CN115359197A - Geological curved surface reconstruction method based on spatial autocorrelation neural network - Google Patents
Geological curved surface reconstruction method based on spatial autocorrelation neural network Download PDFInfo
- Publication number
- CN115359197A CN115359197A CN202210984780.2A CN202210984780A CN115359197A CN 115359197 A CN115359197 A CN 115359197A CN 202210984780 A CN202210984780 A CN 202210984780A CN 115359197 A CN115359197 A CN 115359197A
- Authority
- CN
- China
- Prior art keywords
- curved surface
- neural network
- spatial
- autocorrelation
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/05—Geographic models
-
- 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
Abstract
The invention discloses a geological curved surface reconstruction method based on a spatial autocorrelation neural network, which is applied to the field of oil and gas exploration and development; aiming at the problem that the curved surface obtained by the following technology loses morphological characteristics; firstly, training a constructed spatial autocorrelation neural network based on seismic interpretation data, making a prediction data set, inputting the prediction data set into the training ten-thousand-morning spatial autocorrelation neural network, and obtaining a prediction curved surface model; and for an unreasonable geological curved surface model, morphological characteristic data is extracted and added into a training data set, and a new round of curved surface reconstruction is carried out until a reasonable geological curved surface model is obtained.
Description
Technical Field
The invention belongs to the field of oil-gas exploration and development, and particularly relates to a geological curved surface reconstruction technology.
Background
The geological curved surface reconstruction is one of the applications of the curved surface reconstruction in the field of geological exploration, and is the basis of research works such as oil-gas migration path analysis, reserve calculation, geological structure three-dimensional modeling and the like in oil-gas exploration and development. Due to the fact that some positions which cannot be observed exist in a geographic space or observation points which are difficult to arrange in large quantity exist in the geographic space, a large number of unknown data points exist in a space research area, the reconstruction of the geological curved surface is affected by uncertainty to a large extent, and how to estimate the unknown data through the observation data of the known positions is achieved, so that a reasonable geological curved surface model which is high in fitting degree, good in smoothness and accurate in morphological characteristics is constructed, and the method is a main problem in the field of current geological curved surface reconstruction.
Spatial correlation is the basis for supporting spatial interpolation, and Tobler (1970) proposes the first theorem of geography that "everything is correlated with others, but things that are closer are more correlated than things that are farther", indicating the potential interdependence between observed data of some variables in the same distribution area. Therefore, the accuracy and reliability of spatial interpolation prediction are determined by the solving accuracy of the spatial weight relationship among the data.
However, due to objective factors such as difficulty in acquiring geological exploration data, limitation of geological rules, and insufficient structural interpretation, it is often difficult to accurately fit the spatial weight relationship of a curved surface in a conventional spatial interpolation method, such as an inverse distance weighting method (IDW). In recent years, with the development of deep learning, because a neural network has strong nonlinear fitting capability, part of researchers apply the neural network to point cloud reconstruction and curved surface reconstruction, and extremely remarkable application effect is achieved. However, most deep learning at present needs a large number of training samples to obtain a good model. For the field of few samples and single sample, such as geological exploration and medical imaging, a large amount of sample data is needed for establishing a high-precision model, and the data is difficult to obtain, expensive or impractical.
The prior art has the following defects:
(1) Because input data is sparse and uncertain, the curved surface obtained by the existing method loses morphological characteristics.
(2) Because the geological curved surface has a complex geological structure, the existing method needs to construct a complex topological relation before interpolation, the algorithm efficiency is low, and the method has no universality.
(3) The traditional spatial interpolation method needs prior hypothesis conditions, the model is relatively simple, but the mutual relation between the weight and the spatial distance is difficult to accurately fit.
Disclosure of Invention
In order to solve the technical problems, the invention provides a geological curved surface reconstruction method based on a spatial autocorrelation neural network, which can effectively utilize the spatial correlation among seismic interpretation data and construct a reasonable geological curved surface model with high fitting degree, good smoothness and accurate morphological characteristics by adding a morphological characteristic line which is visually and interactively marked manually.
The technical scheme adopted by the invention is as follows: a geological curved surface reconstruction method based on a spatial autocorrelation neural network comprises the following steps:
s1, acquiring original seismic interpretation data, and manufacturing a training data set according to the acquired original seismic interpretation data;
s2, building a spatial autocorrelation neural network model by adopting a BP neural network, wherein an input layer of the spatial autocorrelation neural network model is set to be n-1 neurons, and n is the total number of sample points in a training data set;
s3, training the autocorrelation neural network model built in the step S2 by adopting the training data made in the step S1;
s4, making a prediction data set;
and S5, inputting the prediction data set into the autocorrelation neural network model trained in the step S3 to obtain a prediction curved surface model.
The method further comprises the steps of carrying out rationality judgment on the predicted surface model, if the obtained predicted surface model is unreasonable, marking morphological characteristic lines unreasonable through visual interaction, then solving the intersection points of the artificially marked morphological characteristic lines and the original explained seed lines to obtain morphological characteristic points, and carrying out least square fitting on the morphological characteristic points to obtain morphological characteristic data; and adding the obtained morphological characteristic data into the training data set in the step S1.
The invention has the beneficial effects that: the method is different from the traditional interpolation method, is based on the spatial autocorrelation neural network, can well process multidimensional data, and can accurately fit the correlation between the weight and the spatial distance. In addition, aiming at the problems of uncertainty and form characteristic loss of the target geological surface, the method introduces geological expert knowledge into network regression in a visual interaction mode to guide the reconstruction process of the constrained surface. The effectiveness of the method is proved by testing a theoretical model. Compared with a common interpolation method, the method provided by the invention is more suitable for analysis and explanation of the geological curved surface. In conclusion, the method provided by the invention can construct a reasonable geological curved surface model with high fitting degree, good smoothness and accurate morphological characteristics, and can effectively promote the development and application of deep learning and signal processing theories in the field of geological curved surface reconstruction.
Drawings
FIG. 1 is a neuronal structure;
FIG. 2 is a multi-level perceptron configuration;
FIG. 3 is a flow chart of an embodiment of the present invention;
FIG. 4 is a model of a spatial autocorrelation neural network;
FIG. 5 is a schematic diagram of the effect of model complexity on under-fit and over-fit;
FIG. 6 is theoretical geological interpretation data;
FIG. 7 is a theoretical surface model without morphological features;
FIG. 8 is a contour diagram of a theoretical curved surface;
FIG. 9 is three-dimensional morphological feature data;
FIG. 10 is a theoretical surface model including morphological features;
FIG. 11 is a layer level explanation data of a certain work area in Chuandong;
FIG. 12 is a curved surface model of a work area in Chuandong;
wherein, (a) is an initial reconstruction curved surface model, and (b) is a final reconstruction curved surface model;
FIG. 13 is a contour diagram of curved surfaces of a work area in Chuandong;
wherein, (a) is the contour map of the initial reconstruction curved surface, and (b) is the contour map of the final reconstruction curved surface.
Detailed Description
In order to facilitate understanding of the technical contents of the present invention by those skilled in the art, a description will first be given of the related art:
1. inverse distance weight method (IDW)
Inverse Distance Weight (IDW) is an interpolation method in which the spatial distance between a point to be interpolated and a known sample point is weighted-averaged as a weight parameter.
For the point to be interpolated, the closer to the known sample point, the greater the weight assigned, whose weight contribution is inversely proportional to the distance, and the general formula is:
wherein the content of the first and second substances,
is an estimated value of the ith point to be interpolated, m represents the total number of points to be interpolated, z j Is the true value of the jth sample point, n represents the total number of sample points, d ij Is the Euclidean distance between the point i to be interpolated and the j point. p is a weighted power exponent that adjusts the shape of the surface of the interpolation function, typically to 2.
The calculation steps of the inverse distance weight method are as follows:
(1) calculating the Euclidean distance d between the point to be interpolated and the known point ij 。
Wherein the content of the first and second substances,
as the coordinates of the point to be interpolated, (x) j ,y j ) Is the known sample point coordinates.
(2) Calculating the space weight w between the point to be interpolated and each sample point by using the weight function i 。
(3) And calculating an estimated value of the point to be interpolated.
The advantage of the inverse distance weight method is that the formula is relatively simple. The method has the disadvantages that the method is a global interpolation calculation method, extreme values in sample points can directly influence the whole interpolation curved surface, and if the number of sampling points participating in interpolation is large, a large amount of calculation is needed for calculating one node.
2. Theory of neural networks
Neurons are the basic units that make up neural networks, the model of which is shown in fig. 1.
Wherein a is k For each component of the input vector, w k Is the weight coefficient of the input, b is the bias of the neuron, f is the activation function, typically a non-linear function, and t is the neuron output. The mathematical representation is:
the multi-layer perceptron (MLP) is a feedforward (BP) artificial neural network model, which introduces one to many hidden layers (hidden layers) on the basis of a single-layer neural network, as shown in fig. 2. The layers of the multilayer perceptron are all connected, the input layer receives input signals, the hidden layer processes the input signals, the activation function is used for carrying out nonlinear conversion on the signals, and finally the output layer outputs results. The multilayer perceptron has strong nonlinear fitting capability and is widely applied to various fields.
The present invention will be further explained with reference to the accompanying drawings.
As shown in fig. 3, the method of the present invention includes the following:
1. data pre-processing
Firstly, the existing data is processed, and the steps are as follows:
11. and preprocessing the seismic interpretation data. According to seismic data obtained from oil fields and geophysical service companies, the interpreter interprets the seismic data by the horizon and fault to obtain original seismic interpretation data. The work area interpretation data often comprise point cloud data of a plurality of horizons and faults and are characterized by dense data in the line measuring direction and sparse data among lines. Therefore, the interpretation data needs to be preprocessed, including data line direction sampling, work area boundary extraction, and data normalization.
12. A training data set is made. The preprocessed interpretation data is three-dimensional coordinate point cloud (x) i ,y i ,z i ) Two-dimensional coordinates (x) of the point cloud i ,y i ) As training data set, elevation value z i As label data for the training model.
2. Spatial autocorrelation neural networks
After the data preparation work is completed, the spatial autocorrelation neural network is required to be built, and the network model is trained according to the overall training framework, and the method comprises the following steps:
21. and building a spatial autocorrelation neural network.
And training by taking the component difference of the Euclidean space distance between the space point clouds as input, and for different geological curved surfaces, performing different training to obtain the neural network.
In the invention, the inverse distance weight method and the BP neural network are combined, the component difference of the Euclidean space distance between the space point clouds is taken as input, and the weight coefficient w is subjected to weight matching through the neural network ij And a spatial distance d ij The nonlinear relation between the two is learned, so that the prediction from the sample point to the point to be interpolated is realized, and the weight function of the neural network is obtained by comparing a space weight calculation formula (3) of an inverse distance weight method as follows:
w i ＝(w i1 ,w i2 ,...,w in )＝f(d i1 ,d i2 ,...,d in ) (6)
wherein w i A spatial weight vector, w, representing the ith point ij Represents the ith pointSpatial weight with j-th point, d ij Representing the Euclidean distance between the ith point and the jth point, and defined as formula (2), wherein f is a spatial weight calculation formula (3).
Meanwhile, the spatial distance between itself and itself is 0, and to prevent overfitting, the weight should be 0, and the formula is as follows:
the product of the spatial weight W and the known sample point is the estimation result of the point to be estimated:
according to the principle, the network model input layer is set to be n-1 neurons, and n is the total number of sample points in the training data set. The number of hidden layers and the number of neurons need to be adjusted in continuous training according to the complexity of the curved surface. Similarly, since the network output is the spatial weight between the point to be estimated and the rest of the sample points, the number of output layer neurons is n-1. The network model design is shown in fig. 4.
22. Build an overall training framework
The loss function can quantify the gap between the actual and predicted values of the target. Usually a non-negative number is chosen as the loss, and a smaller number indicates a smaller loss, which is 0 for perfect prediction. The most commonly used loss function in the regression problem is the root Mean Square Error (MSE), which is given by the following equation:
to measure the quality of the model over the entire data set, the loss mean (also equivalent to the sum) over n samples of the training set needs to be calculated.
In the cross validation process, the input of the network model is the space distance between the point to be estimated and other training sample points, the Mean Square Error (MSE) is used as a loss function, the space weight W is obtained through training, and the space weight W is multiplied by the real elevation value z of the known space sample to obtain the predicted value of the point to be estimated
Because of the problem of few training samples in the geological surface reconstruction, enough data cannot be provided to form a proper verification set, and a popular solution to the problem at present is to adopt k-fold cross verification.
Therefore, the present invention currently uses a ten-fold cross-validation method for model training and validation. Dividing the data set into ten parts, and taking 9 parts as a training data set and 1 part as a verification data set in turn. The training set is used for network training, and parameters and spatial autocorrelation neural network weight coefficients are fitted by continuously reducing training errors. And the verification set does not participate in the training process, and the generalization capability of the model is evaluated according to the obtained generalization error. The complexity of the current model can be judged by observing the direct relationship between the training error and the generalization error, so as to adjust the hyper-parameters of the network model, as shown in fig. 5.
When the training error and the generalization error are both large, the network model is in under-fitting, the capacity of the network model is too small, the number of layers or neurons of the hidden layer can be increased, and the capacity of the network model is improved; when the training error is small and the generalization error is large, the network is in overfitting, and the number of layers of hidden layers or the number of neurons can be properly reduced. Therefore, the number of the hidden layer layers and the number of the hidden layer neurons are continuously adjusted until the training error and the generalization error reach the expected values, and the difference between the training error and the generalization error is small.
The cross-validation method averages the results of different packet training, and reduces the sensitivity to the division of data. The invention adopts 10-fold cross validation, and comprises the following steps:
step1: the data was sampled non-repetitively and was randomly divided into 10 portions.
Step2: for each experiment, 1 part of the total was selected as a test set, and the remaining 9 parts were selected as a training set.
Step3: repeating the second step 10 times, and dividing the training set and the test set.
Step4: and training each training set to obtain a result, testing the test set, and storing the evaluation index of the model.
Step5: the mean of the 10 test results was calculated as the result accuracy estimate.
23. Selecting proper neural network optimization algorithm and initial test parameters
The correct selection of the gradient descent algorithm and the learning step length can lead the convergence of the loss function to be faster and more stable. Meanwhile, in order to prevent overfitting, a Dropout strategy is adopted by the network, so that part of neurons are inactivated, after multiple times of experimental verification, the activation probability of the neurons is determined, and is generally set to be 0.75. In addition, the number of layers of the hidden layer and the number of neurons also need to be adjusted and tested according to the complexity of the curved surface, the number of neurons in each layer also affects the network fitting effect, and under-fitting caused by too few neurons affects the network performance and cannot achieve the expected effect. Too many numbers result in too large calculation amount and low efficiency. An empirical formula is given for this Stackoverflow for reference:
wherein: n is a radical of hydrogen h Is the number of input layer neurons; n is a radical of i Is the number of neurons in the output layer; n is a radical of hydrogen s Is the number of samples of the training set; α is any variable that can be taken from, typically ranging from 2 to 10.
24. Producing a prediction dataset
The neural network obtained after cross validation training has learned proper spatial weight, and then grid data (x) to be interpolated is generated according to the work area boundary i ,y i ) The subdivision interval of the mesh is determined according to the actual task requirement, and the smaller the subdivision interval is, the finer and smoother the generated curved surface is.
3. Morphological feature based surface reconstruction
The main function of the spatial autocorrelation neural network is to perform regression fitting on the interpretation data, and the obtained geological curved surface may lose specific morphological characteristics, resulting in the construction of an unreasonable curved surface model, so that network training needs to be performed under the constraint of morphological characteristics, and the detailed steps are as follows:
31. neural network surface reconstruction
And inputting the prediction data into the trained spatial autocorrelation neural network to obtain a prediction curved surface model, and judging the rationality of the geological curved surface by a geological expert through the visualized curved surface model. If the geological curved surface is unreasonable, visual interaction is needed, morphological characteristic lines are marked artificially at unreasonable positions of the curved surface, wherein morphological characteristics mainly comprise ridges, valleys, saddles, peaks, basins and the like in the field of geological curved surfaces. And then, solving an intersection point between the manually marked morphological characteristic line and the original explained seed line to obtain morphological characteristic points, and performing least square fitting on the morphological characteristic points to obtain morphological characteristic data.
The present invention illustrates this process by designing a theoretical model. In a 1 × 1 × 1 space, two parallel sinusoidal space curves are generated by formula (11), and total 500 points are calculated, thereby simulating geological interpretation data having morphological features such as ridges and valleys, as shown in fig. 6.
The input layer inputs the space distance between the current point to be estimated and the rest sample points, so the number of neurons in the network input layer is 499. Hidden layer number is set to three layers according to the capacity of the model, the hidden layer neuron number is calculated by formula (11), where the activation function is set to ReLu. And outputting the spatial correlation weights of the to-be-estimated point of the horizon and other sample points, wherein the number of the neurons is 499.
Through continuous simulation experiments, appropriate training hyper-parameters are set, and the specific hyper-parameter setting of the whole neural network is shown in table 1.
TABLE 1 neural network hyperparameters
After model training and verification are performed by a ten-fold cross verification method, a trained spatial autocorrelation neural network model is obtained, a 100 × 100 interpolation grid is established as a prediction data set, and finally a curved surface model is obtained through prediction, as shown in fig. 7.
As can be seen from the figure, the predicted surface has high fitting degree with the original data and good smoothness, and the correlation between the weight and the spatial distance is well fitted. However, because there is no data between the two measuring lines, there are many uncertainties in the surface, resulting in uncontrolled reconstructed surface, discontinuous and missing valleys and ridges, and various morphological possibilities for the surface.
32. Manually visualized interaction and marking of two-dimensional characteristic lines
The geology expert can roughly guess the reasonable morphological characteristics of the geological curved surface according to own experience and professional knowledge, visually interacts the predicted curved surface, and manually marks two-dimensional morphological characteristic lines including ridge lines and valley lines on the curved surface contour map, as shown in fig. 8.
33. Extracting three-dimensional morphological feature data
The marked two-dimensional morphological characteristic line adds characteristic information to the curved surface, the intersection point of the characteristic line and the seed line is obtained through pixel coordinate conversion, and the intersection point is fitted through a least square method to obtain three-dimensional morphological characteristic data, as shown in fig. 9:
34. iterative reconstruction
And adding the morphological characteristic data into a training data set, and performing the next round of curved surface reconstruction until a geological curved surface model considered reasonable by a geological expert is obtained. FIG. 10 is a rational curved surface obtained by performing a spatial autocorrelation neural network model on a theoretical model under the guidance of morphological features.
The effects of the present invention are explained below with reference to specific data:
the method provided by the invention is applied to the reconstruction of the geological curved surface of a certain work area in the east Chuandong in China, so as to verify the effectiveness of the method.
The method preprocesses the partial horizon curved surface interpretation data of a certain work area in Chuandong in China to obtain a sample data set with 728 points, as shown in figure 11. The sample data set is input to the spatial autoregressive neural network, and an initial reconstructed curved surface is obtained as shown in fig. 12 (a). Thereafter, the contour map of the initial reconstructed curved surface is manually marked, and a red line is marked at a position where a ridge should be present, and a blue line is marked at a position where a valley is present, as shown in fig. 13 (a). Then, the obtained morphological characteristic data points are added into the sample data set again, the next round of neural network training is carried out, and the curved surface model guided by the morphological characteristics is obtained, as shown in fig. 12 (b) and 13 (b), the method can be seen to have good effect improvement compared with the initial curved surface, and the method can well construct a reasonable geological curved surface model with high fitting degree, good smoothness and accurate morphological characteristics.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Various modifications and alterations to this invention will become apparent to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.
Claims (6)
1. A geological curved surface reconstruction method based on a spatial autocorrelation neural network is characterized by comprising the following steps:
s1, acquiring original seismic interpretation data, and manufacturing a training data set according to the acquired original seismic interpretation data;
s2, building a spatial autocorrelation neural network model by adopting a BP neural network, wherein an input layer of the spatial autocorrelation neural network model is set to be n-1 neurons, and n is the total number of sample points in a training data set;
s3, training the autocorrelation neural network model built in the step S2 by adopting the training data made in the step S1;
s4, making a prediction data set;
and S5, inputting the prediction data set into the autocorrelation neural network model trained in the step S3 to obtain a prediction curved surface model.
2. The geological curved surface reconstruction method based on the spatial autocorrelation neural network is characterized by further comprising the steps of carrying out rationality judgment on a predicted curved surface model, if the obtained predicted curved surface model is unreasonable, marking morphological characteristic lines unreasonably through visual interaction, then solving intersection points of the manually marked morphological characteristic lines and original explained seed lines to obtain morphological characteristic points, and carrying out least square fitting on the morphological characteristic points to obtain morphological characteristic data; and adding the obtained morphological characteristic data into the training data set in the step S1.
3. The method for reconstructing the geological curved surface based on the spatial autocorrelation neural network as claimed in claim 2, wherein the step S1 specifically comprises: preprocessing the acquired original seismic interpretation data, wherein the preprocessed interpretation data is three-dimensional coordinate point cloud (x) i ,y i ,z i ) Two-dimensional coordinates (x) of the point cloud i ,y i ) As a training data set, elevation values z i As tag data.
4. The method for reconstructing the geological curved surface based on the spatial autocorrelation neural network as claimed in claim 3, wherein the step S3 specifically comprises: the component difference of the Euclidean space distance between the space point clouds is used as input, a space autocorrelation neural network model is trained, and a weight system is learned through the space autocorrelation neural network modelNumber w ij And a spatial distance d ij And the non-linear relation between the two points is realized, so that the prediction from the sample point to the point to be interpolated is realized.
5. The method for reconstructing the geological curved surface based on the spatial autocorrelation neural network as claimed in claim 3, wherein a root mean square error is adopted as a loss function in the training process.
6. The method for reconstructing the geological curved surface based on the spatial autocorrelation neural network as claimed in claim 5, wherein the expression of the loss function is:
wherein L (W, b) represents a loss function, W represents a spatial weight, z represents a true elevation value of a known spatial sample,
representing the predicted value of the point to be estimated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210984780.2A CN115359197A (en) | 2022-08-17 | 2022-08-17 | Geological curved surface reconstruction method based on spatial autocorrelation neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210984780.2A CN115359197A (en) | 2022-08-17 | 2022-08-17 | Geological curved surface reconstruction method based on spatial autocorrelation neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115359197A true CN115359197A (en) | 2022-11-18 |
Family
ID=84003398
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210984780.2A Pending CN115359197A (en) | 2022-08-17 | 2022-08-17 | Geological curved surface reconstruction method based on spatial autocorrelation neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115359197A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116660996A (en) * | 2023-07-24 | 2023-08-29 | 山东科技大学 | Drifting type shallow sea local earth sound parameter prediction method based on deep learning |
-
2022
- 2022-08-17 CN CN202210984780.2A patent/CN115359197A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116660996A (en) * | 2023-07-24 | 2023-08-29 | 山东科技大学 | Drifting type shallow sea local earth sound parameter prediction method based on deep learning |
| CN116660996B (en) * | 2023-07-24 | 2023-10-13 | 山东科技大学 | Drifting type shallow sea local earth sound parameter prediction method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109492822B (en) | Air pollutant concentration time-space domain correlation prediction method | |
| CN108399248A (en) | A kind of time series data prediction technique, device and equipment | |
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| CN105260786B (en) | A kind of simulation credibility of electric propulsion system assessment models comprehensive optimization method | |
| CN110689179A (en) | Water bloom prediction method based on space-time sequence mixed model | |
| CN111797364B (en) | Landslide multilayer safety evaluation method based on composite cloud model | |
| CN110751318A (en) | IPSO-LSTM-based ultra-short-term power load prediction method | |
| CN106656357B (en) | Power frequency communication channel state evaluation system and method | |
| CN111461453A (en) | Medium-and-long-term runoff ensemble forecasting method based on multi-model combination | |
| CN112149230B (en) | Method for predicting comfort deterioration of wind-induced train of strong wind railway | |
| CN104634265B (en) | A kind of mineral floating froth bed soft measurement method of thickness based on multiplex images Fusion Features | |
| CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
| CN108416458A (en) | A kind of tunnel rich water rock mass Synthetic Geological Prediction Ahead of Construction method based on BP neural network | |
| CN115471679A (en) | Method and intelligent system for synchronously assimilating water level and flow of natural river | |
| CN115359197A (en) | Geological curved surface reconstruction method based on spatial autocorrelation neural network | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| Mohammad et al. | Short term load forecasting using deep neural networks | |
| CN115421216A (en) | STL-ARIMA-NAR mixed model-based medium-and-long-term monthly rainfall forecasting method | |
| CN113505477B (en) | Process industry soft measurement data supplementing method based on SVAE-WGAN | |
| CN112307410A (en) | Seawater temperature and salinity information time sequence prediction method based on shipborne CTD measurement data | |
| CN116720743A (en) | Carbon emission measuring and calculating method based on data clustering and machine learning | |
| CN115409317A (en) | Transformer area line loss detection method and device based on feature selection and machine learning | |
| CN113657021A (en) | Marine measurement period evaluation method based on BP neural network | |
| Rochac et al. | A Hybrid ConvLSTM Deep Neural Network for Noise Reduction and Data Augmentation for Prediction of Non-linear Dynamics of Streamflow | |
| Yanto et al. | Hybrid Method Air Quality Classification Analysis 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 |