CN115994629A - GN-RBF-based air humidity prediction method and system - Google Patents
GN-RBF-based air humidity prediction method and system Download PDFInfo
- Publication number
- CN115994629A CN115994629A CN202310287141.5A CN202310287141A CN115994629A CN 115994629 A CN115994629 A CN 115994629A CN 202310287141 A CN202310287141 A CN 202310287141A CN 115994629 A CN115994629 A CN 115994629A
- Authority
- CN
- China
- Prior art keywords
- air humidity
- preset
- data
- humidity data
- grid
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 65
- 230000002159 abnormal effect Effects 0.000 claims abstract description 29
- 238000013528 artificial neural network Methods 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 16
- 238000010606 normalization Methods 0.000 claims abstract description 10
- 238000005070 sampling Methods 0.000 claims description 52
- 239000013598 vector Substances 0.000 claims description 32
- 241000257303 Hymenoptera Species 0.000 claims description 27
- 230000008569 process Effects 0.000 claims description 20
- 239000003016 pheromone Substances 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 17
- 230000008859 change Effects 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000004891 communication Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 abstract description 7
- 238000003062 neural network model Methods 0.000 abstract 1
- 238000007781 pre-processing Methods 0.000 abstract 1
- 230000004913 activation Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 3
- 238000005520 cutting process Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000008020 evaporation Effects 0.000 description 2
- 238000001704 evaporation Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 241001465754 Metazoa Species 0.000 description 1
- 230000002547 anomalous effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003631 expected effect Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000002688 persistence Effects 0.000 description 1
- 230000009024 positive feedback mechanism Effects 0.000 description 1
- 238000012887 quadratic function Methods 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Chemical compound O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- 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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an air humidity prediction method and system based on GN-RBF, comprising the following steps: preprocessing data, namely removing abnormal data points by adopting an abnormal data removing method, filling up the vacant values of the time sequence, and obtaining the data of the complete time sequence; performing space-time matching on the air humidity data and simultaneously performing normalization processing on the data; optimizing Gauss-Newton algorithm by ant colony algorithm so as to optimize RBF neural network, establishing GN-RBF neural network time sequence regression prediction model, and performing RBF neural network weight optimization in GN-RBF combined model, wherein the processed air humidity data is used as input; and training a network model according to the sample set, and inputting data to be predicted into the trained neural network model to obtain predicted air humidity data. The invention provides an air humidity prediction method with high efficiency and excellent accuracy, which can greatly improve the quality of air humidity prediction.
Description
Technical Field
The invention belongs to the field of meteorological data prediction, and particularly relates to an air humidity prediction method and system based on GN-RBF.
Background
Air humidity plays an important role in many ways, such as the relative humidity commonly used in meteorology, which reflects the likelihood of rain and fog. In hot weather, high relative humidity can be perceived as hotter by humans and other animals, as it prevents evaporation of perspiration. The relative humidity is determined on the one hand by the absolute humidity and on the other hand by the air temperature. In cold areas and seasons, the air humidity is easily saturated, and in cases where the absolute humidity or the water vapor pressure is not too high, the relative humidity may be high. Under the same absolute humidity conditions, the relative humidity in warm areas and seasons tends to be low, and humans can thus formulate a heat index.
Compared with the prediction methods of other meteorological elements, the prediction method of the air humidity starts later, and is the technical parameter which is most difficult to accurately quantify. The traditional air humidity prediction model uses the same network weight to simulate the space-time relationship between the air humidity and the prediction variable, and the properties of the space unknowing and space compactness convolution kernel play a good role in improving the calculation efficiency and explaining the translation invariance equivalence. However, the space-time capability of the convolution kernel for adapting to the air humidity under different space positions and predictors is ignored, so that the space-time difference of the air humidity cannot be reflected, and most of the problems of inaccurate prediction exist. Therefore, the method for efficiently and accurately predicting the air humidity is of great significance.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the traditional air humidity prediction model uses the same network weight to simulate the space-time relationship between the air humidity and the prediction variable, but ignores the space-time capability of the convolution kernel for adapting to the air humidity under different space positions and prediction factors, so that the space-time difference of the air humidity cannot be reflected, and most of the problems of inaccurate prediction exist.
In order to solve the technical problems, the invention adopts the following technical scheme:
the GN-RBF-based air humidity prediction method comprises the following steps of, for a target area, obtaining the predicted air humidity of the target area in a preset future time period taking the current moment as a starting point:
step A: aiming at a target area, acquiring air humidity data respectively corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area based on a preset sampling interval and preset grid division; removing abnormal air humidity data, filling sequence vacant air humidity data, and updating air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area;
and (B) step (B): carrying out normalization processing on air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid region in the target region, and then carrying out space-time matching to obtain an air humidity data sequence corresponding to each grid region;
step C: based on the air humidity data sequences respectively corresponding to the grid areas, an air humidity prediction model is used, wherein the air humidity data sequences respectively corresponding to the grid areas in a preset historical time period taking the current moment as an end point are used as input, the air humidity data sequences respectively corresponding to the grid areas in a preset future time period taking the current moment as a starting point are used as output, and the predicted air humidity data sequences respectively corresponding to the grid areas are obtained;
step D: based on the predicted air humidity data sequences corresponding to the grid areas respectively, taking the average value of the air humidity data of the grid areas at the same sampling time as the predicted air humidity of the target area at the sampling time, and obtaining the predicted air humidity of the target area in a preset future time period taking the current time as the starting point.
Preferably, in the step C, for the air humidity prediction model, a Gauss-Newton algorithm is optimized by an ant colony algorithm in the training process, and preset parameters to be optimized of the air humidity prediction model are optimized to obtain optimal parameters to be optimized, and then the air humidity prediction model is built based on the training of the optimal parameters to be optimized, and the specific process is as follows:
step C1: parameter vector to be optimized based on preset parameter composition to be optimized
Iterative execution in combination with the objective function of the air humidity prediction modelThe method comprises the following steps:
step C1.1: based on the parameter vector to be optimized
Combining with an objective function of an air humidity prediction model to obtain a jacobian matrix corresponding to a parameter vector to be optimized in the current iteration>
And air humidity prediction model error->
，
Represent the firstnCorresponding parameter vectors to be optimized in the secondary iteration;
step C1.2: jacobian matrix corresponding to parameter vector to be optimized in current iteration
And air humidity prediction model error->
Obtaining the increment corresponding to the parameter vector to be optimized in the current iteration>
；
Step C1.3: aiming at increment corresponding to parameter vector to be optimized in current iteration
Judging->
Whether the data of (a) is smaller than the preset change increment +.>
If->
Wherein each data is smaller than the increment of change +.>
Stopping iteration, and carrying out parameter vector to be optimized in the current iteration>
As optimal parameters to be optimized; if->
The presence data is not less than delta change +.>
B, obtaining an optimal increment direction corresponding to the parameter vector to be optimized through the steps a-d, and further enabling the parameter vector to be optimized based on the optimal increment direction
Returning to the step C1.1;
step a: constructing a directed graph based on a preset number of ants; each node in the directed graph is connected with a preset number of edges representing the increment direction, and the number of the edges is the same as the number of ants;
step b: based on the directed graph, each ant performs direction selection based on the corresponding tabu table and the pheromone of each side representing the increment direction; the first element of the tabu list is the initial placement direction of ants, which is the initial direction of ants
The corresponding point is the center of circle, each ant is placed in the direction of the center of circle;
step c: updating the tabu list, respectively judging whether the tabu list corresponding to each ant contains all the increment directions, if so, ending the current iteration, and executing the step d; if the tabu table corresponding to the ants does not contain all the increments, returning the ants which do not contain all the increments to the executing step b;
step d: the iteration number is added by one, whether the current iteration number reaches the maximum iteration number is judged, if the current iteration number reaches the preset maximum iteration number, the iteration is ended, and the optimal increment direction is obtained based on the current pheromone of each side representing the increment direction; and c, if the maximum iteration number is not reached, updating the pheromone of each side in the directed graph, and returning to the step b.
Preferably, the air humidity prediction model is obtained by training an RBF neural network.
Preferably, the preset parameters to be optimized include connection weights in the network.
Preferably, in the step a, abnormal air humidity data is removed by an isolated forest method, and a front-back data filling method is adopted to fill in the sequence vacant air humidity data.
Preferably, in the step a, abnormal air humidity data is removed by an isolated forest method, and the specific process is as follows:
step A1: constructing a preset number of isolated trees aiming at air humidity data respectively corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area;
step A2: based on the constructed preset quantity of isolated trees, taking air humidity data corresponding to one sampling moment of one grid area as one sample, and obtaining anomaly scores corresponding to the samples respectively through the following formulas:
wherein ,
；
in the formula ,
representation->
Sample ∈ in the individual samples>
At the corresponding anomaly score, ++>
For sample->
From root node to sample in an orphan tree->
Path length of the leaf node where +.>
For sample->
From root node to sample in all orphaned trees ≡ ->
Path Length expected value of the located leaf node, < >>
For use->
Constructing an average path length of an isolated tree by using the samples;
For sample->
From the root node of the tree to sample->
The number of edges experienced during the leaf node;
Representation and sample->
The same number of samples at a leaf node, +.>
Representation +.>
Constructing an average path length of an isolated tree by the strip sample;
For harmonizing the number, add>
，
Is Euler constant;
step A3: and removing samples which do not meet the preset abnormal score based on the abnormal score corresponding to each sample.
Preferably, in the step B, for the air humidity data corresponding to each sampling time in the preset historical time period with the current time as the end point in each grid area in the target area, the following steps are specifically executed, and normalization processing is performed:
in the formula ,
for normalized air humidity data, +.>
For raw air humidity data, +.>
Maximum value in air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point for each grid area>
And the minimum value in the air humidity data corresponding to each sampling time in the preset historical time period taking the current time as the end point is used for each grid area.
The system of the air humidity prediction method based on the GN-RBF comprises a data acquisition module, a data processing module, an air humidity prediction module and an air humidity output module;
aiming at the target area, based on a preset sampling interval and preset grid division, the data acquisition module is used for acquiring air humidity data respectively corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area; removing abnormal air humidity data, filling sequence vacant air humidity data, and updating air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area;
the data processing module is used for carrying out normalization processing on air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid region in the target region, and then carrying out space-time matching to obtain an air humidity data sequence corresponding to each grid region;
the air humidity prediction module is used for obtaining a predicted air humidity data sequence corresponding to each grid region by using an air humidity prediction model taking the air humidity data sequence corresponding to each grid region in a preset historical time period taking the current moment as an end point as input and taking the air humidity data sequence corresponding to each grid region in a preset future time period taking the current moment as a starting point as output based on the air humidity data sequence corresponding to each grid region respectively;
the air humidity output module is used for obtaining the predicted air humidity of the target area in a preset future time period by taking the current moment as a starting point by taking the average value of the air humidity data of each grid area at the same sampling moment as the predicted air humidity of the target area at the sampling moment based on the predicted air humidity data sequences respectively corresponding to each grid area.
The terminal of the GN-RBF-based air humidity prediction method comprises a memory and a processor, wherein the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, so that the GN-RBF-based air humidity prediction method is executed.
The beneficial effects of the invention are as follows: the invention provides an air humidity prediction method and system based on GN-RBF. The searching process adopts a distributed parallel computing mode, each individual in the ant colony performs computation at the same time, so that the computing capacity and the operating efficiency of the algorithm are greatly improved, the optimal increment direction is finally approximated, and the iteration speed and the iteration accuracy of the Gaussian Newton algorithm are improved.
The invention introduces a single hidden layer radial basis RBF feedforward neural network, the RBF network can approach any nonlinear function, can process the regularity that is difficult to analyze in a system, and compared with the traditional network, the RBF network has good generalization capability, and has the problems of optimally approaching and overcoming local minima. The processed air humidity data is directly used as input, so that training time is saved, and training accuracy is improved.
The GN-RBF network algorithm provided by the invention is a feedforward network with excellent performance, simple framework, obvious global convergence capability and strong generalization capability and universality, can obviously reduce learning time of network learning, obtains more excellent variable weight, and further can ensure the requirement of more rapid prediction of air humidity. The algorithm designed herein can be applied not only to the problem of air humidity prediction, but also embedded in other algorithms. Compared with the traditional air humidity prediction model, the novel Gaussian Newton algorithm designed in the method accelerates the convergence speed of the algorithm and the accuracy of prediction, provides a novel method and thinking for the air humidity prediction problem, and further expands the application depth and breadth of the RBF network.
Drawings
FIG. 1 is a schematic flow chart of the present invention;
FIG. 2 is a flow chart of outlier forest rejection in the present invention;
FIG. 3 is a flow chart of an ant colony optimization Gauss-Newton algorithm in the invention;
FIG. 4 is a GN-RBF neural network structure in the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples will provide those skilled in the art with a more complete understanding of the invention, but are not intended to limit the invention in any way.
The GN-RBF-based air humidity prediction method comprises the following steps, as shown in fig. 1, for a target area, of obtaining the predicted air humidity of the target area in a preset future time period with the current moment as a starting point:
step A: aiming at a target area, acquiring air humidity data respectively corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area based on a preset sampling interval and preset grid division; and then eliminating abnormal air humidity data, filling sequence vacant air humidity data, and updating air humidity data corresponding to each sampling time in a preset historical time period taking the current time as the end point in each grid area in the target area.
In the step A, abnormal air humidity data are removed through an isolated forest method, and the sequence vacancy air humidity data are filled through a front-back data filling method, namely, the time sequence vacancy values are filled. In the step A, abnormal air humidity data are removed through an isolated forest method, and the specific process is as follows in the steps A1-A3.
Step A1: and constructing a preset number of isolated trees according to air humidity data respectively corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area.
In this embodiment, as shown in fig. 2, the specific process of constructing the preset number of isolated trees is as follows:
s1: each grid region in the target region within a preset history time period ending at the current timeRandom selection in data set composed of air humidity data corresponding to sampling time
Taking the sample data points as sub-samples, and placing the sub-samples into the root node of an isolated tree;
s2: randomly generating a data cutting point P in the current node data range, wherein the cutting point is generated between the maximum value and the minimum value of the specified dimension in the current node data; in this embodiment, the designated dimension is air humidity;
s3: generating a hyperplane at a cut point p, dividing the current node data space into two subspaces, taking the cut point p as a demarcation point, wherein the left branch of the current node is smaller than the left branch of the current node, and the right branch of the current node is larger than or equal to the right branch of the current node;
s4: repeating the steps S2 and S3 on the split left and right branch nodes, continuously constructing new leaf nodes until the leaf nodes have only one data and cannot be split any more or the tree reaches a preset height, and generating an isolated tree;
s5: repeating the steps S1, S2, S3 and S4 to generate a preset number t of isolated trees.
In this embodiment, the selection of the sample for constructing the isolated tree is not replaced until all the samples are selected, i.e. the construction of the isolated tree is completed, and t isolated trees are generated.
Based on the completion of the construction of the isolated tree, executing the steps A2-A3 to define an evaluation system of the isolated forest algorithm on the abnormal data,
step A2: based on the constructed preset number of isolated trees, taking air humidity data corresponding to one sampling moment of one grid area as one sample, wherein the cutting process is completely random, and after t isolated trees are obtained, the generated isolated trees can be used for evaluating test data, namely, the anomaly score s is calculated. For each sample
Data of->
The anomaly score corresponding to each sample was obtained by the following formula:
wherein ,
；/>
in the formula ,
representation->
Sample ∈ in the individual samples>
At the corresponding anomaly score, ++>
For sample->
From root node to sample in an orphan tree->
Path length of the leaf node where +.>
For sample->
From root node to sample in all orphaned trees ≡ ->
Path Length expected value of the located leaf node, < >>
For use->
Constructing an average path length of an isolated tree by using the samples;
For sample->
From the root node of the tree to sample->
The number of edges experienced during the leaf node;
Representation and sample->
The same number of samples at a leaf node, +.>
Representation +.>
Constructing an average path length of an isolated tree by the strip sample;
For harmonizing the number, add>
，
For Euler constant, get +.>
=0.5772156649。
Step A3: and removing samples which do not meet the preset abnormal score based on the abnormal score corresponding to each sample.
Calculating an anomaly score for each sample point according to the anomaly score calculation formula,
the value range is [0,1 ]]When the observed score approaches 1, the path length is very small and the data points are easily isolated, indicating a high likelihood of data anomalies. When the observed value is less than 0.5, the path length becomes larger, and then we get a normal data point. When all the data anomaly scores s are around 0.5, it can be considered that the data as a whole is not significantly anomalous. Counting the abnormal scores of each historical data point, tightening or loosening abnormal value eliminating conditions by setting different thresholds, eliminating abnormal data according to expected effects, and then averaging and filling the latest front and back normal data of the abnormal data, wherein in the embodiment, the abnormal score +_is selected>
Normal data and abnormal data are the rest.
And (B) step (B): and carrying out normalization processing on air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid region in the target region, and then carrying out space-time matching to obtain an air humidity data sequence corresponding to each grid region.
In the step B, for the air humidity data corresponding to each sampling time in the preset historical time period with the current time as the end point in each grid area in the target area, specifically, the following steps are executed to perform normalization processing:
in the formula ,
for normalized air humidity data, +.>
For raw air humidity data, +.>
Maximum value in air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point for each grid area>
And the minimum value in the air humidity data corresponding to each sampling time in the preset historical time period taking the current time as the end point is used for each grid area. />
Step C: based on the air humidity data sequences respectively corresponding to the grid areas, an air humidity prediction model is used, wherein the air humidity data sequences respectively corresponding to the grid areas in a preset historical time period taking the current moment as an end point are used as input, the air humidity data sequences respectively corresponding to the grid areas in a preset future time period taking the current moment as a starting point are used as output, and the predicted air humidity data sequences respectively corresponding to the grid areas are obtained.
In the step C, aiming at the air humidity prediction model, optimizing Gauss-Newton algorithm through ant colony algorithm in the training process, optimizing preset parameters to be optimized of the air humidity prediction model to obtain optimal parameters to be optimized, further training based on the optimal parameters to be optimized to construct the air humidity prediction model, wherein the air humidity prediction model is obtained through training RBF neural network, and the preset parameters to be optimized comprise connection weights in the network. The air humidity prediction model is established by optimizing Gauss-Newton algorithm through ant colony algorithm and further optimizing RBF neural network.
As shown in fig. 4, the RBF neural network of the present invention converts an input space into an hidden space non-linearly, and converts the hidden space into an output space linearly. Constructing an RBF neural network using formula (1), whereinYRepresenting the output of the RBF neural network,
for hiding the number of layers +.>
Is the network ofiConnection weights of hidden layer nodes and output nodes, all connection weights in the network ∈>
Component vector->
，
Is the network ofiCenter point of activation function of each hidden layer node, center point of all activation functions in network +.>
Component vector->
，
Is an input data vector for a network, wherein +.>
，
Coefficients of a linear polynomial, +.>
As a hidden layerThe activation function is represented by formula (2), wherein +.>
Is the network ofiVariance of gaussian function of each hidden layer node.
The invention optimizes Gauss-Newton algorithm by ant colony algorithm to further optimize weight of RBF neural network
. The optimization objective function of the invention is as follows:
；
wherein ,
namely, the connection weight +.>
Vectors of composition, also->
Dimension to optimize variable->
Is about->
Is a multiple function of->
Is a->
The dimension vector maps to a non-linear function of the scalar. For variables->
Optimizing, i.e. finding a set of suitable +.>
So that the objective function is optimized->
Minimum. We can give a set of initial values +.>
Then use +.>
Local properties around the initial value, i.e. around the initial value +.>
How to change can be made +.>
Smaller and then the change increment of the initial value is obtained +.>
By iteration, when->
When it is small enough, we find the optimal network weight +.>
。
The specific optimization process is as follows:
first, initializing the weight of RBF neural network
And set training error +.>
So that there are:
in the formula ,
representing model training errors; the following optimizing iterative process is provided, namely, the specific process of optimizing Gauss-Newton algorithm and further optimizing RBF neural network by ant colony algorithm is as follows:
step C1: parameter vector to be optimized based on preset parameter composition to be optimized
In combination with the objective function of the air humidity prediction model, given an initial value +.>
Delta->
The following steps are iteratively performed:
step C1.1: based on the parameter vector to be optimized
Combining with an objective function of an air humidity prediction model to obtain a jacobian matrix corresponding to a parameter vector to be optimized in the current iteration>
And air humidity prediction model error->
，
Represent the firstnCorresponding parameter vectors to be optimized in the secondary iteration;
step C1.2: jacobian matrix corresponding to parameter vector to be optimized in current iteration
And air humidity prediction model error->
Obtaining the increment corresponding to the parameter vector to be optimized in the current iteration>
；
In this embodiment, the expansion optimization objective function is:
wherein ,
is->
About->
Is at +.>
The value of the place>
Known as gradient or jacobian. In this formula, only->
Is a variable->
Is a defined vector,/->
And
are all determined, so this is a rule +.>
As a quadratic function of the variables. And quadratic termIs positive, the function has a minimum. The condition for taking the minimum value is that the formula is for +.>
Is equal to zero, i.e.:
and (3) solving to obtain:
and (3) recording:
the delta equation for gauss newton is:
the method comprises the steps of carrying out a first treatment on the surface of the Then->
。
Step C1.3: aiming at increment corresponding to parameter vector to be optimized in current iteration
Judging->
Whether the data of (a) is smaller than the preset change increment +.>
If->
Wherein each data is smaller than the increment of change +.>
Stopping iteration, and carrying out parameter vector to be optimized in the current iteration>
As optimal parameters to be optimized; if->
The presence data is not less than delta change +.>
Obtaining the optimal increment direction corresponding to the parameter vector to be optimized through the steps a-d>
Further based on the optimal increment direction command
Returning to the step C1.1; in this embodiment, delta +.>
。
Step a: constructing a directed graph based on a preset number of ants; each node in the directed graph is connected with a preset number of edges representing the increment direction, and the number of the edges is the same as the number of ants;
step b: based on the directed graph, each ant performs direction selection based on the corresponding tabu table and the pheromone of each side representing the increment direction; the first element of the tabu list is the initial placement direction of ants, which is the initial direction of ants
The corresponding point is the center of circle, each ant is placed in the direction of the center of circle;
step c: updating the tabu list, respectively judging whether the tabu list corresponding to each ant contains all the increment directions, if so, ending the current iteration, and executing the step d; if the tabu table corresponding to the ants does not contain all the increments, returning the ants which do not contain all the increments to the executing step b;
step d: the iteration number is added by one, whether the current iteration number reaches the maximum iteration number is judged, if the current iteration number reaches the preset maximum iteration number, the iteration is ended, and the optimal increment direction is obtained based on the current pheromone of each side representing the increment direction; and c, if the maximum iteration number is not reached, updating the pheromone of each side in the directed graph, and returning to the step b.
In this embodiment, an ant colony algorithm is used to find the optimal incremental direction
. Firstly, initializing parameters, enabling an initial time t=0 and an initial cycle number +.>
Setting the maximum number of loops +.>
In this embodiment, the number of ants is 20, because after the number of ants is increased, the effect of positive feedback of information becomes insignificant, and the convergence speed is slow although the randomness of the search is enhanced, which wastes computing resources; on the contrary, the number of ants is small, the randomness of the search is weakened, and the convergence speed is accelerated, but the global performance of the algorithm is reduced, the stability of the algorithm is poor, and the phenomenon of early stagnation is easy to occur. After selecting the number of ants, m=20 ants are added to the current +.>
The point is the circle center, and an ant is put in the direction of every 18 degrees increment, so that every edge on the directed graph is [ ]i,l) Is>
, wherein cRepresents a preset constant and the initial moment +.>
. The first element of the taboo table of each ant is set to be it +.>
The point is the direction at the center of the circle. At this time, the amount of pheromones on each path is equal, and then each ant inspires according to the amount of residual pheromones on the pathThe formula information selects one direction independently, at time t, antsvFrom direction ofiTransfer to directionlProbability of->
：/>
wherein ,
representing antsvThe next step allows the selected direction set, and the taboo table records antsvThe current direction of travel. The tabu table pointer is modified, i.e. after selection, the ant is moved to a new element and the element is moved to the individual tabu table of the ant. When (when)rWhen all directions are added into the tabu list, i.e. all directions are added into the tabu list, antsvOne round of trip is completed;
Is an index weight used for adjusting the importance of pheromone and distance. In the formula->
Is a heuristic factor representing the direction of antsiTransfer to directionlTo a desired extent. In the ant colony algorithm, < >>
In general direction of orientationiStep length and direction of (2)lReciprocal of the distance between the steps. After all ants complete one round trip, the pheromone on each path is updated according to the following formula:
wherein ,
indicating the evaporation coefficient of the pheromone on the path, +.>
A persistence coefficient representing a pheromone;
Representing the middle edge of the iterationiTo the point oflIncrement of upper pheromone,/->
Represent the firstvOnly ants stay on the side in this iterationiTo the point oflThe amount of pheromone. If ants arevWithout passing by edgeiTo the point oflThen->
The value of (2) is zero.
Expressed as:
wherein ,Qin order to set a positive constant value in advance,
represent the firstvOnly ants walk the step length of the path in the optimizing process.
The specific iteration steps of the ant colony algorithm are as follows:
(1) For each ant, selecting a direction element according to the probability calculated by the state transition probability formula (3) by each ant based on the tabu table corresponding to each antlAnd proceeding, i.e. selecting the direction with highest probability based on the probability formulalAdvancing;
(2) Modifying the tabu list pointer, namely moving ants to a new element, namely a new direction after the ants are selected, and adding the direction into the tabu list of the ants;
(3) If it isrAll are added in the directionWhen entering the taboo table, antsvOne trip is completed and the process jumps to (4); if it isrThe directions are not all added to the tabu list, and the process jumps to (1);
(4) Updating the pheromone on each path according to the formula (4) and the formula (5), namely updating the pheromone in each direction;
(5) The number of cycles is increased by one;
(6) If the end condition is satisfied, i.e. if the number of cycles
The circulation is finished, and an optimization result is output, namely, the path direction with the highest final information quantity is the optimal increment direction; otherwise, the tabu list is cleared and the process jumps to (1).
Searching for optimal increment direction according to the flow chart of FIG. 3, and iterating until reaching the preset cycle number to obtain the optimal increment direction
The best +.>
。
Step D: based on the predicted air humidity data sequences corresponding to the grid areas respectively, taking the average value of the air humidity data of the grid areas at the same sampling time as the predicted air humidity of the target area at the sampling time, and obtaining the predicted air humidity of the target area in a preset future time period taking the current time as the starting point.
The system based on the GN-RBF-based air humidity prediction method comprises a data acquisition module, a data processing module, an air humidity prediction module and an air humidity output module;
aiming at the target area, based on a preset sampling interval and preset grid division, the data acquisition module is used for acquiring air humidity data respectively corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area; removing abnormal air humidity data, filling sequence vacant air humidity data, and updating air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area;
the data processing module is used for carrying out normalization processing on air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid region in the target region, and then carrying out space-time matching to obtain an air humidity data sequence corresponding to each grid region;
the air humidity prediction module is used for obtaining a predicted air humidity data sequence corresponding to each grid region by using an air humidity prediction model taking the air humidity data sequence corresponding to each grid region in a preset historical time period taking the current moment as an end point as input and taking the air humidity data sequence corresponding to each grid region in a preset future time period taking the current moment as a starting point as output based on the air humidity data sequence corresponding to each grid region respectively;
the air humidity output module is used for obtaining the predicted air humidity of the target area in a preset future time period by taking the current moment as a starting point by taking the average value of the air humidity data of each grid area at the same sampling moment as the predicted air humidity of the target area at the sampling moment based on the predicted air humidity data sequences respectively corresponding to each grid area.
The terminal of the GN-RBF-based air humidity prediction method comprises a memory and a processor, wherein the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, so that the GN-RBF-based air humidity prediction method is executed.
The invention designs an air humidity prediction method and system based on GN-RBF, and designs a novel Gaussian Newton algorithm, and the searching process is continuously converged by utilizing ant colony optimization and adopting a positive feedback mechanism. The searching process adopts a distributed parallel computing mode, each individual in the ant colony performs computation at the same time, so that the computing capacity and the operating efficiency of the algorithm are greatly improved, the optimal increment direction is finally approximated, and the iteration speed and the iteration accuracy of the Gaussian Newton algorithm are improved.
The invention introduces a single hidden layer radial basis RBF feedforward neural network, the RBF network can approach any nonlinear function, can process the regularity that is difficult to analyze in a system, and compared with the traditional network, the RBF network has good generalization capability, and has the problems of optimally approaching and overcoming local minima. The processed air humidity data is directly used as input, so that training time is saved, and training accuracy is improved.
The GN-RBF network algorithm provided by the invention is a feedforward network with excellent performance, simple framework, obvious global convergence capability and strong generalization capability and universality, can obviously reduce learning time of network learning, obtains more excellent variable weight, and further can ensure the requirement of more rapid prediction of air humidity. The algorithm designed herein can be applied not only to the problem of air humidity prediction, but also embedded in other algorithms. Compared with the traditional air humidity prediction model, the novel Gaussian Newton algorithm designed in the method accelerates the convergence speed of the algorithm and the accuracy of prediction, provides a novel method and thinking for the air humidity prediction problem, and further expands the application depth and breadth of the RBF network.
Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that the foregoing embodiments may be modified or equivalents substituted for some of the features thereof. All equivalent structures made by the content of the specification and the drawings of the invention are directly or indirectly applied to other related technical fields, and are also within the scope of the invention.
Claims (9)
1. An air humidity prediction method based on GN-RBF is characterized in that: for the target area, performing the following steps to obtain the predicted air humidity of the target area in a preset future time period taking the current moment as a starting point:
step A: aiming at a target area, acquiring air humidity data respectively corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area based on a preset sampling interval and preset grid division; removing abnormal air humidity data, filling sequence vacant air humidity data, and updating air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area;
and (B) step (B): carrying out normalization processing on air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid region in the target region, and then carrying out space-time matching to obtain an air humidity data sequence corresponding to each grid region;
step C: based on the air humidity data sequences respectively corresponding to the grid areas, an air humidity prediction model is used, wherein the air humidity data sequences respectively corresponding to the grid areas in a preset historical time period taking the current moment as an end point are used as input, the air humidity data sequences respectively corresponding to the grid areas in a preset future time period taking the current moment as a starting point are used as output, and the predicted air humidity data sequences respectively corresponding to the grid areas are obtained;
step D: based on the predicted air humidity data sequences corresponding to the grid areas respectively, taking the average value of the air humidity data of the grid areas at the same sampling time as the predicted air humidity of the target area at the sampling time, and obtaining the predicted air humidity of the target area in a preset future time period taking the current time as the starting point.
2. The GN-RBF-based air humidity prediction method as recited in claim 1, wherein: in the step C, aiming at the air humidity prediction model, optimizing the Gauss-Newton algorithm through the ant colony algorithm in the training process, optimizing preset parameters to be optimized of the air humidity prediction model to obtain optimal parameters to be optimized, and further training and constructing the air humidity prediction model based on the optimal parameters to be optimized, wherein the specific process is as follows:
step C1: parameter vector to be optimized based on preset parameter composition to be optimized
In combination with the objective function of the air humidity prediction model, the following steps are iteratively executed:
step C1.1: based on the parameter vector to be optimized
Combining with an objective function of an air humidity prediction model to obtain a jacobian matrix corresponding to a parameter vector to be optimized in the current iteration>
And air humidity prediction model error->
，
Represent the firstnCorresponding parameter vectors to be optimized in the secondary iteration;
step C1.2: jacobian matrix corresponding to parameter vector to be optimized in current iteration
And air humidity prediction model error->
Obtaining the increment corresponding to the parameter vector to be optimized in the current iteration>
；
Step C1.3: aiming at increment corresponding to parameter vector to be optimized in current iteration
Judging->
Whether the data of (a) is smaller than the preset change increment +.>
If->
Wherein each data is smaller than the increment of change +.>
Stopping iteration, and carrying out parameter vector to be optimized in the current iteration>
As optimal parameters to be optimized; if->
The presence data is not less than delta change +.>
B, obtaining the optimal increment direction corresponding to the parameter vector to be optimized through the steps a-d, and further enabling ++based on the optimal increment direction>
Returning to the step C1.1;
step a: constructing a directed graph based on a preset number of ants; each node in the directed graph is connected with a preset number of edges representing the increment direction, and the number of the edges is the same as the number of ants;
step b: based on the directed graph, each ant performs direction selection based on the corresponding tabu table and the pheromone of each side representing the increment direction; the first element of the tabu list is the initial placement direction of ants, which is the initial direction of ants
The corresponding point is the center of circle, each ant is placed in the direction of the center of circle;
step c: updating the tabu list, respectively judging whether the tabu list corresponding to each ant contains all the increment directions, if so, ending the current iteration, and executing the step d; if the tabu table corresponding to the ants does not contain all the increments, returning the ants which do not contain all the increments to the executing step b;
step d: the iteration number is added by one, whether the current iteration number reaches the maximum iteration number is judged, if the current iteration number reaches the preset maximum iteration number, the iteration is ended, and the optimal increment direction is obtained based on the current pheromone of each side representing the increment direction; and c, if the maximum iteration number is not reached, updating the pheromone of each side in the directed graph, and returning to the step b.
3. The GN-RBF-based air humidity prediction method as recited in claim 1, wherein: the air humidity prediction model is obtained by training an RBF neural network.
4. The GN-RBF-based air humidity prediction method as claimed in claim 2, wherein: the preset parameters to be optimized comprise the connection weights in the network.
5. The GN-RBF-based air humidity prediction method as recited in claim 1, wherein: in the step A, abnormal air humidity data are removed through an isolated forest method, and a front-back data filling method is adopted to fill in sequence vacant air humidity data.
6. The GN-RBF-based air humidity prediction method as claimed in claim 5, wherein: in the step A, abnormal air humidity data is removed by an isolated forest method, and the specific process is as follows:
step A1: constructing a preset number of isolated trees aiming at air humidity data respectively corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area;
step A2: based on the constructed preset quantity of isolated trees, taking air humidity data corresponding to one sampling moment of one grid area as one sample, and obtaining anomaly scores corresponding to the samples respectively through the following formulas:
wherein ,
；
in the formula ,
representation->
Sample ∈ in the individual samples>
At the corresponding anomaly score, ++>
For sample->
From root node to sample in an orphan tree->
Path length of the leaf node where +.>
For sample->
From root node to sample in all orphaned trees ≡ ->
Path Length expected value of the located leaf node, < >>
For use->
Constructing an average path length of an isolated tree by using the samples;
For sample->
From the root node of the tree to sample->
The number of edges experienced during the leaf node;
Representation and sample->
The same number of samples at a leaf node, +.>
Representation +.>
Constructing an average path length of an isolated tree by the strip sample;
For harmonizing the number, add>
，
Is Euler constant;
step A3: and removing samples which do not meet the preset abnormal score based on the abnormal score corresponding to each sample.
7. The GN-RBF-based air humidity prediction method as recited in claim 1, wherein: in the step B, for the air humidity data corresponding to each sampling time in the preset historical time period with the current time as the end point in each grid area in the target area, specifically, the following steps are executed to perform normalization processing:
in the formula ,
for normalized air humidity data, +.>
For raw air humidity data, +.>
Maximum value in air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point for each grid area>
And the minimum value in the air humidity data corresponding to each sampling time in the preset historical time period taking the current time as the end point is used for each grid area.
8. A system based on the GN-RBF based air humidity prediction method as recited in any of claims 1 to 7, characterized by: the system comprises a data acquisition module, a data processing module, an air humidity prediction module and an air humidity output module;
aiming at the target area, based on a preset sampling interval and preset grid division, the data acquisition module is used for acquiring air humidity data respectively corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area; removing abnormal air humidity data, filling sequence vacant air humidity data, and updating air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid area in the target area;
the data processing module is used for carrying out normalization processing on air humidity data corresponding to each sampling time in a preset historical time period taking the current time as an end point in each grid region in the target region, and then carrying out space-time matching to obtain an air humidity data sequence corresponding to each grid region;
the air humidity prediction module is used for obtaining a predicted air humidity data sequence corresponding to each grid region by using an air humidity prediction model taking the air humidity data sequence corresponding to each grid region in a preset historical time period taking the current moment as an end point as input and taking the air humidity data sequence corresponding to each grid region in a preset future time period taking the current moment as a starting point as output based on the air humidity data sequence corresponding to each grid region respectively;
the air humidity output module is used for obtaining the predicted air humidity of the target area in a preset future time period by taking the current moment as a starting point by taking the average value of the air humidity data of each grid area at the same sampling moment as the predicted air humidity of the target area at the sampling moment based on the predicted air humidity data sequences respectively corresponding to each grid area.
9. A terminal of an air humidity prediction method based on GN-RBF, characterized in that: the air humidity prediction method based on the GN-RBF comprises a memory and a processor, wherein the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, so that the air humidity prediction method based on the GN-RBF is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310287141.5A CN115994629B (en) | 2023-03-23 | 2023-03-23 | GN-RBF-based air humidity prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310287141.5A CN115994629B (en) | 2023-03-23 | 2023-03-23 | GN-RBF-based air humidity prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115994629A true CN115994629A (en) | 2023-04-21 |
| CN115994629B CN115994629B (en) | 2023-06-20 |
Family
ID=85995302
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310287141.5A Active CN115994629B (en) | 2023-03-23 | 2023-03-23 | GN-RBF-based air humidity prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115994629B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117313555A (en) * | 2023-11-28 | 2023-12-29 | 南京信息工程大学 | Distributed storage-based adaptive OATS-AJSA improved GRU humidity prediction method |
| CN117436005A (en) * | 2023-12-21 | 2024-01-23 | 山东汇力环保科技有限公司 | Abnormal data processing method in automatic ambient air monitoring process |
Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080053240A1 (en) * | 2006-08-28 | 2008-03-06 | Invensys Systems, Inc. | Wet Gas Measurement |
| RU2009111287A (en) * | 2006-08-28 | 2010-10-10 | Инвенсис Системз, Инк. (Us) | WATER GAS MEASUREMENT |
| CN103426026A (en) * | 2013-09-10 | 2013-12-04 | 信阳师范学院 | Method for forecasting hybrid neural network and recognizing scenic spot meteorological elements |
| CN103927460A (en) * | 2014-05-05 | 2014-07-16 | 重庆大学 | Wind power plant short-term wind speed prediction method based on RBF |
| WO2015066564A1 (en) * | 2013-10-31 | 2015-05-07 | Cancer Prevention And Cure, Ltd. | Methods of identification and diagnosis of lung diseases using classification systems and kits thereof |
| CN105260800A (en) * | 2015-10-26 | 2016-01-20 | 国网浙江省电力公司电力科学研究院 | Photovoltaic module temperature prediction method and device |
| CN109506655A (en) * | 2018-10-19 | 2019-03-22 | 哈尔滨工业大学（威海） | Improvement ant colony path planning algorithm based on non-homogeneous modeling |
| CN109670628A (en) * | 2018-11-09 | 2019-04-23 | 昆明理工大学 | A kind of micro-grid load prediction technique of the neural network based on LSSVM |
| CN110399928A (en) * | 2019-07-29 | 2019-11-01 | 集美大学 | Voltage of solid oxide fuel cell prediction technique, terminal device and storage medium |
| CN111144663A (en) * | 2019-12-31 | 2020-05-12 | 上海电力大学 | Ultra-short-term wind power prediction method for offshore wind farm considering output fluctuation process |
| CN111260126A (en) * | 2020-01-13 | 2020-06-09 | 燕山大学 | Short-term photovoltaic power generation prediction method considering correlation degree of weather and meteorological factors |
| CN111832814A (en) * | 2020-07-01 | 2020-10-27 | 北京工商大学 | Air pollutant concentration prediction method based on graph attention machine mechanism |
| CN111967696A (en) * | 2020-10-23 | 2020-11-20 | 北京国新智电新能源科技有限责任公司 | Neural network-based electric vehicle charging demand prediction method, system and device |
| CN112634018A (en) * | 2020-12-23 | 2021-04-09 | 百维金科(上海)信息科技有限公司 | Overdue monitoring method for optimizing recurrent neural network based on ant colony algorithm |
| CN112966853A (en) * | 2021-02-04 | 2021-06-15 | 南通大学 | Urban road network short-term traffic flow prediction method based on space-time residual error mixed model |
| CN113919231A (en) * | 2021-10-25 | 2022-01-11 | 北京航天创智科技有限公司 | PM2.5 concentration space-time change prediction method and system based on space-time diagram neural network |
| CN114154702A (en) * | 2021-11-25 | 2022-03-08 | 深圳中兴网信科技有限公司 | Pollutant concentration prediction method and device based on multi-granularity graph space-time neural network |
| CN114881323A (en) * | 2022-05-07 | 2022-08-09 | 沈阳工业大学 | Foundation pit dewatering area underground water level prediction and updating method based on deep neural network |
-
2023
- 2023-03-23 CN CN202310287141.5A patent/CN115994629B/en active Active
Patent Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080053240A1 (en) * | 2006-08-28 | 2008-03-06 | Invensys Systems, Inc. | Wet Gas Measurement |
| RU2009111287A (en) * | 2006-08-28 | 2010-10-10 | Инвенсис Системз, Инк. (Us) | WATER GAS MEASUREMENT |
| CN103426026A (en) * | 2013-09-10 | 2013-12-04 | 信阳师范学院 | Method for forecasting hybrid neural network and recognizing scenic spot meteorological elements |
| WO2015066564A1 (en) * | 2013-10-31 | 2015-05-07 | Cancer Prevention And Cure, Ltd. | Methods of identification and diagnosis of lung diseases using classification systems and kits thereof |
| CN103927460A (en) * | 2014-05-05 | 2014-07-16 | 重庆大学 | Wind power plant short-term wind speed prediction method based on RBF |
| CN105260800A (en) * | 2015-10-26 | 2016-01-20 | 国网浙江省电力公司电力科学研究院 | Photovoltaic module temperature prediction method and device |
| CN109506655A (en) * | 2018-10-19 | 2019-03-22 | 哈尔滨工业大学（威海） | Improvement ant colony path planning algorithm based on non-homogeneous modeling |
| CN109670628A (en) * | 2018-11-09 | 2019-04-23 | 昆明理工大学 | A kind of micro-grid load prediction technique of the neural network based on LSSVM |
| CN110399928A (en) * | 2019-07-29 | 2019-11-01 | 集美大学 | Voltage of solid oxide fuel cell prediction technique, terminal device and storage medium |
| CN111144663A (en) * | 2019-12-31 | 2020-05-12 | 上海电力大学 | Ultra-short-term wind power prediction method for offshore wind farm considering output fluctuation process |
| CN111260126A (en) * | 2020-01-13 | 2020-06-09 | 燕山大学 | Short-term photovoltaic power generation prediction method considering correlation degree of weather and meteorological factors |
| CN111832814A (en) * | 2020-07-01 | 2020-10-27 | 北京工商大学 | Air pollutant concentration prediction method based on graph attention machine mechanism |
| CN111967696A (en) * | 2020-10-23 | 2020-11-20 | 北京国新智电新能源科技有限责任公司 | Neural network-based electric vehicle charging demand prediction method, system and device |
| CN112634018A (en) * | 2020-12-23 | 2021-04-09 | 百维金科(上海)信息科技有限公司 | Overdue monitoring method for optimizing recurrent neural network based on ant colony algorithm |
| CN112966853A (en) * | 2021-02-04 | 2021-06-15 | 南通大学 | Urban road network short-term traffic flow prediction method based on space-time residual error mixed model |
| CN113919231A (en) * | 2021-10-25 | 2022-01-11 | 北京航天创智科技有限公司 | PM2.5 concentration space-time change prediction method and system based on space-time diagram neural network |
| CN114154702A (en) * | 2021-11-25 | 2022-03-08 | 深圳中兴网信科技有限公司 | Pollutant concentration prediction method and device based on multi-granularity graph space-time neural network |
| CN114881323A (en) * | 2022-05-07 | 2022-08-09 | 沈阳工业大学 | Foundation pit dewatering area underground water level prediction and updating method based on deep neural network |
Non-Patent Citations (3)
| Title |
|---|
| 刘雅云: "基于机器学习的降水数据分析算法的研究", 《中国优秀硕士学位论文全文数据库基础科学辑》, no. 01, pages 009 - 187 * |
| 张鹏: "基于神经网络的探空湿度太阳辐射误差预测及查询网站设计", 《基于神经网络的探空湿度太阳辐射误差预测及查询网站设计》, no. 02, pages 009 - 98 * |
| 郑林涛: "基于机器学习方法的珠三角地区商场空调全年负荷预测研究", 《中国博士学位论文全文数据库工程科技Ⅱ辑》, no. 06, pages 038 - 47 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117313555A (en) * | 2023-11-28 | 2023-12-29 | 南京信息工程大学 | Distributed storage-based adaptive OATS-AJSA improved GRU humidity prediction method |
| CN117313555B (en) * | 2023-11-28 | 2024-03-08 | 南京信息工程大学 | GRU humidity prediction method based on self-adaptive OATS-AJSA |
| CN117436005A (en) * | 2023-12-21 | 2024-01-23 | 山东汇力环保科技有限公司 | Abnormal data processing method in automatic ambient air monitoring process |
| CN117436005B (en) * | 2023-12-21 | 2024-03-15 | 山东汇力环保科技有限公司 | Abnormal data processing method in automatic ambient air monitoring process |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115994629B (en) | 2023-06-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115994629B (en) | GN-RBF-based air humidity prediction method and system | |
| CN111027732B (en) | Method and system for generating multi-wind power plant output scene | |
| CN111144663B (en) | Ultra-short-term wind power prediction method for offshore wind farm considering output fluctuation process | |
| CN111611692B (en) | Equal-reliability-based design flood calculation method and system under climate change situation | |
| CN110969290A (en) | Runoff probability prediction method and system based on deep learning | |
| CN113326969A (en) | Short-term wind speed prediction method and system based on improved whale algorithm for optimizing ELM | |
| CN112733997B (en) | Hydrological time series prediction optimization method based on WOA-LSTM-MC | |
| CN113469426A (en) | Photovoltaic output power prediction method and system based on improved BP neural network | |
| CN114169434A (en) | Load prediction method | |
| CN114139783A (en) | Wind power short-term power prediction method and device based on nonlinear weighted combination | |
| CN113221447A (en) | Soil humidity prediction method for optimizing BP neural network based on improved genetic algorithm | |
| CN114897204A (en) | Method and device for predicting short-term wind speed of offshore wind farm | |
| CN118332496A (en) | Winter abnormal snowfall prediction method based on meteorological data and statistical model | |
| CN110851959A (en) | Wind speed interval prediction method integrating deep learning and quantile regression | |
| CN114547964A (en) | Improved HHO (Hilbert-Huang-Quadrature-order) optimization DELM (Del-enhanced dead mass) based debutanizer soft measurement modeling method | |
| CN116544931B (en) | Power load distribution prediction method based on integrated fragment transformation and time convolution network | |
| CN112801364A (en) | Reference crop evapotranspiration amount prediction method based on MIV and improved bat algorithm | |
| CN117595238A (en) | Training method of wind power prediction model and wind power prediction method | |
| CN110322050B (en) | Wind energy resource data compensation method | |
| CN116562455A (en) | Air temperature forecast data processing method and device of wind driven generator and computer equipment | |
| CN116054144A (en) | Distribution network reconstruction method, system and storage medium for distributed photovoltaic access | |
| CN113052379A (en) | Tomato water demand prediction method based on BAS-RBF | |
| CN107196328A (en) | A kind of power distribution network light stores up association system access point vulnerability index forecasting method | |
| Jain et al. | Simulation Model Calibration with Dynamic Stratification and Adaptive Sampling | |
| CN117057255B (en) | Pre-training model acquisition method for online synchronization of digital twin model |
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 |