, real-time data relating to the prediction of fog visibility includes actually collected data of air pressure, temperature and relative humidity for the predicted time period.
3. The short-term adjacent fog visibility prediction method as claimed in claim 1, wherein: and the second step of preprocessing the real-time data comprises the steps of sequentially carrying out data cleaning, data merging on all stations and minimizing the similar fog visibility prediction attribute data according to the actual fog visibility prediction attribute threshold.
4. The short-term adjacent fog visibility prediction method as claimed in claim 1, wherein: the model construction method for predicting fog visibility in the third step is as follows:
s1, acquiring and processing an original data set related to fog visibility prediction;
s2, preprocessing the original data in the step S1 to obtain all characteristic attribute data related to fog visibility prediction;
s3, marking to obtain all characteristic attribute data related to fog visibility prediction, and establishing fog visibility prediction sample training data at different early warning times according to the marks corresponding to the prediction range;
and S4, training the fog visibility prediction samples at different moments of the training data in the step S3 by using an XGboost method, obtaining the relation between each predicted fog visibility value and a meteorological attribute feature space at different early warning times, and establishing fog visibility prediction models at different early warning times according to the relation.
5. The short-term adjacent fog visibility prediction method as claimed in claim 4, wherein:
s2, the method for obtaining the new fog visibility prediction attribute in the step comprises the following steps: pretreatment: the real-time data method comprises the following steps of sequentially carrying out data cleaning, station data merging and similar fog visibility prediction attribute data minimization according to an actual fog visibility prediction attribute threshold value:
data addition characteristics: adding different time attributes to each fog visibility prediction data attribute, and calculating a new attribute by using the relationship among the attributes of the fog visibility prediction data;
the labeling method in the step S3 is as follows: determining the fog moment according to a specific threshold value; and marking the fog visibility value of the fog visibility prediction data according to the obtained fog-existence moments and different early warning times, and obtaining fog visibility prediction training samples at different early warning times.
6. The short-term adjacent fog visibility prediction method as claimed in claim 3 or 5, wherein:
actual data attributes relevant for fog visibility prediction include barometric pressure, temperature, and relative humidity.
7. The short-term adjacent fog visibility prediction method as claimed in claim 5, wherein: the method for obtaining all characteristic attribute data related to the fog visibility prediction comprises the following steps:
the method for establishing the time window adds different time attributes to each fog visibility prediction data attribute, the method for establishing the time window obtains a new time by subtracting the time corresponding to the current data from the corresponding threshold range according to each set time window model threshold range, finds out the line number of the data corresponding to the data set according to the obtained new time, and adds some fog visibility prediction attributes in the line data into the current time as the new attributes of the current time;
the time window model threshold range is set to one hour and two hours; the method for calculating the relationship among the attributes comprises the steps of subtracting the attributes at the same moment in each piece of data and calculating the ratio; the new attributes obtained from each of the fog visibility prediction data include temperature dew point difference, temperature dew point ratio, and span temperature variation.
8. The short-term adjacent fog visibility prediction method as claimed in claim 4, wherein: in the step S3, the method for establishing different early warning time fog prediction training samples includes:
firstly, determining the time with or without fog, wherein the determining method is to meet the requirement of a certain time period, set a threshold value for the characteristics of fog and fog, and judge the initial time of the time period as the time with or without fog if the data of the time period meet the requirement of the threshold value;
respectively establishing fog visibility prediction training samples at different moments at the moment of judging whether fog exists and the early warning moment corresponding to the moment of judging whether fog exists, subtracting the range of the early warning time threshold from the moment corresponding to the fog data of each station to obtain a new moment for each early warning time threshold, marking the data of each station corresponding to the obtained new moment, and forming a fog prediction training sample data set of appointed early warning time by all characteristic data of the marking and the corresponding moment; the label of the fog prediction training sample represents the fog visibility value condition of the attribute data at the moment in the future early warning time, and the data set is used for the next XGboost method and model training.
9. The short-term adjacent fog visibility prediction method as claimed in claim 4, wherein: in the step S4, the XGBoost method is a method for predicting visibility in fog in a future time period by training sample data, that is, the fog visibility prediction sample data at different early warning moments are respectively input to the XGBoost for training, a threshold segmentation point related to the division of each meteorological attribute feature value and a leaf node related to the visibility value are obtained by using each round of additive training fitting residual error and greedy algorithm, and each sub-tree is formed, each obtained sub-tree corresponds to a feature space, and the integrated learning method model based on trees is used for analyzing each meteorological attribute while predicting the visibility in fog.
10. The method of predicting visibility of short-term neighboring fog according to claim 9, wherein:
the expression of the XGboost is shown as formula (1); the XGboost is used for fitting data in each round of training through an addition model, an objective function of the XGboost adopts an expression mode of Taylor second-order expansion, and the Taylor expansion of the objective function is shown as the following formula (2); optimizing an objective function and dividing a feature space by adopting a greedy algorithm to obtain an optimal residual error of each round; the specific formula of the optimal subtree of each XGboost formed in each round is shown as (3);
(1) (2) and (3) respectively representing a general expression of the XGboost, an objective function based on Taylor second-order expansion and an optimal decision tree in each round, wherein k represents the number of the decision trees forming the XGboost integrated learning method, F represents the decision tree space forming the XGboost, X represents training sample data, T represents the number of leaf nodes in the XGboost, w represents the number of the leaf nodes in the XGboostjA score representing a leaf j node; where λ is the regularization coefficient, γ is the coefficient controlling the number of leaf nodes, preventing overfitting, giFor the first derivative, h, of the new decision tree under XGboost additive trainingiFor the second derivative of the iterative decision tree under XGboost additive training, IjFor the sample where the jth leaf node in the kth leaf is located, fkRepresenting the kth sub-tree constituting the method of the integration tree.
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910731635 | 2019-08-08 | ||
| CN2019107316351 | 2019-08-08 | ||
| CN2019107310251 | 2019-08-08 | ||
| CN201910731025 | 2019-08-08 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111784023A true CN111784023A (en) | 2020-10-16 |
Family
ID=72753512
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010409472.8A Active CN111784022B (en) | 2019-08-08 | 2020-05-14 | Short-time adjacent large fog prediction method based on combination of Wrapper method and SVM method |
| CN202010409474.7A Pending CN111784023A (en) | 2019-08-08 | 2020-05-14 | Short-time adjacent fog visibility prediction method |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010409472.8A Active CN111784022B (en) | 2019-08-08 | 2020-05-14 | Short-time adjacent large fog prediction method based on combination of Wrapper method and SVM method |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN111784022B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801202A (en) * | 2021-02-10 | 2021-05-14 | 延锋汽车饰件系统有限公司 | Vehicle window fogging prediction method, system, electronic device, and storage medium |
| CN115169592A (en) * | 2022-06-30 | 2022-10-11 | 深圳信息职业技术学院 | Visibility prediction method, device, storage medium and computer equipment |
| CN115662149A (en) * | 2022-12-13 | 2023-01-31 | 合肥远捷信息技术有限公司 | Information system based on highway group fog early warning management and control |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114280696B (en) * | 2021-12-23 | 2024-09-17 | 无锡九方科技有限公司 | Sea fog grade intelligent forecasting method and system |
| CN114999180B (en) * | 2022-05-12 | 2023-12-26 | 广东省韶关市气象局 | Expressway bad weather traffic early warning system and method based on Internet of things |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108375808A (en) * | 2018-03-12 | 2018-08-07 | 南京恩瑞特实业有限公司 | Dense fog forecasting procedures of the NRIET based on machine learning |
| US20180300639A1 (en) * | 2017-04-13 | 2018-10-18 | Oracle International Corporation | Novel autonomous artificially intelligent system to predict pipe leaks |
| CN109190670A (en) * | 2018-08-02 | 2019-01-11 | 大连理工大学 | A kind of charging pile failure prediction method based on expansible boosted tree |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100914518B1 (en) * | 2008-02-19 | 2009-09-02 | 연세대학교 산학협력단 | System for generating genre classification taxonomy, and method therefor, and the recording media storing the program performing the said method |
| CN102629305B (en) * | 2012-03-06 | 2015-02-25 | 上海大学 | Feature selection method facing to SNP (Single Nucleotide Polymorphism) data |
| CN104657574B (en) * | 2014-06-13 | 2017-10-31 | 苏州大学 | The method for building up and device of a kind of medical diagnosismode |
| CN104504475A (en) * | 2014-12-24 | 2015-04-08 | 上海大学 | AR*-SVM (support vector machine) hybrid modeling based haze time series prediction method |
| CN107729716B (en) * | 2017-11-27 | 2020-10-27 | 西安建筑科技大学 | Coal mine water inrush prediction method based on long-time and short-time memory neural network |
-
2020
- 2020-05-14 CN CN202010409472.8A patent/CN111784022B/en active Active
- 2020-05-14 CN CN202010409474.7A patent/CN111784023A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180300639A1 (en) * | 2017-04-13 | 2018-10-18 | Oracle International Corporation | Novel autonomous artificially intelligent system to predict pipe leaks |
| CN108375808A (en) * | 2018-03-12 | 2018-08-07 | 南京恩瑞特实业有限公司 | Dense fog forecasting procedures of the NRIET based on machine learning |
| CN109190670A (en) * | 2018-08-02 | 2019-01-11 | 大连理工大学 | A kind of charging pile failure prediction method based on expansible boosted tree |
Non-Patent Citations (1)
| Title |
|---|
| 王智 等: ""基于提升树的PM2.5浓度预测模型"", 《软件》, vol. 39, no. 10, pages 156 - 163 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801202A (en) * | 2021-02-10 | 2021-05-14 | 延锋汽车饰件系统有限公司 | Vehicle window fogging prediction method, system, electronic device, and storage medium |
| CN112801202B (en) * | 2021-02-10 | 2024-03-05 | 延锋汽车饰件系统有限公司 | Vehicle window fogging prediction method, system, electronic equipment and storage medium |
| CN115169592A (en) * | 2022-06-30 | 2022-10-11 | 深圳信息职业技术学院 | Visibility prediction method, device, storage medium and computer equipment |
| CN115662149A (en) * | 2022-12-13 | 2023-01-31 | 合肥远捷信息技术有限公司 | Information system based on highway group fog early warning management and control |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111784022A (en) | 2020-10-16 |
| CN111784022B (en) | 2024-01-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111784023A (en) | Short-time adjacent fog visibility prediction method | |
| CN113919448B (en) | Method for analyzing influence factors of carbon dioxide concentration prediction at any time-space position | |
| CN110555561B (en) | Medium-and-long-term runoff ensemble forecasting method | |
| Martel et al. | Role of natural climate variability in the detection of anthropogenic climate change signal for mean and extreme precipitation at local and regional scales | |
| Wang et al. | An improved ARIMA model for precipitation simulations | |
| CN111665575B (en) | Medium-and-long-term rainfall grading coupling forecasting method and system based on statistical power | |
| CN102880755B (en) | Method and system for quantitatively forecasting extreme rainfall | |
| CN111191193A (en) | Long-term soil temperature and humidity high-precision prediction method based on autoregressive moving average model | |
| CN106600037B (en) | Multi-parameter auxiliary load prediction method based on principal component analysis | |
| CN113496314B (en) | Method for predicting road traffic flow by neural network model | |
| CN113435630B (en) | Basin hydrological forecasting method and system with self-adaptive runoff yield mode | |
| CN112116149A (en) | Multi-station medium and long term runoff rolling probability prediction method considering forecast uncertainty associated evolution characteristics | |
| CN113239971B (en) | Wind field-oriented fusion system for proximity forecast and short-term forecast | |
| CN111861206A (en) | Industrial industry prosperity index obtaining method based on enterprise electric power big data | |
| CN115526298A (en) | High-robustness comprehensive prediction method for concentration of atmospheric pollutants | |
| CN113836808A (en) | PM2.5 deep learning prediction method based on heavy pollution feature constraint | |
| CN114564487B (en) | Meteorological raster data updating method combining forecast prediction | |
| CN113821895B (en) | Method and device for constructing power transmission line icing thickness prediction model and storage medium | |
| Mangaraj et al. | A Markov chain analysis of daily rainfall occurrence at western Orissa of India | |
| CN111967677A (en) | Prediction method and device for unconventional resource dessert distribution | |
| CN114004426B (en) | Dynamic adjustment method of short-time rainstorm forecast release model | |
| CN116184533A (en) | Novel weight coefficient-based hour-by-hour time lag set precipitation prediction correction method | |
| CN114723166A (en) | Surface temperature reconstruction method based on XGboost model | |
| CN112380778A (en) | Weather drought forecasting method based on sea temperature | |
| Xiang et al. | GA-ARIMA model-based analysis of arrival time at bus stop |
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 |