CN110598860A - Multi-station online wave cycle data prediction diagnosis method - Google Patents

Multi-station online wave cycle data prediction diagnosis method Download PDF

Info

Publication number
CN110598860A
CN110598860A CN201910719441.XA CN201910719441A CN110598860A CN 110598860 A CN110598860 A CN 110598860A CN 201910719441 A CN201910719441 A CN 201910719441A CN 110598860 A CN110598860 A CN 110598860A
Authority
CN
China
Prior art keywords
model
space
time
training
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910719441.XA
Other languages
Chinese (zh)
Other versions
CN110598860B (en
Inventor
李文庆
宋苗苗
陈世哲
王文彦
付晓
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Oceanographic Instrumentation Research Institute Shandong Academy of Sciences
Original Assignee
Oceanographic Instrumentation Research Institute Shandong Academy of Sciences
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Oceanographic Instrumentation Research Institute Shandong Academy of Sciences filed Critical Oceanographic Instrumentation Research Institute Shandong Academy of Sciences
Priority to CN201910719441.XA priority Critical patent/CN110598860B/en
Publication of CN110598860A publication Critical patent/CN110598860A/en
Application granted granted Critical
Publication of CN110598860B publication Critical patent/CN110598860B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C13/00—Surveying specially adapted to open water, e.g. sea, lake, river or canal
    • G01C13/002—Measuring the movement of open water
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Hydrology & Water Resources (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Testing Or Calibration Of Command Recording Devices (AREA)

Abstract

The invention discloses a multi-station online wave cycle data prediction diagnosis method, which comprises a space-time sample test training stage and a space-time model dynamic calculation adjusting stage, wherein the space-time sample test training stage comprises two processes of time model test training and space model test training; the dynamic calculation and adjustment stage of the space-time model comprises four processes of mixed calculation diagnosis of a time model and a space model, weight adjustment of the space-time model, dynamic adjustment of the time model and dynamic adjustment of the space model; the time model adopts an RBF neural network model, the space model adopts a linear neural network model, and the method disclosed by the invention increases the space mutual analysis among a plurality of stations on the basis of the self-analysis of the single-station time sequence data, effectively combines the single-station time analysis and the multi-station space analysis, and improves the precision and the accuracy of the predictive diagnosis.

Description

Multi-station online wave cycle data prediction diagnosis method
Technical Field
The invention relates to the field of ocean wave monitoring, in particular to a multi-station online wave period data prediction and diagnosis method.
Background
At present, aiming at online wave cycle data monitored by a marine wave monitoring single station, only simple sending and receiving verification in a sea-land communication process can be carried out to judge whether a communication link has data errors. The existing online data receiving software can only simply judge the data range (such as more than or equal to 0 and less than 30 seconds), and cannot perform substantial diagnosis on specific data. The method aims at the defects of the prediction and diagnosis method of whether the data measured by the marine wave meter is abnormal or not, and the abnormal and vacant data of the marine wave meter cannot be timely and effectively found and supplemented, and the targeted judgment and identification are needed manually.
The existing data prediction diagnosis technology mainly comprises an autoregressive moving average model, a neural network model and the like, and the neural network model for time series prediction mainly comprises a linear neural network, a BP neural network, an RBF neural network and the like.
Most of the existing autoregressive moving average models and neural network models are used for processing existing data, and the models are often static and unchangeable or complex and time-consuming, so that the model preparation rate is low or the running speed is low, real-time updating of data in the online monitoring process is not considered, abnormal data need to be timely and effectively judged in the online ocean monitoring process, and the data in the absence of time need to be timely predicted and supplemented.
Meanwhile, marine monitoring data are related among a plurality of marine monitoring stations with similar space geographic distances, however, the relevance of the data among the plurality of stations is not considered in the prior art, so that the relationship among the stations is not strong, and the data analysis is not complete and accurate enough.
Disclosure of Invention
In order to solve the technical problems, the invention provides a method for predicting and diagnosing multi-station online wave cycle data, which aims to increase the spatial mutual analysis among a plurality of stations on the basis of the self-analysis of single-station time sequence data, effectively combine the single-station time analysis and the multi-station spatial analysis and improve the precision and the accuracy of predictive diagnosis.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a multi-station online wave cycle data prediction diagnosis method comprises a space-time sample test training stage and a space-time model dynamic calculation adjusting stage, wherein the space-time sample test training stage comprises a time model test training stage and a space model test training stage; the dynamic calculation and adjustment stage of the space-time model comprises four processes of mixed calculation diagnosis of a time model and a space model, weight adjustment of the space-time model, dynamic adjustment of the time model and dynamic adjustment of the space model; the time model adopts an RBF neural network model, and the space model adopts a linear neural network model.
In the above scheme, the method for predicting diagnosis specifically includes the following steps:
(1) training by using time samples to obtain an RBF neural network initial time model, and training by using space samples to obtain a linear neural network initial space model;
(2) loading a trained RBF neural network initial time model and a trained linear neural network initial space model, performing actual prediction calculation one by one according to the latest time sequence input and space sequence input of real-time ocean waves obtained by a wave sensor to respectively obtain a time prediction result and a space prediction result, and obtaining a space-time comprehensive prediction result according to the time prediction result and the space prediction result by weighting calculation;
(3) judging whether the data is abnormal or not by using the obtained space-time comprehensive prediction result and the actual measurement result at the next moment, and supplementing a predicted value as a current value if the data at the current moment is lack of measurement or the data is abnormal; and taking the measured value judged to be the effective value as the latest data to form a new time sample and a new space sample, and respectively using the new time sample and the new space sample for retraining and adjusting the RBF neural network dynamic time model and the retraining and adjusting the linear neural network dynamic space model, and performing dynamic prediction diagnosis on the data in the continuous training and adjusting process of the model.
In a further technical scheme, in the step (1), the training of the initial time model of the RBF neural network includes the following processes:
selecting wave period data of a single station as a processing training sample set, and setting training parameters of the RBF neural network time model, wherein the training parameters comprise the number of initial input layers, the number of hidden layer nodes and iteration precision; the testability training determines the number of input layers, the number of training samples and the number of hidden layer central nodes; and training to obtain an initial time model of the RBF neural network.
In a further technical solution, in the step (1), the training of the linear neural network initial space model includes the following processes:
1) selecting N space monitoring stations, and normalizing wave period data of each station to obtain a space sample set;
2) selecting 1 station P as an object for space model calculation, and using the rest N-1 station data as the input of a linear neural network;
3) performing testability training by using different numbers of training samples, and determining a spatial model Ms which enables the training precision to be highest;
4) the number of samples corresponding to Ms is Np, and Np is the number of training samples during dynamic adjustment of the spatial model.
According to the technical scheme, the multi-station online wave periodic data prediction diagnosis method provided by the invention integrates the time and space characteristics of marine monitoring data, adopts an RBF neural network model for the time sequence data of a single station, adopts a linear neural network model for the space distribution data of a plurality of stations, and performs weighted calculation on the time calculation result and the space calculation result to form a final prediction value. And using the final prediction value for real-time diagnosis of the ocean online monitoring activity data. The invention has the following beneficial effects:
1. neural networks have advantages over traditional models
The neural network has strong nonlinear mapping capability. The method can complete highly complex input and output nonlinear mapping, and has good nonlinear fitting capability, which is incomparable with the traditional prediction method based on the least square method.
The neural network fitting result is more accurate than the mathematical model fitting result. The ARMA model requires that a time sequence is stable, so that logarithm is firstly taken and then difference is made on a data set during modeling, the establishment of a modeling neural network model is complex and troublesome, and a regular approximation real model can be found through training under the condition that the neural network does not know an internal mathematical model.
2. High operation speed, high adaptability and high accuracy
The RBF neural network time model has high operation speed and the dynamic learning and adjusting process of the model is high; the linear neural network space model has the advantages that the number and the scale of the spatial data point positions determine that the input of the space model is small, the calculation speed is high, the amount of dynamic training samples is small, and the high dynamic training speed of the linear neural network space model is guaranteed.
In the actual marine monitoring data diagnosis process, new data are continuously fed back to the learning and adjustment of the RBF neural network time model and the linear neural network space model along with the arrival of the new data, so that the whole calculation model has strong adaptability, and the effectiveness and the accuracy of data diagnosis in long-term marine online monitoring activities are ensured.
3. Temporal and spatial characteristics of marine surveillance data effectively combined
The space-time analysis is combined, the problem that the deviation of a simulation calculation result is large when the distance between a single station time sequence and a sample set and a central set is large is solved, and the problem that the error of a space analysis result is generally large due to the fact that the space factor is only emphasized in the single multi-station space analysis is also solved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below.
FIG. 1 is a schematic diagram illustrating a method for diagnosing online cycle data according to multiple stations according to the present invention;
FIG. 2 is a flowchart illustrating a time model sample testing training process according to an embodiment of the present invention;
FIG. 3 is a flowchart illustrating a spatial model sample testing training process according to an embodiment of the present invention;
fig. 4 is a schematic flowchart of a method for diagnosing the multi-station online wave cycle data in a predictive manner according to an embodiment of the present invention (shown in two pages).
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
The invention provides a method for predicting and diagnosing multi-station online wave cycle data, as shown in fig. 1, the specific embodiment is as follows:
the method for predicting and diagnosing the multi-station online wave cycle data comprises a space-time sample test training stage and a space-time model dynamic calculation adjusting stage, wherein the space-time sample test training stage comprises a time model test training stage and a space model test training stage; the dynamic calculation and adjustment stage of the space-time model comprises four processes of mixed calculation diagnosis of the time model and the space model, weight adjustment of the space-time model, dynamic adjustment of the time model and dynamic adjustment of the space model; the time model adopts an RBF neural network model, and the space model adopts a linear neural network model.
One, time-space sample test training phase
And training by using the time samples to obtain an RBF neural network initial time model, and training by using the space samples to obtain a linear neural network initial space model.
1. Time model test training, as shown in fig. 2:
1) representative sample data (30 days) with continuous single-station monitoring time is selected and divided into a training sample and a testing sample, wherein the two samples are 15 days respectively.
2) The sample data is processed into a data sequence with the length of 1 hour and equal time intervals, namely 360 groups of training samples and test samples, and normalization processing is carried out (the numerical range is processed to be between-1 and 1).
3) The number of training samples is set to be 200, the iteration precision is set to be 0.005, the maximum iteration number is 500, the iteration step length is set to be 0.01, and the clustering segmentation factor is set to be 0.85 when the RBF neural network time model is trained.
4) And respectively training and testing the number of input layers of the RBF neural network time model by taking 24, 36, 48 and 60 as the number of input layers of the RBF neural network time model, obtaining the number of the input layers corresponding to the model with the best test accuracy, recording the number as Ni, and taking the number of the input layers of the model as Ni.
5) And (5) training and testing by using different training sample numbers with 10 as step length between [200 and 360], recording the training sample number corresponding to the model with the best test accuracy as Ns, and taking the training sample number as Ns.
6) And (3) training and testing by using different segmentation factors with 0.01 as a step length between [0.5 and 2], obtaining the segmentation factor corresponding to the model with the best test accuracy and marking as pp, wherein the number of the corresponding hidden layer central nodes is Nt when the value of the clustering segmentation factor is taken as pp.
7) And training to obtain an initial time model Mt of the RBF neural network by taking the number of input layers as Ni, the number of samples as Ns, the iteration precision as 0.005, the maximum iteration number as 500, the iteration step length as 0.01 and the number of nodes of the hidden layer as Nt (a clustering partition factor is pp).
2. The training phase of the spatial model test is shown in fig. 3:
1) n (N >3) ocean monitoring stations with similar spatial positions are selected, and representative sample data (3 days) with continuous monitoring time is selected for each station.
2) The sample data is processed into data sequences with the length of 1 hour and equal time intervals, namely 120 groups, and normalization processing (processing the numerical range to be between-1 and 1) is carried out.
3) Selecting 1 station position P from N ocean monitoring station positions as an object of linear neural network space model calculation of on-line wave period data, and taking wave period data of each moment of the rest N-1 station positions as N-1 inputs of a linear neural network.
4) And (4) training and testing by using different training sample numbers with 1 as a step length between [2 and 72] to obtain a model (containing various parameters) Ms with the best test accuracy, wherein the training sample number corresponding to the Ms is recorded as Np. In the dynamic adjustment calculation stage of the spatial model, the number of dynamic training samples is Np.
Second, the spatio-temporal model dynamically calculates and adjusts the phase (taking P monitoring station as an example), as shown in fig. 4:
in the dynamic calculation and adjustment stage of the space-time model, the pre-trained time model, space model and space-time model weight coefficient of the P station are used as actual calculation diagnosis. In the actual monitoring activity of the ocean waves of the P station, new wave period data are continuously obtained, time model calculation and space model calculation are continuously carried out, a prediction result is calculated according to a space-time weight coefficient, an actual measurement value and a predicted value are compared, whether the actual measurement value is an effective value or not is judged, if the actual measurement value is not measured or is judged to be an invalid value, the predicted value is used as a current effective value, and if the actual measurement value is judged to be an effective value, the actual measurement value is used for dynamic feedback adjustment of the space-time weight, the time model and the space model, so that the latest data characteristics and the change rule are continuously adapted. And during the P-station ocean wave monitoring activity, the space-time model calculation and adjustment are carried out at each monitoring moment until the P-station ocean wave monitoring activity is finished.
The whole process comprises the following steps:
1) and starting to perform space-time model dynamic calculation and adjustment when the P station has continuous Ni effective monitoring values. And loading an RBF neural network initial time model Mt and a linear neural network initial space model Ms, recording the initial time sample set of Mt as S, and then the sample number of S is the optimal time sample number Ns obtained in the training stage of the time model.
2) The number of the newly added samples of the time model is set to 0, that is, Nc is 0.
3) And finding the latest Np monitoring moments when all the N station bit data are complete and effective according to the current moment. Np measured values Ca for P-site; respectively calculating Np time predicted values Ct of the P station positions by using Mt, wherein the time predicted value at each moment t is obtained by inputting Ns continuous effective values before t (not containing t) into the Mt for calculation; respectively calculating Np space predicted values Cs of the P station positions by using Ms, wherein the space predicted value of each time t is obtained by inputting N-1 effective values of N-1 station positions (not containing the P station positions) at the time t into the Ms.
4) When all values of Cs and Ct are identical, W is taken to be 0.5; when Cs and Ct are not identical, calculating to obtain optimal W within the range of [0,1], and enabling the distance between W and Cs plus (1-W) and Ca to be minimum, wherein Cs, Ct and Ca all contain Np values, and the monitoring time of corresponding values among Cs, Ct and Ca is identical.
5) tn records the current latest data time. Inputting effective values of Ni latest moments before tn (without tn) of the P station position into the Mt to obtain a time prediction result Rt, and finally obtaining a predicted value Rp-Rt. And judging whether other N-1 stations except P have data missing at the tn moment, and turning to 7) if the data missing exists.
6) And inputting the N-1 monitoring values of the rest N-1 stations (without P stations) at the time tn into Ms to calculate a spatial prediction value Rs. And calculating a space-time predicted value Rp ═ W ═ Rs + (1-W) × Rt.
7) Checking and judging whether the P station position has wave period actual measurement data at the new time tn, if the wave period actual measurement data is lacked, using Rp as a P station position tn time effective value, and turning to 21); if the measured data is available, the measured data is marked as Ra, if abs (Rp-Ra) >0.25, the measured value is diagnosed to be invalid, and Rp is used as the effective value of the P station tn, and the operation is switched to 21).
8) And judging whether other N-1 stations except P have data missing at the tn moment, and turning to 13) if the data missing exists. Ra replaced the oldest data in Ca; rt replaces the longest time data in the Ct; rs replaces the oldest time data in Cs.
9) When all values of Cs and Ct are identical, W is taken to be 0.5; when Cs and Ct are not identical, recalculating the optimal space-time weight Wn in the range of [0,1] to minimize the distance between Wn Cs plus (1-Wn) Ct and Ca, wherein Cs, Ct and Ca all contain Np values, and the monitoring time of corresponding values among Cs, Ct and Ca is identical. W is adjusted so that W is 0.5+ Wn 0.5.
10) And taking N-1 data of N-1 station positions except P at the tn time as space model sample input, and taking Ra as space model sample output to form a latest space sample B.
11) Except for the tn time, all N stations take the historical values of Np-1 latest times. The missing time data is checked and removed from Np-1 time data sets, and B is added to form Nps spatial sample sets (Nps > -1).
12) The linear neural network spatial model is trained with Nps spatial samples, resulting in a new spatial model and updating the replacement Ms.
13) And taking Ni latest time data of the P station except tn time as time model sample input, and taking Ra as time model sample output to form latest time sample A.
14) The shortest distance DisS between the sample input of A and each sample input of S is calculated. Go to 21 if dis < Ni 0.005).
15) Adding A into a time sample set S, and enabling Nc to be Nc + 1; if Nc > -Ni 0.3, go to 18).
16) And calculating the minimum distance DisT between the sample input of the A and each central node in the T, and calculating the minimum distance DisM between each central node in the T. When DisT > DisM 0.7 and DisS > DisM 0.5, the sample input of A is used as a new center to add T, the oldest sample Sh is found from the time sample set, the center with the minimum distance to the sample input of Sh is found from T and deleted, and the number of the centers of T is kept unchanged.
17) Sh is removed from the time sample set S and a transition is made to 20).
18) The longest time sample Sh was found from S and removed.
19) Finding the longest sample S from the current S0All samples in S are taken as distance S0The distances of (D) are sorted from small to large (S)0First sample) and use L0,L1,...,LNs-1Represents the sorted samples and S0Calculating a distance difference D1＝L1-L0，D2＝L2-L1,…,DNs-1＝LNs-1-LNs-2. To D1,D2,...,DNs-1Taking Nt maximum values, and using the set of Nt sample inputs corresponding to the Nt values as a new central node set T.
20) And recalculating the time model output weight Wd according to the latest time sample set and the hidden layer center set T, and updating and adjusting the time model Mt.
21) If the ocean wave monitoring activity is over, go to 22); otherwise, waiting for the next time to arrive, and going to 5).
22) And ending the dynamic calculation diagnosis process of the space-time model of the P station.
In the above process steps, the hybrid computation diagnosis process of the temporal model and the spatial model includes steps 3) -7); the dynamic space-time weight adjusting process comprises steps 8) and 9); the multi-station space model training and adjusting process comprises steps 10) -12); the single-site time model dynamic adjustment process includes steps 13) -20).
Wherein, in steps 4) and 9), when Cs and Ct are not identical, calculating the optimal W within the range of [0,1] to minimize the distance between W and Cs plus (1-W) Ct and Ca, and the method comprises the following steps:
recording Np values of n-Np and Cs as x1, x2, … and xn in the monitoring time sequence; np numerical values of Ct are y1, y2, … and yn according to the sequence of monitoring time; the Np values of Ca are z1, z2, … and zn according to the monitoring time sequence.
Solving W with the smallest distance between W Cs + (1-W) Ct and Ca, namely solving W with the smallest distance between W Cs + Ct and CaThe smallest W.
③
④There is a minimum value for the quadratic polynomial.
TakingJudging whether the W value is [0,1]]When W is less than 0, taking W as 0; when W is more than 1, taking W as 1.
On-line ocean wave cycle data is continuously obtained at fixed time intervals of 10 minutes, 30 minutes, 60 minutes and the like, and 100% transmission efficiency is difficult to guarantee when the data is transmitted to a software processing system on land from the sea by processes of wave sensors, data packaging, data wireless transmission, data unpacking and the like.
When a multi-station space model is calculated, the model input of a certain station is the monitoring value of other stations at the same time, and when any station data is missing at the present time, the calculation of the space model can not be carried out. And establishing an RBF neural network time model for each station, diagnosing and filling abnormal and missing data, and using complete and continuous data of each station for calculation of a linear neural network space model so as to further perform mixed calculation of the time model and the space model.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (4)

1. A multi-station online wave cycle data prediction diagnosis method is characterized by comprising a space-time sample test training stage and a space-time model dynamic calculation adjusting stage, wherein the space-time sample test training stage comprises a time model test training stage and a space model test training stage; the dynamic calculation and adjustment stage of the space-time model comprises four processes of mixed calculation diagnosis of a time model and a space model, weight adjustment of the space-time model, dynamic adjustment of the time model and dynamic adjustment of the space model; the time model adopts an RBF neural network model, and the space model adopts a linear neural network model.
2. The method for predictively diagnosing multi-site online wave cycle data according to claim 1, wherein the method specifically includes the following steps:
(1) training by using time samples to obtain an RBF neural network initial time model, and training by using space samples to obtain a linear neural network initial space model;
(2) loading a trained RBF neural network initial time model and a trained linear neural network initial space model, performing actual prediction calculation one by one according to the latest time sequence input and space sequence input of real-time ocean waves obtained by a multi-station wave sensor to respectively obtain a time prediction result and a space prediction result, and obtaining a space-time comprehensive prediction result according to the time prediction result and the space prediction result by weighting calculation;
(3) judging whether the data is abnormal or not by using the obtained space-time comprehensive prediction result and the actual measurement result at the next moment, and supplementing a predicted value as a current value if the data at the current moment is lack of measurement or the data is abnormal; and taking the measured value judged to be the effective value as the latest data to form a new time sample and a new space sample, and respectively using the new time sample and the new space sample for retraining and adjusting the RBF neural network dynamic time model and the retraining and adjusting the linear neural network dynamic space model, and performing dynamic prediction diagnosis on the data in the continuous training and adjusting process of the model.
3. The method for predictively diagnosing the multi-station online wave cycle data as recited in claim 2, wherein the training of the initial time model of the RBF neural network in step (1) includes the following steps:
selecting wave period data of a single station as a processing training sample set, and setting training parameters of the RBF neural network time model, wherein the training parameters comprise the number of initial input layers, the number of hidden layer nodes and iteration precision; the testability training determines the number of input layers, the number of training samples and the number of hidden layer central nodes; and training to obtain an initial time model of the RBF neural network.
4. The method for predictively diagnosing the multi-site online wave cycle data as recited in claim 2, wherein the training of the initial spatial model of the linear neural network in step (1) comprises the following processes:
1) selecting N space monitoring stations, and normalizing wave period data of each station to obtain a space sample set;
2) selecting 1 station P as an object for space model calculation, and using the rest N-1 station data as the input of a linear neural network;
3) performing testability training by using different numbers of training samples, and determining a spatial model Ms which enables the training precision to be highest;
4) the number of samples corresponding to Ms is Np, and Np is the number of training samples during dynamic adjustment of the spatial model.
CN201910719441.XA 2019-08-06 2019-08-06 Multi-station online wave cycle data prediction diagnosis method Active CN110598860B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910719441.XA CN110598860B (en) 2019-08-06 2019-08-06 Multi-station online wave cycle data prediction diagnosis method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910719441.XA CN110598860B (en) 2019-08-06 2019-08-06 Multi-station online wave cycle data prediction diagnosis method

Publications (2)

Publication Number Publication Date
CN110598860A true CN110598860A (en) 2019-12-20
CN110598860B CN110598860B (en) 2023-02-24

Family

ID=68853565

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910719441.XA Active CN110598860B (en) 2019-08-06 2019-08-06 Multi-station online wave cycle data prediction diagnosis method

Country Status (1)

Country Link
CN (1) CN110598860B (en)

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004028616A (en) * 2002-06-21 2004-01-29 Nec Corp Marine environment measuring interval determination method and device
CN103489034A (en) * 2013-10-12 2014-01-01 山东省科学院海洋仪器仪表研究所 Method and device for predicting and diagnosing online ocean current monitoring data
CN103678883A (en) * 2013-11-21 2014-03-26 上海海洋大学 Space sampling method oriented to multisource marine environmental monitoring data
CN105787592A (en) * 2016-02-26 2016-07-20 河海大学 Wind turbine generator set ultra-short period wind power prediction method based on improved RBF network
CN106599577A (en) * 2016-12-13 2017-04-26 重庆邮电大学 ListNet learning-to-rank method combining RBM with feature selection
CN106897705A (en) * 2017-03-01 2017-06-27 上海海洋大学 A kind of oceanographic observation big data location mode based on incremental learning
CN107364548A (en) * 2017-06-01 2017-11-21 中国海洋石油总公司 Internal wave of ocean observes buoy and its typhoon protection design method, typhoon protection method
CN108036961A (en) * 2017-12-07 2018-05-15 国家海洋局南海环境监测中心 Marine environmental monitoring Multitask Data acquisition method, storage medium and electronic equipment
US20180313321A1 (en) * 2015-10-27 2018-11-01 IFP Energies Nouvelles Method for predicting a characteristic resulting from the swell on a floating system for at least two future time steps
CN108761497A (en) * 2018-03-06 2018-11-06 北京交通大学 A kind of train satellite positioning local completeness monitoring method and system
CN108898223A (en) * 2018-07-11 2018-11-27 国家海洋技术中心 A kind of ocean observation device operating status method for detecting abnormality and device
CN109947879A (en) * 2019-01-29 2019-06-28 中国海洋大学 A kind of oceanographic observation big data visual analysis method based on complex network

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004028616A (en) * 2002-06-21 2004-01-29 Nec Corp Marine environment measuring interval determination method and device
CN103489034A (en) * 2013-10-12 2014-01-01 山东省科学院海洋仪器仪表研究所 Method and device for predicting and diagnosing online ocean current monitoring data
CN103678883A (en) * 2013-11-21 2014-03-26 上海海洋大学 Space sampling method oriented to multisource marine environmental monitoring data
US20180313321A1 (en) * 2015-10-27 2018-11-01 IFP Energies Nouvelles Method for predicting a characteristic resulting from the swell on a floating system for at least two future time steps
CN105787592A (en) * 2016-02-26 2016-07-20 河海大学 Wind turbine generator set ultra-short period wind power prediction method based on improved RBF network
CN106599577A (en) * 2016-12-13 2017-04-26 重庆邮电大学 ListNet learning-to-rank method combining RBM with feature selection
CN106897705A (en) * 2017-03-01 2017-06-27 上海海洋大学 A kind of oceanographic observation big data location mode based on incremental learning
CN107364548A (en) * 2017-06-01 2017-11-21 中国海洋石油总公司 Internal wave of ocean observes buoy and its typhoon protection design method, typhoon protection method
CN108036961A (en) * 2017-12-07 2018-05-15 国家海洋局南海环境监测中心 Marine environmental monitoring Multitask Data acquisition method, storage medium and electronic equipment
CN108761497A (en) * 2018-03-06 2018-11-06 北京交通大学 A kind of train satellite positioning local completeness monitoring method and system
CN108898223A (en) * 2018-07-11 2018-11-27 国家海洋技术中心 A kind of ocean observation device operating status method for detecting abnormality and device
CN109947879A (en) * 2019-01-29 2019-06-28 中国海洋大学 A kind of oceanographic observation big data visual analysis method based on complex network

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
E. B. L. MACKAY 等: "A parametric model for ocean wave period from Ku band altimeter data", 《GEOPHYSICAL RESEARCH: OCEANS 》 *
JIGNESHKUMAR R. KSHATRIYA等: "Determination of Ocean Wave Period from Altimeter Data Using Wave-Age Concept", 《MARINE GEODESY 》 *
WENQING LI 等: "A VR-Ocean system for interactive geospatial analysis and 4D visualization of the marine environment around Antarctica", 《COMPUTERS & GEOSCIENCES 》 *
李民 等: "海洋环境定点平台观测技术概述及发展态势分析", 《海洋技术学报 》 *
杨晓华等: "水环境非线性时序预测的高精度RBF网络模型", 《水科学进展》 *
王美娜 等: "海洋测深中垂直分置海底混响信号的仿真分析", 《海洋测绘》 *
黄铭等: "基于径向基函数的海堤监控模型及预测置信度判别", 《工业建筑》 *

Also Published As

Publication number Publication date
CN110598860B (en) 2023-02-24

Similar Documents

Publication Publication Date Title
CN110689179A (en) Water bloom prediction method based on space-time sequence mixed model
CN108920812B (en) Machining surface roughness prediction method
CN110570013B (en) Single-station online wave period data prediction diagnosis method
CN110346844A (en) Quantitative Precipitation estimating and measuring method of the NRIET based on cloud classification and machine learning
CN108171259A (en) A kind of wisdom tip water quality monitoring system and method based on NB-IoT
CN111539515A (en) Complex equipment maintenance decision method based on fault prediction
WO2022021726A1 (en) Pmu-based power system state estimation performance evaluation method
CN109919356B (en) BP neural network-based interval water demand prediction method
CN110533239B (en) Smart city air quality high-precision measurement method
CN112926265A (en) Atmospheric porous probe measurement calibration method based on genetic algorithm optimization neural network
CN110222411B (en) Non-measurement-area solar radiation estimation method based on mRMR-DBN algorithm
CN111767517A (en) BiGRU multi-step prediction method and system applied to flood prediction and storage medium
CN111814956A (en) Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction
CN113592144A (en) Medium-and-long-term runoff probability forecasting method and system
CN113895272A (en) Electric vehicle alternating current charging state monitoring and fault early warning method based on deep learning
CN112070103B (en) Method for inverting atmospheric visibility through microwave link network gridding self-adaptive variable scale
CN113505534A (en) Load prediction method considering demand response
CN114154619A (en) Ship track prediction method based on CNN and BILSTM
CN113189623A (en) Satellite navigation system signal quality grade evaluation method
CN116227689A (en) Photovoltaic panel pollution degree state evaluation and cleaning decision system
CN113095389B (en) Method and system for calibrating monitoring network of gridding sensor
CN114638358A (en) Extinction coefficient-to-particulate matter concentration method, device and computer storage medium
CN110598860B (en) Multi-station online wave cycle data prediction diagnosis method
CN115408483B (en) Beidou navigation service positioning performance intelligent prediction method and device
CN113627074B (en) Ground wave propagation delay prediction method based on transfer learning

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