CN117313555B - GRU humidity prediction method based on self-adaptive OATS-AJSA - Google Patents
GRU humidity prediction method based on self-adaptive OATS-AJSA Download PDFInfo
- Publication number
- CN117313555B CN117313555B CN202311599596.7A CN202311599596A CN117313555B CN 117313555 B CN117313555 B CN 117313555B CN 202311599596 A CN202311599596 A CN 202311599596A CN 117313555 B CN117313555 B CN 117313555B
- Authority
- CN
- China
- Prior art keywords
- jellyfish
- data
- representing
- data set
- gru
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 69
- 230000002159 abnormal effect Effects 0.000 claims abstract description 29
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 22
- 238000003860 storage Methods 0.000 claims abstract description 10
- 238000005457 optimization Methods 0.000 claims abstract description 7
- 241000242583 Scyphozoa Species 0.000 claims description 117
- 238000009826 distribution Methods 0.000 claims description 27
- 230000006870 function Effects 0.000 claims description 26
- 230000003044 adaptive effect Effects 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000007246 mechanism Effects 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000002474 experimental method Methods 0.000 claims description 4
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000005520 cutting process Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 230000009467 reduction Effects 0.000 abstract description 3
- 238000012545 processing Methods 0.000 description 8
- 238000013459 approach Methods 0.000 description 4
- 230000003247 decreasing effect Effects 0.000 description 4
- 238000010845 search algorithm Methods 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 235000013305 food Nutrition 0.000 description 2
- 238000012804 iterative process Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000010998 test method Methods 0.000 description 2
- 241000209761 Avena Species 0.000 description 1
- 235000007319 Avena orientalis Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000003973 irrigation Methods 0.000 description 1
- 230000002262 irrigation Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000004899 motility Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Chemical compound O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06N3/08—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)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Environmental & Geological Engineering (AREA)
- Atmospheric Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Medical Informatics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Environmental Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a distributed storage self-adaptive OATS-AJSA-based improved GRU humidity prediction method, which specifically comprises the following steps: acquiring historical multi-element meteorological data and storing the data in a MinIO database in a distributed manner; identifying abnormal values in the multi-element meteorological data set by using an isolated forest method, and replacing the abnormal values; performing dimension reduction on the replaced data set by adopting a t-SNE method to obtain a dimension reduced data set, and normalizing the dimension reduced data set to obtain a normalized data set; establishing an OATS-AJSA-GRU humidity prediction model, and optimizing the super parameters in the GRU by using an OATS-AJSA optimization algorithm to obtain optimized super parameters; and placing the optimized super-parameters and the normalized data set into the GRU for prediction to obtain a humidity prediction result. The method realizes higher accuracy, stability and efficiency in the aspect of humidity prediction, and simultaneously has automatic super-parameter optimization and wide practical application potential.
Description
Technical Field
The invention relates to an improved GRU humidity prediction method based on self-adaptive OATS-AJSA, and belongs to the technical field of meteorological data prediction.
Background
The humidity prediction plays a very key role in weather disaster early warning, is an important research content in the fields of weather science and environmental science, and has important significance in understanding and predicting the change of the moisture content in the atmosphere. Humidity has profound effects on aspects such as weather, agriculture, energy, ecosystem, health, and the like. Humidity is one of the important factors in meteorology, in predicting weather and climate change, affecting cloud formation, precipitation generation and energy transfer in the atmosphere. In the agricultural field, humidity prediction can help farmers to rationally arrange irrigation and crop management to maximize yield. The energy sector needs accurate humidity prediction to optimize power production and distribution, as humidity changes can affect the water vapor content in the air, and thus the efficiency of wind power generation and solar power generation. Accurate humidity prediction can early warn in advance, and loss caused by disasters is greatly reduced. The conventional method for predicting humidity is low in prediction accuracy because the problem of time series is not considered. Meanwhile, with the progress of technology, the collection speed and the collection quantity of multi-element meteorological data are continuously increased, and challenges are brought to the traditional data storage and processing method.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the improved GRU humidity prediction method based on the adaptive OATS-AJSA is provided, and the accuracy of humidity prediction is further improved.
The invention adopts the following technical scheme for solving the technical problems:
the improved GRU humidity prediction method based on the adaptive OATS-AJSA comprises the following steps:
step 1, acquiring historical multi-element meteorological data and storing the data in a MinIO database in a distributed manner;
step 2, extracting multi-element meteorological data from a MinIO database as a prediction data set, identifying abnormal values in the prediction data set by using an isolated forest method, and replacing the abnormal values by using a linear interpolation method to obtain a replaced data set;
step 3, reducing the dimension of the replaced data set obtained in the step 2 by adopting a t distribution-random adjacent embedding method to obtain a dimension-reduced data set, and normalizing the dimension-reduced data set to obtain a normalized data set;
step 4, an OATS-AJSA-GRU humidity prediction model is established, and the ultra-parameters in the GRU are optimized by utilizing an OATS-AJSA optimization algorithm to obtain optimized ultra-parameters;
and 5, putting the optimized super parameters and the normalized data set obtained in the step 3 into GRU for prediction, and obtaining a humidity prediction result.
As a preferable scheme of the invention, the specific process of the step 1 is as follows:
1) Accessing the ECMWF by using the API and downloading the NC data file;
2) Analyzing the downloaded NC data file by using a xarray library of Python to obtain a CSV data file;
3) Configuring and connecting to a MinIO server, uploading the CSV data file obtained in the step 2 to a MinIO socket by using a MinIO SDK, and taking the CSV data file as a historical multi-element weather database, wherein the multi-element weather data comprises temperature, humidity, wind speed and pressure data;
4) Storing user information by using a MySQL database, creating an API endpoint by using a Python Web framework, and calling the historical multi-element weather data in the historical multi-element weather database by the user stored in the MySQL database.
As a preferred scheme of the present invention, in step 2, the identifying abnormal values in the predicted data set by using the isolated forest method specifically includes:
1) Setting the prediction data set as X, each data point X in X i Consists of d meteorological features, i.e. x i =(x i1 ,…,x id ),x i1 、x id Representing data point x i Is randomly selected from XThe data points form a subset X ', and the subset X' is taken as a root node of the isolated tree;
2) Randomly designating a dimension m, and m < d, randomly generating a cut point n in the subset X', as follows:
min(x ij ,j=m,x ij ∈X′)<n<max(x ij ,j=m,x ij ∈X′)
wherein x is ij Representing data point x i Is the jth meteorological feature of (2);
3) Creating a hyperplane according to the cutting point n, and dividing the sub-set X ' by utilizing the hyperplane, namely placing data points smaller than n in the sub-set X ' into a left sub-node, and placing data points larger than or equal to n in the sub-set X ' into a right sub-node;
4) Repeating the steps 2) and 3) until all leaf nodes of the isolated tree contain only one data point or the height of the isolated tree reaches a preset upper height limit, namely generating an isolated tree;
5) Repeating the steps 1) -4) to obtain an isolated forest;
6) For each data point X in X i Traversing each isolated tree in the isolated forest to obtain a data point x i The heights in each isolated tree are averaged to obtain an average heightThe average height of all data points is normalized;
7) Using the normalized average height g (x i ) Calculating outlier scoreTaking data points corresponding to abnormal value scores which are larger than or equal to a preset threshold value as abnormal values; the outlier score is calculated as follows:
wherein,ξ=0.5772156649;E(g(x i ) For data point x) i Calculating average heights in all the isolated trees, and taking an index after normalization; />Representing a parametric function->Represents the Euler-Mascheroni constant plus +.>Natural logarithm of (a).
As a preferred scheme of the present invention, in step 2, the replacing of the outlier by using the linear interpolation method specifically includes:
1) Finding adjacent data points of the abnormal value, namely two normal values before and after the abnormal value;
2) A straight line is constructed by using two normal values before and after the abnormal value, namely:
x estimated =(1-α)*x before +α*x after
wherein x is estimated Alternative value, x, representing outlier before 、x after Respectively before the abnormal value,The subsequent normal value, alpha represents an abnormal value of x before And x after The position ratio of the two components is more than or equal to 0 and less than or equal to 1;
3) And finding out the corresponding ordinate value along the constructed straight line according to the position of the outlier on the abscissa, namely, the replacement value of the outlier.
As a preferred scheme of the present invention, in step 3, the dimension of the replaced data set obtained in step 2 is reduced by adopting a t distribution-random adjacent embedding method, so as to obtain a dimension-reduced data set, which specifically includes the following steps:
1) Setting the high-dimensional data set as { x } which is the data set after replacement 1 ,x 2 ,…,x n The dimensionality-reduced data set is { y } 1 ,y 2 ,…,y n },x 1 ,x 2 ,…,x n Representing high-dimensional data points in a high-dimensional dataset, y 1 ,y 2 ,…,y n Representing low-dimensional data points in the dimensionality reduced data set, and calculating high-dimensional similarity probability P by using Gaussian distribution ij The method comprises the steps of carrying out a first treatment on the surface of the Wherein, the Gaussian distribution formula is as follows:
wherein p is j|i Representing a high-dimensional data point x i Generating x j Conditional probability, sigma of i 2 Variance parameter, ||x, representing distance calculation i -x j II represents x i And x j Euclidean distance, x k 、x l Representing high-dimensional data points in a high-dimensional dataset;
2) Calculating low-dimensional similarity probability Q using t distribution ij Wherein the t distribution formula is as follows:
wherein q j|i Representing low-dimensional data points y i Generating y j Conditional probability of y k 、y l Representing reduced-dimension data setsLow dimension data points;
3) Using the difference between the KL-divergence Heng Lianggao-dimensional probability distribution and the low-dimensional probability distribution, the formula is as follows:
wherein KL (p||q) represents KL divergence;
4) Updating the low-dimensional representation using a gradient descent method, the updating rule being as follows:
where beta represents the learning rate, theta represents the global parameter,is a parameter used to scale the gradient when updating the low-dimensional representation;
5) Repeating 2) -4) until a convergence condition is reached, i.e. a preset number of iterations is reached or the difference between the KL-divergence of the current iteration and the KL-divergence of the last iteration is smaller than a preset threshold.
As a preferred scheme of the present invention, the specific process of the step 4 is as follows:
step 41, initializing parameters of an artificial jellyfish searching algorithm by using an orthogonal table testing method, wherein the parameters of the artificial jellyfish searching algorithm comprise population quantity, searching radius and a population initializing strategy;
step 42, taking the super parameters of the GRU as jellyfish, calculating the fitness value of the jellyfish by using a cross entropy loss function in each iteration, and taking the jellyfish position with the highest fitness value as the optimal individual position of the jellyfish population in the current iteration; the fitness value is calculated as follows:
wherein L (i, L, d, h) represents a cross entropy loss function, i is the iteration number, L is the learning rate, d is the dropout proportion, h is the number of hidden layer neurons, N is the number of samples, y i Representing the actual label of the i-th sample,representing the prediction probability of the model to the ith sample, and F represents the fitness;
step 43, a time control mechanism is introduced to determine that the current iteration jellyfish enters the ocean current stage or jellyfish group stage, namely:
wherein C (t) represents a control function of the time control mechanism, t represents the current iteration number, rand (0, 1) represents a random number between 0 and 1, t max Representing the initially set maximum iteration number;
the maximum iteration number is dynamically adjusted by using an adaptive adjustment method, and the formula is as follows:
wherein t is rem Representing the number of remaining iterations, t min Representing a minimum number of iterations;
step 44, judging according to the time control function, if C (t) is more than or equal to 0.5, enabling jellyfish to enter a ocean current stage, and updating the individual position according to the following formula:
X i (t+1)=X i (t)+rand(0,1)×(X * -σ×rand(0,1))×μ
wherein X is i (t+1) represents the position after the i-th jellyfish is updated, X i (t) represents the current iteration position of the ith jellyfish, X * Representing the optimal individual position in the current iteration jellyfish population, and sigma is a distribution systemNumber, σ=3, μ represents the average position of all jellyfish in the population;
if C (t) <0.5, the jellyfish enters a jellyfish group stage, whether the jellyfish actively moves or passively moves in the jellyfish group stage is further judged according to whether rand (0, 1) is larger than or equal to 0.5, if rand (0, 1) is larger than or equal to 0.5, the jellyfish is passively moved in the jellyfish group stage, and the individual position is updated according to the following formula:
X i (t+1)=X i (t)+α×rand(0,1)×(U b -L b )
where α is the motion coefficient, α=0.1, u b 、L b Respectively representing an upper bound and a lower bound of jellyfish population search space;
if rand (0, 1) <0.5, indicating that jellyfish is actively moving in the jellyfish group phase, the individual position is updated according to the following formula:
wherein,indicates the direction of jellyfish movement, +.>Represents the distance that jellyfish moves along the movement direction, X j (t) represents the current iteration position of the jth jellyfish randomly selected from the jellyfish population, f (X) i )、f(X j ) Respectively representing the fitness values of the ith jellyfish and the jth jellyfish at the current iteration position;
step 45, correcting the jellyfish position according to the clip function:
X new =clip(X,X min ,X max )
wherein X is new Represents the corrected jellyfish position, X min 、X max Respectively representing minimum and maximum allowable values of the position;
and 46, judging the magnitudes of the current iteration times T and the threshold constant T, if the current iteration times T are smaller than the threshold constant T, continuing iteration, otherwise, outputting the optimal value of the super parameter.
As a preferred embodiment of the present invention, the specific process of step 41 is as follows:
411, determining parameter ranges of three parameters, namely population quantity, search radius and population initialization strategy, and selecting different combinations of the three parameters, namely population quantity, search radius and population initialization strategy, according to the parameter ranges;
412, performing an ASJA experiment on each combination of the L16 orthogonal table, and selecting the corresponding combination when the experimental result is optimal as the optimal combination of three parameters, namely, the optimal parameter of the artificial jellyfish searching algorithm, such as population number, searching radius and population initializing strategy.
A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, which when executed implements the steps of the adaptive oat-AJSA based improved GRU humidity prediction method as described above.
A computer readable storage medium storing a computer program which when executed by a processor implements the steps of an adaptive oat-AJSA based improved GRU humidity prediction method as described above.
Compared with the prior art, the technical scheme provided by the invention has the following technical effects:
1. according to the invention, the GRU model is optimized by using the distributed storage and self-adaptive OATS-AJSA algorithm, so that the storage problem of massive meteorological data can be solved, and the humidity can be predicted more accurately, thereby improving the accuracy of meteorological prediction.
2. Compared with the method of adopting random initialization for the initialization of the optimization algorithm, the method of the invention uses the orthogonal table test method to initialize the parameters of the artificial jellyfish search algorithm, which can improve the efficiency, accuracy and stability of the algorithm, and is helpful to find the optimal parameter combination faster instead of relying on random attempt.
3. The invention adopts distributed storage and efficient data processing flow, so that the data acquisition and processing are more efficient. In addition, by means of self-adaptive adjustment of the maximum iteration times and the like, the parameter searching process is optimized, and the computing efficiency is improved.
4. According to the invention, an artificial jellyfish search algorithm is introduced to automatically find the optimal GRU model super-parameters, so that manual adjustment is not needed, and the time and labor cost are saved.
5. The method realizes higher accuracy, stability and efficiency in the aspect of humidity prediction, and simultaneously has automatic super-parameter optimization and wide practical application potential, thereby having remarkable technical advantages in the field of weather prediction.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a flow chart of the present invention for constructing an orphan forest;
FIG. 3 is a flow chart of the data dimension t-reduction distribution-random proximity embedding method of the present invention;
FIG. 4 is a flow chart of the OATS-AJSA-GRU predictive moisture model of the present invention.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below by referring to the drawings are exemplary only for explaining the present invention and are not to be construed as limiting the present invention.
The improved GRU humidity prediction method based on the adaptive OATS-AJSA comprises the following specific steps:
as shown in fig. 1, the multi-element weather data is downloaded and stored in a database in a distributed manner, and a user can call the multi-element weather data in the database, which comprises the following specific steps:
step 1: acquiring meteorological data from ECMWF, using API to access data service and downloading data file and storing in database;
step 2: analyzing the downloaded NC data file by using a xarray library of Python, and extracting the required meteorological variables and geographic coordinates;
step 3: configuring and connecting to a MinIO server, and uploading the obtained CSV file to a MinIO socket by using a MinIO SDK to serve as a historical multi-element meteorological database;
step 4: the MySQL database is used to store user information data for use in authenticating a user to log into the system. An API endpoint is created using a Python's Web framework (e.g., a flash framework) for a user to request weather data within a specified date range, thereby creating a callable interface to call historical multi-element weather data from a Minio database.
The invention uses historical multi-element meteorological data in a Minio database as a data source of a method for improving GRU prediction humidity by self-adapting OATS-AJSA.
The method of using the isolated forest processes the meteorological data stored in the database with outliers, according to the specific steps of fig. 2, as follows:
in an orphan tree, each node is of two types: external nodes and internal nodes. The external node has no child nodes, while the internal node has two child nodes (T left And T right ) And a test condition consisting of the attribute and the division point. Sample points are assigned to either the left or right subtree depending on the test conditions. For sample point x, its path length in the orphan tree h (x) is defined as the number of edges that pass from the root node to the leaf node. In other words, h (x) represents the number of test conditions that need to be passed when determining the sample point x in the orphan tree.
After defining the isolated tree and path length, an isolated forest needs to be built, specifically as follows:
step 1: the meteorological data set extracted from the database is X, wherein each data point X i Consists of d meteorological features, i.e. x i =(x i1 ,…,x id ). Randomly select from XAnd (3) forming a subset X' by using the sample points as the initial data of the root node of the isolated tree.
Randomly assigning a dimension m (m < d) from d dimensions, randomly generating a cut point n in the subset X', as follows:
min(x ij ,j=m,x ij ∈X′)<n<max(x ij ,j=m,x ij ∈X′)
in the above, x ij Representing data point x i Is the j-th dimension of (2).
Step 2: using the cut point n, a hyperplane is created dividing the subset X' into two subspaces. In this division, sample points smaller than n in a specific dimension are put into the left child node, and sample points greater than or equal to n are put into the right child node. This partitioning facilitates the distribution of data samples into different child nodes in the tree structure according to the cut point setting.
Step 3: by repeatedly performing the above procedure, the cut point and hyperplane can be continuously applied into the data space until one of two conditions is reached: one condition is that all leaf nodes contain only one sample point, and the other condition is that the height of the orphan tree reaches a pre-specified threshold. Until a complete orphan tree is successfully generated.
Step 4: and repeatedly executing the steps 1-3 until the isolated forest is successfully generated. This process is an iterative process in which the algorithm performs data cutting, hyperplane generation and node construction again and again until an isolated forest is formed.
For each data point x i Each isolated tree in the forest is traversed and the path length (or height) of the data point in the tree, i.e., the number of edges traversed from the root node to the leaf node of the tree, is calculated. All these path lengths (heights) will be collected and their average calculated to get an average height. Normalizing the average height of all data points for further analysis or classificationWhether the class data point is an outlier.
The outlier score is calculated according to the following formula:
wherein the method comprises the steps ofThe formula is as follows:
representing meteorological data point x i In the course of->Degree of abnormality in isolated tree constructed of individual samples, E (g (x i ) For data point x) i Calculating the average height in all the isolated trees, then taking the index, ++>Is a parameter function whose value depends on +.> Is the Euler-Mascheroni constant plus +.>Natural logarithm of (a).
To represent the degree of anomaly of data point x in the orphan tree. This degree is based on the average height of the data points in the plurality of isolated trees and is normalized and adjusted by a series of parameters. Its value ranges from 0,1]Between them. When the majority of training samples +.>The values are all close to 0.5, meaning that the meteorological data has no obvious abnormal condition in the construction process of the isolated tree. In other words, if the degree of abnormality of most training samples approaches the intermediate value of 0.5, it can be concluded that the meteorological data as a whole is free from prominent abnormalities.
Wherein,for each meteorological data point x i Th is a threshold.
For the method that the isolated forest is used for identifying possible abnormal points in meteorological data and marking the abnormal points as the abnormal points, the abnormal points are replaced by adopting a linear interpolation method, and the method specifically comprises the following steps:
two normal values before and after the outlier, i.e., adjacent data points of the outlier, are found.
A straight line is constructed using these two normal values, connecting the two points.
And finding the ordinate value of the corresponding position along the straight line according to the position of the outlier on the abscissa, namely, the estimated outlier substitution value. The calculation formula of the specific linear interpolation is as follows:
x estimated =(1-α)*x before +α*x after
wherein the outlier is x abnormal Its previous normal value is x before The latter normal value is x after . Assume that the position of the outlier in the data is t abnormal It is at x before And x after The position ratio between the two is alpha (alpha is more than or equal to 0 and less than or equal to 1). X is x estimated Representing an estimated replacement value for the outlier.
This formula is expressed by the ratio at x before And x after Linear interpolation between, and the proportion of alpha is used for weighting calculation of the estimated value. The value of alpha represents the relative position of the outlier between the preceding and following data points, with the closer to the previous normal value xbefore, the closer alpha to 0, and the closer to the next normal value x after The closer α is to 1.
And carrying out one-hot coding on the analyzed meteorological data according to months, wherein the specific coding form is shown in table 1.
TABLE 1
Month of month | Encoding content |
One month | 100000000000 |
February month | 010000000000 |
March (March) | 001000000000 |
April (April) | 000100000000 |
July of five months | 000010000000 |
June (June) | 000001000000 |
July of Ju | 000000100000 |
August of August | 000000010000 |
September (September) | 000000001000 |
October (October) | 000000000100 |
October month | 000000000010 |
December month | 000000000001 |
In order to achieve the following objectives in processing meteorological data: the method comprises the steps of storing most information, reducing the data size and dimension, removing noise, and performing dimension reduction operation by adopting a t-distribution-random adjacent embedding method (t-distributed Stochastic Neighbor Embedding, t-SNE) method, so that a lower-dimension representation is obtained. This approach facilitates rapid processing of data while avoiding the effects of duplicate or redundant information. According to fig. 3, the specific steps are as follows:
first, consider the original high-dimensional dataset to be represented as { x } 1 ,x 2 ,…,x n X, where x i Is a high-dimensional data point, the invention refers to multi-element meteorological data processed by an isolated forest, and the goal of the t-SNE is to map the high-dimensional data point into a low-dimensional space, which is assumed hereThe dimension is D, and the mapped low-dimensional dataset is represented as { y } 1 ,y 2 ,…,y n -wherein y i Is the corresponding low-dimensional data point.
Step 1: calculating high-dimensional similarity probability P ij For high dimension data point x i And x j The similarity between them is measured using a gaussian distribution. The gaussian distribution formula is as follows:
wherein p is j|i Representing data point x i Generating x j Conditional probability, sigma of i 2 A variance parameter representing the distance calculation. P is p j|i Represented in a high-dimensional space as data point x i Generating data point x j And P ij Representing a high-dimensional probability of similarity for measuring data point x i And x j Similarity between them. P is p j|i The probability of Pij is estimated using a gaussian distribution. II x i -x j ‖ 2 The square of the euclidean distance (Euclidean distance) is represented and used to measure the distance between two vectors.
Step 2: calculating a low-dimensional similarity probability Q ij For low-dimensional data point y i And y j The t-distribution is used to measure the similarity between them. the t distribution formula is as follows:
wherein q j|i To represent the data point y i Generating y j Conditional probability of (2).
Step 3: the KL-divergence is calculated for the difference between the Heng Lianggao-dimensional probability distribution P and the low-dimensional probability distribution Q. The calculation formula is as follows:
wherein KL (P Q) represents the divergence, P ij Representing high-dimensional similarity probability, Q ij Representing a low dimensional similarity probability.
Step 4: updating a low-dimensional representation { y } using gradient descent 1 ,y 2 ,…,y n To reduce KL divergence. The updated rules are:
where β is the learning rate, the step size updated in each iteration is controlled. θ is a global parameter for controlling the probability of similarity P in high-dimensional space ij Is a distribution of (a). It is used in the t-SNE algorithm to balance the parameters of local and global similarity. Higher θ values will result in more high-dimensional data points being considered similar, affecting the global structure of the t-SNE.Is for each low-dimensional data point y i For controlling the probability of similarity Q in a low-dimensional space ij Is a distribution of (a). In the t-SNE algorithm, +.>Is variable and is updated continuously with the iterative process to adjust for the similarity between the low-dimensional data points. Specifically, & gt>Is a parameter used to scale the gradient when updating the low-dimensional representation that helps maintain the relative distance between the data points.
And (3) repeating the steps 2 to 4 until reaching a convergence condition, so that the variation of certain iteration times or KL divergence is small. When the iteration is ended, the resulting low-dimensional representation is the final result.
The weather data after the t-SNE dimension reduction is normalized, and the specific formula is as follows:
wherein x is * Representing the normalized value of meteorological data, x min Representing the minimum value of meteorological data, x max Representing the maximum value of the meteorological data.
The structure of the OATS-AJSA-GRU humidity prediction model was established according to fig. 4: the method comprises the following steps:
step 1: the over-parameters (iteration range, dropout ratio, number of neurons in hidden layer (hidden-size), learning rate size) in the GRU are optimized optimally using the OATS-ASJA optimization algorithm. Iteration range [1, 50], dropout ratio [0.01,0.20], hidden-size range [10, 200], learning rate range [0.001,0.01].
Step 2: parameters of an artificial jellyfish search algorithm (Artificial Jellyfish Search Algorithm, AJSA) are initialized by using an orthogonal table test method (Orthogonal Array Testing Strategy, OATS), and the parameters are selected from population quantity, search radius and population initialization strategies.
Step 3: determining the number of the population, searching the radius, initializing the parameter range of the strategy of the population, and filling in the orthogonal table. The L16 orthogonal table is selected to fill in the number of the population, the searching radius and the combination of three parameters of the population initialization strategy, and the specific orthogonal table is shown in the table 2.
TABLE 2
Step 4: ASJA experiments are carried out according to the combination of the orthogonal tables, the population quantity, the searching radius and the population initialization strategy are optimized. The results of each experiment were recorded and the optimal parameter configuration was obtained.
Step 5: calculating the fitness value of jellyfish, and calculating the fitness value of jellyfish by using a cross entropy loss function, wherein the specific calculation formulas of the cross entropy loss function and the fitness value of jellyfish are as follows:
where i is the number of iterations (iterations), l is the learning rate (learning_rate), d is the dropout ratio (dropout_rate), h is the number of hidden layer neurons (hidden_neurons), N is the number of samples, y i Is the actual label of the i-th sample,is the model's predictive probability for the ith sample, and F is the fitness.
Step 6: and (3) determining the optimal individual positions of the population, sorting all jellyfish individuals according to the fitness value calculated in the step (5), and arranging the jellyfish individuals from high to low according to the fitness value. The optimal individual is the jellyfish individual with the highest fitness value. Thus, after ranking, the first individual (highest fitness) is the optimal individual. The position is the optimal individual position of the population.
Step 7: because the ocean current contains a large amount of food, jellyfish gathers in the ocean current, the jellyfish can form jellyfish groups along with the movement of the ocean current along with the time, and the jellyfish becomes an ocean current stage, and the position of each jellyfish in the ocean current movement can be updated, and the specific formula is as follows:
X i (t+1)=X i (t)+rand(0,1)×(X * -σ×rand(0,1))×μ
wherein X is i (t+1) is the position of the ith jellyfish update, X i (t) is the position of the ith jellyfish, X * Is the optimal position in the current jellyfish population, σ is the distribution coefficient, σ=3. Mu is the average position of all jellyfish in the population and rand (0, 1) is a random number from 0 to 1.
However, the temperature and wind direction of the ocean current will affect the distribution of jellyfish groups, so as to form new jellyfish groups, and at this time, jellyfish will move inside the jellyfish groups, which is called as jellyfish group stage. In the initial generation stage of jellyfish population, most jellyfish will follow the jellyfish population, namely jellyfish will move around its own position and become passive movement, so the position of jellyfish will be updated, and the specific formula is as follows:
X i (t+1)=X i (t)+α×rand(0,1)×(U b -L b )
wherein U is b 、L b Is the upper and lower bounds of the search space of the jellyfish population, α is the motion coefficient, α=0.1.
Over time, jellyfish gradually develops subjective motility itself and gradually approaches to a companion who can find more foods, which becomes active movement, and the specific calculation formulas of the movement direction of jellyfish and the position after update are as follows:
wherein,is the direction of jellyfish movement,/->Is the distance X of jellyfish moving along the moving direction j (t) is the position of the j-th jellyfish randomly selected from the jellyfish population, f (X) i )、f(X j ) And respectively representing the fitness values of the ith jellyfish and the jth jellyfish at the current position.
In order to regulate the movement of jellyfish between following ocean current movement and movement inside jellyfish groups, a time control mechanism is introduced to control the active movement or the passive movement during switching, and a specific formula of a time control function is as follows:
wherein C (t) is a control function of the time control mechanism, specifically a control function which is in interval [0, 1] along with the iteration number t]Random value in, t is the current iteration number, t max Is the maximum number of iterations initially set. The rule for the number of iterations is as follows:
dynamically adjusting the maximum iteration number t during the operation of the AJSA algorithm by using adaptive adjustment max In the initial stage of the algorithm, the invention sets the initial iteration number to 50 times, calculates the proportion of the current iteration number and the total iteration number in each iteration, gradually reduces the iteration number, and dynamically adjusts the rest iteration number according to the proportion of the current iteration number by using a linear decreasing function. The linearly decreasing function ensures that the number of iterations is relatively large at the early stages of the algorithm in order to fully explore the global search space, then gradually decreasing the number of iterations, and finally allowing the algorithm to explore more intensively in the local search space. When the number of remaining iterations decreases to a set threshold, it may be selected to stop the algorithm's iterations. The linear decreasing iteration formula is as follows:
wherein t is min Is the minimum number of iterations, and t is the current number of iterations. The above formula dynamically calculates the remaining iteration number t according to the current iteration number rem Thereby gradually reducing the iteration times in the iteration process. As t increases, t rem Will approach t min So that the local search is more focused at a later stage of the algorithm.
Step 8: judging according to the time control function, if C (t) is more than or equal to 0.5, enabling jellyfish to enter into the ocean current stage, and updating the individual position according to a formula. If C (t) <0.5, the jellyfish enters a jellyfish group stage, whether the jellyfish actively moves or passively moves in the jellyfish group stage is judged according to whether rand (0, 1) is larger than or equal to 0.5, the jellyfish actively moves when the number of the jellyfish is larger than or equal to 0.5, the jellyfish passively moves when the number of the jellyfish is smaller than 0.5, and the individual position is updated according to a formula.
Step 9: the jellyfish can judge whether the position of the jellyfish exceeds the defined searching range in the searching process, and correct the jellyfish, and the specific judging and correcting method is as follows:
defining clip functions, wherein the content is as follows:
clip(value,v min ,v max )
where value is the value to be limited, v min Is the minimum allowed value, and if the value is smaller than this value, it is corrected to v min ,v max Is the maximum allowed, and if value is greater than this value, it is corrected to v max . In short, the clip function will ensure that value falls at v min And v max If it is outside this range, it is corrected to the nearest boundary value.
The formula for correcting jellyfish position according to clip function is:
X new =clip(X,X min ,X max )
wherein X is new Is the calculated new position, X min 、X max Is the minimum and maximum allowable value of the position.
And updating the corrected jellyfish position, performing GRU model training, recalculating the fitness value, and updating the optimal individual position of the jellyfish population.
Step 10: and judging the iteration times T and the threshold constant T, if T is smaller than or equal to T, carrying out iteration again, otherwise, outputting an optimal solution and an optimal value.
Step 11: and putting the processed meteorological data and the GRU super parameters after optimizing into a GRU unit for humidity prediction.
Based on the same inventive concept, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the adaptive oat-AJSA based GRU humidity prediction method described above when the processor executes the computer program.
Based on the same inventive concept, embodiments of the present application provide a computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of the adaptive oat-AJSA based improved GRU humidity prediction method described above.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow in the flowchart, and combinations of flows in the flowchart, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows.
The above embodiments are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereto, and any modification made on the basis of the technical scheme according to the technical idea of the present invention falls within the protection scope of the present invention.
Claims (8)
1. The improved GRU humidity prediction method based on the adaptive OATS-AJSA is characterized by comprising the following steps of:
step 1, acquiring historical multi-element meteorological data and storing the data in a MinIO database in a distributed manner;
step 2, extracting multi-element meteorological data from a MinIO database as a prediction data set, identifying abnormal values in the prediction data set by using an isolated forest method, and replacing the abnormal values by using a linear interpolation method to obtain a replaced data set;
step 3, reducing the dimension of the replaced data set obtained in the step 2 by adopting a t distribution-random adjacent embedding method to obtain a dimension-reduced data set, and normalizing the dimension-reduced data set to obtain a normalized data set;
step 4, an OATS-AJSA-GRU humidity prediction model is established, and the ultra-parameters in the GRU are optimized by utilizing an OATS-AJSA optimization algorithm to obtain optimized ultra-parameters; the specific process is as follows:
step 41, initializing parameters of an artificial jellyfish searching algorithm by using an orthogonal table testing method, wherein the parameters of the artificial jellyfish searching algorithm comprise population quantity, searching radius and a population initializing strategy;
step 42, taking the super parameters of the GRU as jellyfish, calculating the fitness value of the jellyfish by using a cross entropy loss function in each iteration, and taking the jellyfish position with the highest fitness value as the optimal individual position of the jellyfish population in the current iteration; the fitness value is calculated as follows:
wherein L (i, L, d, h) represents a cross entropy loss function, i is the iteration number, L is the learning rate, d is the dropout proportion, h is the number of hidden layer neurons, N is the number of samples, y i Representing the actual label of the i-th sample,representing the prediction probability of the model to the ith sample, and F represents the fitness;
step 43, a time control mechanism is introduced to determine that the current iteration jellyfish enters the ocean current stage or jellyfish group stage, namely:
wherein C (t) represents a control function of the time control mechanism, t represents the current iteration number, rand (0, 1) represents a random number between 0 and 1, t max Representing the initially set maximum iteration number;
the maximum iteration number is dynamically adjusted by using an adaptive adjustment method, and the formula is as follows:
wherein t is rem Representing the number of remaining iterations, t min Representing a minimum number of iterations;
step 44, judging according to the time control function, if C (t) is more than or equal to 0.5, enabling jellyfish to enter a ocean current stage, and updating the individual position according to the following formula:
X i (t+1)=X i (t)+rand(0,1)×(X * -σ×rand(0,1))×μ
wherein X is i (t+1) represents the position after the i-th jellyfish is updated, X i (t) represents the current iteration position of the ith jellyfish, X * Representing the optimal individual position in the current iteration jellyfish population, wherein sigma is a distribution coefficient, sigma=3, and mu represents the average position of all jellyfishes in the population;
if C (t) <0.5, the jellyfish enters a jellyfish group stage, whether the jellyfish actively moves or passively moves in the jellyfish group stage is further judged according to whether rand (0, 1) is larger than or equal to 0.5, if rand (0, 1) is larger than or equal to 0.5, the jellyfish is passively moved in the jellyfish group stage, and the individual position is updated according to the following formula:
X i (t+1)=X i (t)+α×rand(0,1)×(U b -L b )
where α is the motion coefficient, α=0.1, u b 、L b Respectively representing an upper bound and a lower bound of jellyfish population search space;
if rand (0, 1) <0.5, indicating that jellyfish is actively moving in the jellyfish group phase, the individual position is updated according to the following formula:
wherein,indicates the direction of jellyfish movement, +.>Represents the distance that jellyfish moves along the movement direction, X j (t) represents the current iteration position of the jth jellyfish randomly selected from the jellyfish population, f (X) i )、f(X j ) Respectively representing the fitness values of the ith jellyfish and the jth jellyfish at the current iteration position;
step 45, correcting the jellyfish position according to the clip function:
X new =clip(X,X min ,X max )
wherein X is new Represents the corrected jellyfish position, X min 、X max Respectively representing minimum and maximum allowable values of the position;
step 46, judging the magnitudes of the current iteration times T and the threshold constant T, if the current iteration times T are smaller than the threshold constant T, continuing iteration, otherwise, outputting the optimal value of the super parameter;
and 5, putting the optimized super parameters and the normalized data set obtained in the step 3 into GRU for prediction, and obtaining a humidity prediction result.
2. The adaptive oat-AJSA-based improved GRU humidity prediction method of claim 1 wherein the specific procedure of step 1 is as follows:
1) Accessing the ECMWF by using the API and downloading the NC data file;
2) Analyzing the downloaded NC data file by using a xarray library of Python to obtain a CSV data file;
3) Configuring and connecting to a MinIO server, uploading the CSV data file obtained in the step 2 to a MinIO socket by using a MinIO SDK, and taking the CSV data file as a historical multi-element weather database, wherein the multi-element weather data comprises temperature, humidity, wind speed and pressure data;
4) Storing user information by using a MySQL database, creating an API endpoint by using a Python Web framework, and calling the historical multi-element weather data in the historical multi-element weather database by the user stored in the MySQL database.
3. The adaptive oat-AJSA-based improved GRU humidity prediction method according to claim 1, wherein the identifying abnormal values in the predicted dataset by using the isolated forest method in step 2 is specifically:
1) Setting the prediction data set as X, each data point X in X i Consists of d meteorological features, i.e. x i =(x i1 ,…,x id ),x i1 、x id Representing data point x i Is randomly selected from XThe data points form a subset X ′ Will subset X ′ As the root node of an orphan tree;
2) Randomly designating dimension m, and m<d, at subset X ′ A cutting point n is randomly generated, and the formula is as follows:
min(x ij ,j=m,x ij ∈X ′ )<n<max(x ij ,j=m,x ij ∈X ′ )
wherein x is ij Representing data point x i Is the jth meteorological feature of (2);
3) Creating a hyperplane from the cut point n, utilizing the hyperplane to pair sub-planes X ′ Dividing, i.e. subset X ′ Data points smaller than n are put into the left sub-node, and the subset X is obtained ′ The data points with the number of n or more are put into a right child node;
4) Repeating the steps 2) and 3) until all leaf nodes of the isolated tree contain only one data point or the height of the isolated tree reaches a preset upper height limit, namely generating an isolated tree;
5) Repeating the steps 1) -4) to obtain an isolated forest;
6) For each data point X in X i Traversing each isolated tree in the isolated forest to obtain a data point x i The heights in each isolated tree are averaged to obtain the average height, and the average heights of all data points are normalized;
7) Using the normalized average height g (x i ) Calculating outliersScore ofTaking data points corresponding to abnormal value scores which are larger than or equal to a preset threshold value as abnormal values; the outlier score is calculated as follows:
wherein,ξ=0.5772156649;E(g(x i ) For data point x) i Calculating average heights in all the isolated trees, and taking an index after normalization; />Representing a parametric function->Represents the Euler-Mascheroni constant plus +.>Natural logarithm of (a).
4. The adaptive oat-AJSA-based improved GRU humidity prediction method according to claim 1, wherein the replacing of the outlier with the linear interpolation method in step 2 is specifically:
1) Finding adjacent data points of the abnormal value, namely two normal values before and after the abnormal value;
2) A straight line is constructed by using two normal values before and after the abnormal value, namely:
x estimated =(1-α)*x before +α*x after
wherein x is estimated Alternative value, x, representing outlier before 、x after The normal values before and after the abnormal value are respectively represented, and alpha represents that the abnormal value is x before And x after The position ratio of the two components is more than or equal to 0 and less than or equal to 1;
3) And finding out the corresponding ordinate value along the constructed straight line according to the position of the outlier on the abscissa, namely, the replacement value of the outlier.
5. The adaptive oat-AJSA-based improved GRU humidity prediction method according to claim 1, wherein in step 3, the t distribution-random proximity embedding method is adopted to reduce the dimension of the replaced data set obtained in step 2, so as to obtain a dimension-reduced data set, which is specifically as follows:
1) Setting the high-dimensional data set as { x } which is the data set after replacement 1 ,x 2 ,…,x n The dimensionality-reduced data set is { y } 1 ,y 2 ,…,y n },x 1 ,x 2 ,…,x n Representing high-dimensional data points in a high-dimensional dataset, y 1 ,y 2 ,…,y n Representing low-dimensional data points in the dimensionality reduced data set, and calculating high-dimensional similarity probability P by using Gaussian distribution ij The method comprises the steps of carrying out a first treatment on the surface of the Wherein, the Gaussian distribution formula is as follows:
wherein p is j|i Representing a high-dimensional data point x i Generating x j Conditional probability, sigma of i 2 Variance parameter, ||x, representing distance calculation i -x j II represents x i And x j Euclidean distance, x k 、x l Representing high-dimensional data points in a high-dimensional dataset;
2) Calculating low-dimensional similarity probability Q using t distribution ij Wherein the t distribution formula is as follows:
wherein q j|i Representing low-dimensional data points y i Generating y j Conditional probability of y k 、y l Representing low-dimensional data points in the reduced-dimension dataset;
3) Using the difference between the KL-divergence Heng Lianggao-dimensional probability distribution and the low-dimensional probability distribution, the formula is as follows:
wherein KL (p||q) represents KL divergence;
4) Updating the low-dimensional representation using a gradient descent method, the updating rule being as follows:
where beta represents the learning rate, theta represents the global parameter,is a parameter used to scale the gradient when updating the low-dimensional representation;
5) Repeating 2) -4) until a convergence condition is reached, i.e. a preset number of iterations is reached or the difference between the KL-divergence of the current iteration and the KL-divergence of the last iteration is smaller than a preset threshold.
6. The adaptive oat-AJSA-based improved GRU humidity prediction method of claim 1 wherein the specific procedure of step 41 is as follows:
411, determining parameter ranges of three parameters, namely population quantity, search radius and population initialization strategy, and selecting different combinations of the three parameters, namely population quantity, search radius and population initialization strategy, according to the parameter ranges;
step 412, performing an ASJA experiment on each combination of the L16 orthogonal table, and selecting the corresponding combination when the experimental result is optimal as the optimal combination of three parameters, namely, the optimal parameter of the artificial jellyfish searching algorithm, such as population number, searching radius and population initializing strategy.
7. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor, when executing the computer program, implements the steps of the adaptive oat-AJSA-based improved GRU humidity prediction method of any of claims 1 to 6.
8. A computer readable storage medium storing a computer program, which when executed by a processor implements the steps of the adaptive oat-AJSA-based improved GRU humidity prediction method of any one of claims 1 to 6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311599596.7A CN117313555B (en) | 2023-11-28 | 2023-11-28 | GRU humidity prediction method based on self-adaptive OATS-AJSA |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311599596.7A CN117313555B (en) | 2023-11-28 | 2023-11-28 | GRU humidity prediction method based on self-adaptive OATS-AJSA |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117313555A CN117313555A (en) | 2023-12-29 |
CN117313555B true CN117313555B (en) | 2024-03-08 |
Family
ID=89260717
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311599596.7A Active CN117313555B (en) | 2023-11-28 | 2023-11-28 | GRU humidity prediction method based on self-adaptive OATS-AJSA |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117313555B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114707421A (en) * | 2022-04-28 | 2022-07-05 | 河北工业大学 | IJS-SVR model-based short-term wind power prediction method |
CN114781688A (en) * | 2022-03-21 | 2022-07-22 | 广东电网有限责任公司广州供电局 | Method, device, equipment and storage medium for identifying abnormal data of business expansion project |
CN115643189A (en) * | 2022-10-10 | 2023-01-24 | 科大国创软件股份有限公司 | Network anomaly detection method based on group intelligent algorithm and isolated forest |
CN115994629A (en) * | 2023-03-23 | 2023-04-21 | 南京信息工程大学 | GN-RBF-based air humidity prediction method and system |
CN116245015A (en) * | 2023-01-09 | 2023-06-09 | 四川通信科研规划设计有限责任公司 | Data change trend prediction method and system based on deep learning |
CN116976192A (en) * | 2023-06-24 | 2023-10-31 | 北京工业大学 | JS-BP model-based die forging defect accurate repair process parameter decision method |
-
2023
- 2023-11-28 CN CN202311599596.7A patent/CN117313555B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114781688A (en) * | 2022-03-21 | 2022-07-22 | 广东电网有限责任公司广州供电局 | Method, device, equipment and storage medium for identifying abnormal data of business expansion project |
CN114707421A (en) * | 2022-04-28 | 2022-07-05 | 河北工业大学 | IJS-SVR model-based short-term wind power prediction method |
CN115643189A (en) * | 2022-10-10 | 2023-01-24 | 科大国创软件股份有限公司 | Network anomaly detection method based on group intelligent algorithm and isolated forest |
CN116245015A (en) * | 2023-01-09 | 2023-06-09 | 四川通信科研规划设计有限责任公司 | Data change trend prediction method and system based on deep learning |
CN115994629A (en) * | 2023-03-23 | 2023-04-21 | 南京信息工程大学 | GN-RBF-based air humidity prediction method and system |
CN116976192A (en) * | 2023-06-24 | 2023-10-31 | 北京工业大学 | JS-BP model-based die forging defect accurate repair process parameter decision method |
Also Published As
Publication number | Publication date |
---|---|
CN117313555A (en) | 2023-12-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wang et al. | Correlation aware multi-step ahead wind speed forecasting with heteroscedastic multi-kernel learning | |
CN110619360A (en) | Ultra-short-term wind power prediction method considering historical sample similarity | |
CN110782658B (en) | Traffic prediction method based on LightGBM algorithm | |
Mohan et al. | Deep learning based weighted SOM to forecast weather and crop prediction for agriculture application | |
CN112906298B (en) | Blueberry yield prediction method based on machine learning | |
CN112418498B (en) | Temperature prediction method and system for intelligent greenhouse | |
CN107169621A (en) | A kind of Dissolved Oxygen in Water Forecasting Methodology and device | |
CN116596044B (en) | Power generation load prediction model training method and device based on multi-source data | |
CN113361785A (en) | Power distribution network short-term load prediction method and device, terminal and storage medium | |
Lee et al. | An efficient prediction for heavy rain from big weather data using genetic algorithm | |
WO2023024615A1 (en) | System and method for performing precise irrigation on the basis of plant wilting degree | |
CN113221447A (en) | Soil humidity prediction method for optimizing BP neural network based on improved genetic algorithm | |
CN115049024A (en) | Training method and device of wind speed prediction model, electronic equipment and storage medium | |
CN113344243B (en) | Wind speed prediction method and system for optimizing ELM (ELM) based on improved Harris eagle algorithm | |
CN114091745A (en) | Industry power consumption prediction method based on improved multi-storage pool echo state network | |
CN118013457A (en) | Wind speed prediction method and system based on multi-mode data | |
CN117313555B (en) | GRU humidity prediction method based on self-adaptive OATS-AJSA | |
CN113962819A (en) | Method for predicting dissolved oxygen in industrial aquaculture based on extreme learning machine | |
Shi et al. | Dissolved oxygen prediction using regularized extreme learning machine with clustering mechanism in a black bass aquaculture pond | |
CN115713144A (en) | Short-term wind speed multi-step prediction method based on combined CGRU model | |
CN114511078A (en) | BP neural network prediction method and device based on multi-strategy sparrow search algorithm | |
CN112906871A (en) | Temperature prediction method and system based on hybrid multilayer neural network model | |
Rungyaem et al. | Comparison of 3D Rice Organs Point Cloud Classification Techniques | |
Islam et al. | Machine learning models to predict soil moisture for irrigation schedule | |
Jayashree et al. | Prediction of reference crop evapotranspiration: Empirical and machine learning approaches |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |