CN113688770B - Method and device for supplementing long-term wind pressure missing data of high-rise building - Google Patents
Method and device for supplementing long-term wind pressure missing data of high-rise building Download PDFInfo
- Publication number
- CN113688770B CN113688770B CN202111026123.9A CN202111026123A CN113688770B CN 113688770 B CN113688770 B CN 113688770B CN 202111026123 A CN202111026123 A CN 202111026123A CN 113688770 B CN113688770 B CN 113688770B
- Authority
- CN
- China
- Prior art keywords
- wind pressure
- data
- pressure data
- sequence
- missing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 230000007774 longterm Effects 0.000 title claims abstract description 23
- 230000001502 supplementing effect Effects 0.000 title claims abstract description 16
- 238000003062 neural network model Methods 0.000 claims abstract description 52
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 46
- 239000011159 matrix material Substances 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 31
- 230000000295 complement effect Effects 0.000 claims description 14
- 238000013527 convolutional neural network Methods 0.000 claims description 14
- 238000005259 measurement Methods 0.000 claims description 12
- 238000012216 screening Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 7
- 239000000284 extract Substances 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 description 18
- 238000010586 diagram Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005251 gamma ray Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a method and a device for supplementing long-term wind pressure missing data of a high-rise building, which comprise the steps of collecting wind pressure data of the high-rise building, determining missing data measuring points, historical wind pressure data and surrounding wind pressure data, and respectively integrating the historical wind pressure data and the surrounding wind pressure data into a wind pressure data sequence matrix according to time sequence; decomposing the wind pressure data sequence matrix by adopting an empirical mode decomposition algorithm to obtain an EMD decomposition data sample; inputting EMD decomposition data samples and missing data measuring points into a CNN-BiGRU neural network model for model training; when the real-time wind pressure data is missing, the missing data is complemented by wind pressure prediction data obtained through the prediction of the CNN-BiGRU neural network model; the invention is realized based on the EMD algorithm and the CNN-BiGRU model, so that the prediction accuracy of wind pressure prediction data is high, and the surface wind pressure of the high-rise building can be accurately estimated.
Description
Technical Field
The invention relates to the technical field of wind pressure prediction of building structures, in particular to a method and a device for supplementing long-term wind pressure missing data of a high-rise building.
Background
Wind pressure sensors are of great significance for assessing building surface pressure, but some wind pressure sensors fail or abnormality in large-scale wind pressure data during long-term operation, resulting in large-scale data loss of building surface wind pressure over a long period of time, and these data are difficult to recover.
At present, the method for predicting the missing wind pressure data of the high-rise building based on the artificial intelligence method is mainly divided into two types: one is a "shallow" machine learning method, the wind pressure data has high nonlinearity and non-stationarity, the model of "shallow" has certain limitation to the short-term prediction of wind load, can't process massive monitoring data and the accuracy is lower; the other method is a traditional neural network model, has the characteristics of universality, high efficiency and the like, and the accuracy is still to be further improved. Therefore, it is necessary to further improve the accuracy of the traditional neural network model and develop the wind pressure prediction work of the high-rise building accurately and in real time.
Disclosure of Invention
In view of the above, the invention aims to provide a method and a device for supplementing long-term wind pressure missing data of a high-rise building, which are used for solving the problem that wind pressure data on the surface of the high-rise building is difficult to accurately predict and supplement after large-scale missing in the prior art.
In order to achieve the above purpose, an aspect of the present invention provides a method for supplementing long-term wind pressure missing data for a high-rise building, which specifically includes the following steps:
s1: collecting wind pressure data of a high-rise building, determining missing data measuring points, selecting historical wind pressure data of the missing data measuring points and measured surrounding wind pressure data of surrounding data measuring points corresponding to each missing data measuring point, and integrating the historical wind pressure data and the surrounding wind pressure data thereof into a wind pressure data sequence matrix according to time sequence;
s2: decomposing the wind pressure data sequence matrix into an IMF component and a RES component by adopting an empirical mode decomposition algorithm to obtain an EMD decomposed data sample;
s3: inputting the EMD decomposition data sample and the historical wind pressure data measured by the missing data measurement points into a CNN-BiGRU neural network model for iterative training, and optimizing model parameters to obtain a trained CNN-BiGRU neural network model;
s4: when the real-time wind pressure data is lost, the wind pressure data before the data loss and the wind pressure data around the wind pressure data are input into a trained CNN-BiGRU neural network model, wind pressure prediction data are obtained through the CNN-BiGRU neural network model prediction, and the lost data are complemented by the wind pressure prediction data.
Further, in the step S2, the wind pressure data sequence matrix includes a historical wind pressure data sequence and surrounding wind pressure data sequences thereof, the historical wind pressure data sequence and surrounding wind pressure data sequences thereof are recorded as original wind pressure data sequences, and each original wind pressure data sequence matrix is subjected to empirical mode decomposition, which specifically includes the steps of:
s201: screening all maximum value points and minimum value points in the original wind pressure data sequence, and respectively fitting the maximum value points and the minimum value points to obtain an upper envelope curve and a lower envelope curve of the original wind pressure data sequence;
s202: calculating the average value envelope of the original wind pressure data sequence according to the upper envelope line and the lower envelope line, and calculating the difference value between the original wind pressure data sequence and the average value envelope to obtain a first wind pressure data sequence;
s203: judging whether the first wind pressure data sequence meets the condition that the IMF component is met, if yes, executing step S204, and if not, repeatedly executing steps S201-S202 on the first wind pressure data sequence until the ith wind pressure data sequence obtained after m times of repetition meets the condition that the IMF component is met;
s204: taking the ith wind pressure data sequence as a first IMF component of the original wind pressure data sequence, and separating the first IMF component from the original wind pressure data sequence to obtain a first residual wind pressure data sequence;
s205: and judging whether the first residual wind pressure data sequence is a monotonic function, if so, completing the decomposition, otherwise, repeatedly executing the steps S201-S204 on the first residual wind pressure data sequence until the j residual wind pressure data sequence obtained by repeating p times is a monotonic function, so as to decompose and obtain p IMF components and one RES component.
Further, in the step S203, the condition that the IMF component is satisfied is:
in the whole time interval, the number of maximum value points and minimum value points on the first wind pressure data sequence is equal to or at most different from the number of zero crossing points;
and the average value of the upper envelope curve and the lower envelope curve at any point on the first wind pressure data sequence is zero in the whole time interval.
Further, the step S3 includes the following steps:
s301: constructing a CNN-BiGRU neural network model comprising a CNN network and a BiGRU network, and inputting the EMD decomposition data sample obtained in the step S2 and the historical wind pressure data measured by the missing data measuring points into the constructed CNN-BiGRU neural network model;
s302: the CNN network extracts a space correlation characteristic sequence between surrounding wind pressure data and missing data measuring points based on a wind pressure data sequence matrix and the missing data measuring points;
s303: the BiGRU network takes the spatial correlation characteristic sequence extracted by the CNN network as input, further extracts the time front-back dependency characteristic between a wind pressure data sequence matrix and a missing data measuring point, and outputs wind pressure prediction data;
s304: and reversely transmitting wind pressure prediction data to the CNN-BiGRU neural network model for iterative training, and completing training until the loss function of the CNN-BiGRU neural network model tends to be stable or reaches the maximum iteration times, thereby obtaining a trained CNN-BiGRU neural network model.
Further, in step S302, the CNN network includes a convolution layer and a pooling layer, where the convolution layer learns the local features of each historical wind pressure data sequence in the wind pressure data sequence matrix comprehensively through convolution operation to obtain global features, and the pooling layer eliminates redundant feature information to extract the spatial correlation feature sequence between surrounding wind pressure data and missing data measurement points.
Further, in step S303, the CNN-biglu neural network model includes two GRU sub-models with the same structure, and the two GRU sub-models process the input IMF component along the positive sequence and the negative sequence of time respectively, and the training process is as follows:
s3031: the spatial correlation characteristic sequences extracted in the step S302 are respectively input into two GRU sub-models, and a reset gate and an update gate of the two GRU sub-models are calculated;
s3032: calculating candidate activation states of the two GRU sub-models respectively, and calculating to obtain hidden layer output of the corresponding GRU sub-model according to the candidate activation states, the reset gate and the update gate;
s3033: and fusing the calculated hidden layer outputs of the two GRU submodels to obtain the time front-back dependency characteristics between the EMD decomposition data sample and the missing data measuring point, and further outputting a wind pressure predicted value.
Further, in step S3033, the wind pressure predicted value y t The method comprises the following steps:
wherein:outputting a hidden layer which propagates forward for the GRU submodel at the moment t; />Outputting a hidden layer which is transmitted backward for another GRU submodel at the moment t; alpha t ,β t Weights output by hidden layers of forward propagation and backward propagation of the two GRU submodels at the moment t respectively; b t And (5) hiding the bias quantity corresponding to the layer state for the BiGRU network at the moment t.
Further, in step S304, the loss function uses a mean square error function, where the expression is:
wherein: x is x t Inputting the spatial correlation characteristic of the BiGRU network at the moment t, y t The method is characterized in that wind pressure prediction data of the high-rise building to be predicted at the time t is obtained, and N is the number of iterations of model training.
Another aspect of the present invention provides a long-term wind pressure missing data complement apparatus for a high-rise building, comprising:
the data acquisition module is used for acquiring wind pressure data of the high-rise building, wherein the wind pressure data comprises historical wind pressure data and real-time wind pressure data;
the data screening and integrating module is used for determining missing data measuring points, selecting historical wind pressure data at the moment before the missing data measuring points and surrounding wind pressure data of each historical wind pressure data, and integrating the historical wind pressure data and the surrounding wind pressure data thereof into a wind pressure data sequence matrix according to a time sequence;
the data decomposition module is used for performing empirical mode decomposition on the wind pressure data sequence matrix output by the data screening and integrating module to obtain a plurality of IMF components and one RES component, so as to form an EMD decomposition data sample;
the model training module is used for inputting the EMD decomposition data sample obtained by the data decomposition module and the historical wind pressure data measured by the missing data measuring points into the CNN-BiGRU neural network model for iterative training to obtain a trained CNN-BiGRU neural network model;
the prediction data output module is used for outputting wind pressure prediction data; and
and the missing data complement module is used for complementing the wind pressure data after the data missing by utilizing the wind pressure prediction data output by the prediction data output module when the real-time wind pressure data is missing.
Further, the wind pressure prediction device further comprises a display module for displaying the wind pressure prediction data output by the prediction data output module.
According to the invention, based on an empirical mode decomposition algorithm and a CNN-BiGRU neural network model, nonlinear wind pressure data are decomposed into linear combinations of limited IMF components with frequencies from high to low through the empirical mode decomposition algorithm, each decomposed IMF component contains local characteristic signals of different time scales of original wind pressure data, and then the decomposed IMF components and original missing data measuring points are input into the CNN-BiGRU neural network model for extracting a spatial correlation characteristic sequence and a time dependency characteristic, so that prediction of large-scale wind pressure missing data is realized, the missing data is complemented by predicted wind pressure prediction data, and accurate assessment of the surface wind pressure of a high-rise building is facilitated.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the instrumentalities and combinations particularly pointed out in the specification.
Drawings
Fig. 1 is a flowchart of a method for supplementing long-term wind pressure missing data of a high-rise building according to an embodiment of the invention.
Fig. 2 is a flowchart of step S2.
Fig. 3 is a flowchart of step S3.
Fig. 4 is a schematic diagram of the CNN network in step S302.
Fig. 5 is a schematic diagram of the structure of the biglu network in step S303.
Fig. 6 is a flowchart of step S303.
Fig. 7 is a schematic structural diagram of the GRU sub-model in step S3031.
Fig. 8 is a schematic diagram illustrating the process of the pressure loss data complement in step S4.
Fig. 9 is a system block diagram of a high-rise building long-term wind pressure missing data complement device according to a second embodiment of the invention.
Detailed Description
The following is a further detailed description of the embodiments:
example 1
As shown in fig. 1, a flowchart of a method for supplementing long-term wind pressure missing data of a high-rise building according to the present embodiment is shown. The method for supplementing the long-term wind pressure missing data of the high-rise building specifically comprises the following steps:
s1: and (5) data acquisition.
Collecting wind pressure data of a high-rise building, and determining missing data measuring points x t (t is the missing data measurement point x t Data loss time of (2) selecting a missing data measurement point x t The historical wind pressure data at a previous time (i.e., the historical wind pressure data before time t) forms a historical wind pressure data set X, which may be expressed as:
X＝{x 1 ,x 2 ,…,x t-1 } (1)
selecting surrounding wind pressure data corresponding to each historical wind pressure data in the historical wind pressure data set X to form a surrounding wind pressure data set Y, wherein the surrounding wind pressure data is corresponding to each missing data measuring point X t The wind pressure data measured by the corresponding surrounding data measurement points can be selected according to the prediction accuracy to delete the data measurement point x in the embodiment t Points within the radius are designated for the center of the circle as ambient data points. The ambient wind pressure data set Y may be expressed as:
wherein: y is Y 1 ,Y 2 …Y N Surrounding wind pressure data sets corresponding to the historical wind pressure data in the historical wind pressure data set X are respectively; n is the number of surrounding wind pressure data sets, i.e. the number of historical wind pressure data in the historical wind pressure data set X.
Integrating the historical wind pressure data sets X according to time sequence to form a historical wind pressure data sequence X (t), and integrating the surrounding wind pressure data sets Y according to time sequence to form a surrounding wind pressure data sequence Y 1 (t),Y 2 (t)…Y N (t) combining said historical wind pressure data sequence X (t) with a surrounding wind pressure data sequence Y 1 (t),Y 2 (t)…Y N (t) forming a wind pressure data sequence matrix E (t), which may be expressed as:
s2: and (5) data decomposition.
Specifically, an Empirical Mode Decomposition (EMD) algorithm is used to divide the historical wind pressure data sequence X (t) and each surrounding wind pressure data sequence Y in the wind pressure data sequence matrix E (t) 1 (t),Y 2 (t)…Y N (t) decomposing into a plurality of IMF components and a RES component, respectively. In this embodiment, since the wind pressure data of the high-rise building is random, the obtained original wind pressure data sequence shows a tendency of non-steady rising or falling, so that the fluctuation and the tendency of different scales in the original wind pressure data sequence can be decomposed step by adopting the EMD algorithm to generate a series of data sequences with different characteristic scales, and each data sequence is an IMF component.
As shown in FIG. 2, the historical wind pressure data sequence X (t) and the surrounding wind pressure data sequence Y 1 (t),Y 2 (t)…Y N (t) denoted as the original wind pressure data sequence E 0 (t), said step S2 is performed on the original wind pressure data sequence E 0 (t) the specific steps of empirical mode decomposition are as follows:
s201: and calculating an upper envelope curve and a lower envelope curve of the original wind pressure data sequence.
Firstly, the original wind pressure data sequence E is screened out 0 All maxima points max [ E ] in (t) 0 (t)]Using cubic spline function to make all maximum points max E 0 (t)]Fitting the original wind pressure data sequence E 0 Upper envelope m of (t) 1 (t) max The method comprises the steps of carrying out a first treatment on the surface of the Then, the original wind pressure data sequence E is screened out 0 All minimum points min [ E ] in (t) 0 (t)]Using cubic spline function to make all minimum value points min [ E 0 (t)]Fitting the original wind pressure data sequence E 0 Lower envelope m of (t) 1 (t) min 。
S202: and calculating the mean envelope to obtain a first wind pressure data sequence.
According to the upper envelope m 1 (t) max And lower envelope line m 1 (t) min Calculating the original wind pressure data sequence E 0 Mean envelope m of (t) 1 (t)：
Based on the mean envelope m obtained in the above formula (2) 1 (t) calculating the original wind pressure data sequence E 0 (t) and mean envelope m 1 Obtaining a first wind pressure data sequence d from the difference between (t) 1 (t)：
d 1 (t)＝E 0 (t)-m 1 (t) (5)
S203: and judging the condition of the first wind pressure data sequence.
Judging the first wind pressure data sequence d obtained in the step S202 1 (t) whether two conditions for the IMF component to be satisfied are satisfied:
the first wind pressure data sequence d 1 (t) maximum point max [ d ] 1 (t)]And minimum point min [ d ] 1 (t)]The number of the zero crossing points is equal to or at most one different from the number of the zero crossing points;
the first wind pressure data sequence d 1 (t) an upper envelope m at any point on 2 (t) max And lower envelope line m 2 (t) min Mean envelope m of (2) 2 (t)＝0。
If the first wind pressure data sequence d 1 (t) satisfying the condition that the IMF component is satisfied, the step S204 is continued to be executed.
If the first wind pressure data sequence d 1 (t) if the condition for the IMF component to be satisfied is not satisfied, the first wind pressure data sequence d 1 (t) repeating steps S201-S202 as another original wind pressure data sequence until the ith wind pressure data sequence d obtained after repeating m times i (t) (where i=1, 2, …, m, m is the number of times steps S201-S202 are performed, i.e. d 1 (t) is a wind pressure data sequence obtained after performing the steps S201-S202 once, d m (t) is a wind pressure data sequence obtained after m times of S201-S202 are executed) until the condition that the IMF component is satisfied.
S204: and separating IMF components, and calculating a residual wind pressure data sequence of the original wind pressure data sequence.
The wind pressure data sequence d meeting the IMF component satisfaction condition obtained in the step S203 i (t) as a first IMF component, noted IMF 1 (t) and IMF the first IMF component 1 (t) from the raw wind pressure data sequence E 0 Separating in (t) to obtain a first residual wind pressure data sequence r 1 (t)：
r 1 (t)＝E 0 (t)-IMF 1 (t) (6)
S205: and judging the condition of the residual wind pressure data sequence.
Judging the first residual wind pressure data sequence r obtained by decomposition in the step S204 according to the convergence condition of the EMD algorithm 1 (t) whether or not it is a monotonic function.
If yes, the original wind pressure data sequence E is completed 0 (t) decomposing, otherwise, the remaining wind pressure data sequence r 1 (t) repeating steps S2031 to S2034 as a new original wind pressure data sequence until the jth remaining wind pressure data obtained by repeating n timesSequence r j (t) (where j=1, 2, …, n, n is the number of times steps S201-S204 are performed, i.e. r 1 (t) is the residual wind pressure data sequence obtained after performing the steps S201-S204 once, r n (t) is the remaining wind pressure data sequence obtained after n times of S201-S204 are executed) is a monotonic function. According to the convergence condition of EMD algorithm, when the residual wind pressure data sequence obtained by decomposition is a monotonic function, its time period will be greater than original wind pressure data sequence E 0 The recorded length of (t) so that the last remaining wind pressure data sequence r obtained by decomposition can be used n (t) as an original wind pressure data sequence E 0 (t), namely RES component, and the original wind pressure data sequence E 0 (t) can be expressed as:
namely, n IMF components obtained by decomposing a first wind pressure data sequence in a wind pressure data sequence matrix E (t) i (t)＝{IMF 1 (t),IMF 2 (t),…,IMF n (t) } and a RES component.
Decomposing the next wind pressure data sequence according to the steps S201-S205 until all wind pressure data sequences in the wind pressure data sequence matrix E (t) are completely decomposed, and obtaining an EMD decomposed data sample S:
wherein: p=n, the number of IMF components decomposed for each wind pressure data sequence; q=n, which is the number of wind pressure data sequences in the wind pressure data sequence matrix E (t); namely S 1,1 For the first IMF component of the first wind pressure data sequence, S 2,1 The first IMF component of the second wind pressure data sequence, and so on.
And then inputting the EMD decomposition data sample S into a CNN-BiGRU neural network model in the following step S3 for prediction.
S3: and (5) model training.
And inputting the EMD decomposition data sample obtained by decomposition and the historical wind pressure data measured by the missing data measurement points into a CNN-BiGRU neural network model, respectively extracting spatial correlation characteristics and time front-rear dependency characteristics of the wind pressure data, predicting to obtain wind pressure prediction data at the current moment, reversely inputting the wind pressure prediction data into the CNN-BiGRU neural network model for iterative training, and optimizing model parameters to obtain a trained CNN-BiGRU neural network model.
As shown in fig. 3, the specific process of training the CNN-biglu neural network model in step S3 is as follows:
s301: and constructing a CNN-BiGRU neural network model.
In large-scale wind pressure data prediction, from the space dimension, wind pressure data at a certain moment is associated with wind pressure data around the wind pressure data, and from the time dimension, the wind pressure data at a certain moment is also associated with the previous moment and the future moment at the same time, so that when large-scale missing of the wind pressure data occurs, the past wind pressure data influence factors, the future wind pressure data influence factors and the wind pressure data around are associated with the wind pressure data prediction at the current moment, the wind pressure prediction data can be more accurate, and the completion of missing data is further realized.
Because the CNN network has the characteristics of local connection, weight sharing and the like, the spatial correlation characteristics between the IMF component and the missing data measuring point can be reserved and extracted, and the BiGRU network can adaptively sense the up-down time sequence characteristic information, a CNN-BiGRU neural network model is constructed based on the CNN network and the BiGRU network, and the EMD decomposition data sample and the missing data measuring point which are obtained through decomposition in the step S2 are input into the constructed CNN-BiGRU neural network model.
S302: and extracting spatial correlation characteristics.
And the CNN extracts spatial correlation features between surrounding wind pressure data and missing data measuring points based on the EMD decomposition data samples and the historical wind pressure data measured by the missing data measuring points to form a spatial correlation feature sequence.
As shown in fig. 4, the CNN network includes a convolution layer and a pooling layer, where the convolution layer comprehensively learns local features of each historical wind pressure data sequence in the wind pressure data sequence matrix (i.e., features of different scales included in each IMF component in the EMD decomposed data sample) through convolution operation, so as to obtain global features. And then the pooling layer eliminates redundant characteristic information, and reserves space correlation characteristics so as to achieve the purpose of extracting the space correlation characteristics between surrounding wind pressure data and missing data measuring points, and integrates the space correlation characteristics into a space correlation characteristic sequence a (t) according to time sequence:
a(t)＝{a 1 ,a 2 ,…,a t } (9)
s303: and extracting the dependency characteristics before and after time, and outputting wind pressure prediction data.
The BiGRU network takes the spatial correlation characteristic sequence a (t) extracted by the CNN network as input, further extracts the time front-back dependency characteristic between the wind pressure data sequence matrix and the missing data measuring point, and outputs wind pressure prediction data.
Specifically, as shown in fig. 5, the structure diagram of the biglu network is shown, the biglu network is formed by combining two GRU sub-models with the same structure, the two GRU sub-models are respectively marked as a forward GRU sub-model and a backward GRU sub-model, the forward GRU sub-model processes the spatial correlation characteristic sequence a (t) of the input biglu network along a positive sequence, and the backward GRU sub-model processes the spatial correlation characteristic sequence a (t) of the input biglu network along a reverse sequence. The BiGRU network comprises an input layer, a hidden layer and an output layer, wherein the input layer is used for inputting a spatial correlation characteristic sequence a (t), the output layer is used for outputting wind pressure prediction data, and the output of the hidden layer can be obtained by weighted summation of the output of a forward hidden layer of a forward GRU sub-model and the output of a backward hidden layer of a backward GRU sub-model. Thus, the processing of the input spatial correlation feature sequence a (t) by the biglu network can also be regarded as the processing of the spatial correlation feature sequence a (t) by the two GRU sub-models, respectively.
As shown in fig. 6, the specific steps of outputting wind pressure data in step S303 are:
s3031: the reset gate and update gate of the GRU sub-model are calculated.
Specifically, as shown in fig. 7, the spatial correlation feature sequence a (t) extracted in step S302 is input into a forward GRU sub-model and a backward GRU sub-model, and reset gates r corresponding to the forward GRU sub-model and the backward GRU sub-model at time t are calculated respectively according to hidden layer output of the forward GRU sub-model and the backward GRU sub-model at t-1 t ：
Respectively calculating to obtain update gates z corresponding to the forward GRU sub-model and the backward GRU sub-model at the time t t ：
Wherein: w (W) r ,W z ,U r ,U z The model parameter matrix to be trained; sigma is a Sigmoid activation function; x is x t Inputting GRU submodel space correlation characteristics for the time t;the hidden layer outputs of the forward GRU sub-model and the backward GRU sub-model at the time t-1 are respectively carried out; the → indicates processing along the time preamble, +..
S3032: and calculating hidden layer output of the GRU submodel.
Inputting a forward GRU sub-model and a backward GRU sub-model according to the time tThe spatial correlation characteristic of the model, the reset gate r at time t is calculated in step S301 t And outputting hidden layers of the forward GRU sub-model and the backward GRU sub-model at the time t-1, and respectively calculating to obtain candidate activation states corresponding to the forward GRU sub-model and the backward GRU sub-model at the time t
Based on the candidate activation state at time tReset gate r t Updating door z t Calculating to obtain hidden layer output h of corresponding forward GRU sub-model and backward GRU sub-model at t moment t ：
Wherein: w, U is a model parameter matrix to be trained; tanh is a hyperbolic tangent function;representing the product of the matrices.
S3033: wind pressure prediction data is calculated.
Outputting the forward hidden layer of the forward GRU submodel calculated in the step S302 at the time tAnd the backward hidden layer output of the backward GRU submodel at the time t>Weighted summation is carried out, and hidden layer output H of the BiGRU network is obtained through calculation t ：
Wherein: gamma ray t The weight output by the forward hidden layer of the forward GRU submodel at the moment t is beta t The weight is output by a backward hidden layer of the backward GRU submodel at the time t; b t And (5) hiding the bias quantity corresponding to the layer state for the BiGRU network at the moment t.
Hidden layer output H of the BiGRU network t Namely wind pressure prediction data y output by the CNN-BiGRU neural network model t 。
S304: and (5) performing iterative training.
The wind pressure prediction data y obtained in step S3033 is used t And reversely transmitting the training data to the CNN-BiGRU neural network model for iterative training until the loss function of the CNN-BiGRU neural network model tends to be stable or reaches the maximum iteration times, and completing training to obtain a trained CNN-BiGRU neural network model. In this embodiment, the loss function uses a mean square error function, and the expression is:
wherein: s is S t Inputting an IMF component of a CNN-BiGRU neural network model at the moment t, y t The method is characterized in that wind pressure prediction data of the high-rise building to be predicted at the time t is obtained, and M is the number of iterations of model training.
S4: wind pressure missing data complement.
As shown in fig. 8, real-time wind pressure data of a high-rise building is collected, when the real-time wind pressure data is lost, the wind pressure data before the data loss and the wind pressure data around the wind pressure data are input into a trained CNN-biglu neural network model, wind pressure prediction data are obtained through the prediction of the CNN-biglu neural network model, and the lost data are complemented by the wind pressure prediction data.
In this embodiment, an EMD algorithm is adopted to decompose fluctuation and trend of wind pressure data with different scales in a wind pressure data sequence matrix step by step, so as to generate a series of data sequences with different characteristic scales, the data sequences are IMF components obtained by decomposition, then the IMF components and missing data measuring points are used as inputs of a CNN-biglu neural network model to perform model training, finally wind pressure prediction data is obtained, in the prediction of wind pressure data, the CNN-biglu neural network model can not only link surrounding wind pressure data influencing factors with wind pressure data prediction at the current moment in a spatial dimension, but also link past wind pressure data influencing factors and future wind pressure data influencing factors with wind pressure data prediction at the current moment in a time dimension, so that wind pressure prediction data is more accurate.
Example two
As shown in fig. 9, a system block diagram of the high-rise building long-term wind pressure missing data complement apparatus of the present embodiment is shown. The high-rise building long-term wind pressure missing data complement device of the embodiment comprises a data acquisition module 100, a data screening and integrating module 200, a data decomposition module 300, a model training module 400, a predicted data output module 500, a missing data complement module 600 and a display model 700, so as to complement the high-rise building long-term wind pressure missing data.
The data acquisition module 100 is used for acquiring wind pressure data of a high-rise building, wherein the wind pressure data comprises historical wind pressure data and real-time wind pressure data; the data acquisition module 100 is further configured to transmit the historical wind pressure data to the data screening and integration module 200 and transmit the real-time wind pressure data to the prediction data output module 500. In this embodiment, the data acquisition module 100 is preferably a wind pressure sensor disposed on an outer facade of a high-rise building.
The data screening and integrating module 200 is configured to determine missing data measurement points, select historical wind pressure data at a time before the missing data measurement points and surrounding wind pressure data of each historical wind pressure data, and integrate the historical wind pressure data and surrounding wind pressure data thereof into a wind pressure data sequence matrix according to a time sequence, and transmit the wind pressure data sequence matrix to the data decomposition module 300.
The data decomposition module 300 is configured to perform empirical mode decomposition on the wind pressure data sequence matrix output by the data screening and integration module, obtain a plurality of IMF components and one RES component, form an EMD decomposed data sample, and input the EMD decomposed sample data into the model training module 400 for model training.
The model training module 400 is configured to predict wind pressure data by using the EMD decomposition data sample obtained by the data decomposition module 300 and the historical wind pressure data measured by the missing data measurement point as input of the CNN-biglu neural network model, and reversely transfer the predicted wind pressure prediction data to the CNN-biglu neural network model for iterative training until the loss function of the CNN-biglu neural network model tends to be stable or reaches the maximum iteration number, i.e., stopping training to obtain the trained CNN-biglu neural network model. In this embodiment, the loss function uses a mean square error function.
The prediction data output module 500 is configured to input the real-time wind pressure data collected by the data collection module 100 into the trained CNN-biglu neural network model for prediction, and output the final wind pressure prediction data to the missing data complement module 600 and the display module 700.
The missing data complement module 600 is configured to complement the missing data by using the final wind pressure predicted data output by the predicted data output module 500 when the real-time wind pressure data is missing.
The display module 700 is configured to display the wind pressure prediction data output by the prediction data output module 500, so as to realize visualization of the wind pressure prediction data.
In this embodiment, the data screening and integrating module 200, the data decomposing module 300, the model training module 400, the predicted data output module 500, the missing data complementing module 600 and the display module 700 can be integrated on a computer, so that the structure is simple, and the processing requirements of the EMD decomposing algorithm and the CNN-BiGRU model on hardware and calculation are not high, so that the cost of the device is greatly reduced.
The foregoing is merely exemplary embodiments of the present invention, and specific structures and features that are well known in the art are not described in detail herein. It should be noted that modifications and improvements can be made by those skilled in the art without departing from the structure of the present invention, and these should also be considered as the scope of the present invention, which does not affect the effect of the implementation of the present invention and the practical applicability of the present invention.
Claims (9)
1. The method for supplementing the long-term wind pressure missing data of the high-rise building is characterized by comprising the following steps of:
s1: collecting wind pressure data of a high-rise building, determining missing data measuring points, selecting historical wind pressure data of the missing data measuring points and measured surrounding wind pressure data of surrounding data measuring points corresponding to each missing data measuring point, and integrating the historical wind pressure data and the surrounding wind pressure data thereof into a wind pressure data sequence matrix according to time sequence;
s2: adopting an empirical mode decomposition algorithm to decompose the wind pressure data sequence matrix into an IMF component and a RES component to obtain an EMD decomposed data sample;
s3: inputting the EMD decomposition data sample and the historical wind pressure data measured by the missing data measurement points into a CNN-BiGRU neural network model for iterative training, and optimizing model parameters to obtain a trained CNN-BiGRU neural network model;
s4: when the real-time wind pressure data is lost, inputting the wind pressure data before the data loss and the wind pressure data around the wind pressure data into a trained CNN-BiGRU neural network model, predicting by the CNN-BiGRU neural network model to obtain wind pressure prediction data, and complementing the lost data by using the wind pressure prediction data;
the step S3 includes the steps of:
s301: constructing a CNN-BiGRU neural network model comprising a CNN network and a BiGRU network, and inputting the EMD decomposition data sample obtained in the step S2 and the historical wind pressure data measured by the missing data measuring points into the constructed CNN-BiGRU neural network model;
s302: the CNN network extracts a space correlation characteristic sequence between surrounding wind pressure data and missing data measuring points based on a wind pressure data sequence matrix and the missing data measuring points;
s303: the BiGRU network takes the spatial correlation characteristic sequence extracted by the CNN network as input, further extracts the time front-back dependency characteristic between a wind pressure data sequence matrix and a missing data measuring point, and outputs wind pressure prediction data;
s304: and reversely transmitting wind pressure prediction data to the CNN-BiGRU neural network model for iterative training, and completing training until the loss function of the CNN-BiGRU neural network model tends to be stable or reaches the maximum iteration times, thereby obtaining a trained CNN-BiGRU neural network model.
2. The method for supplementing long-term wind pressure missing data of high-rise building according to claim 1, wherein in the step S2, the wind pressure data sequence matrix includes a historical wind pressure data sequence and surrounding wind pressure data sequences thereof, the historical wind pressure data sequence and surrounding wind pressure data sequences thereof are recorded as original wind pressure data sequences, and each original wind pressure data sequence matrix is subjected to empirical mode decomposition, and the specific steps are as follows:
s201: screening all maximum value points and minimum value points in the original wind pressure data sequence, and respectively fitting the maximum value points and the minimum value points to obtain an upper envelope curve and a lower envelope curve of the original wind pressure data sequence;
s202: calculating the average value envelope of the original wind pressure data sequence according to the upper envelope line and the lower envelope line, and calculating the difference value between the original wind pressure data sequence and the average value envelope to obtain a first wind pressure data sequence;
s203: judging whether the first wind pressure data sequence meets the condition that the IMF component is met, if yes, executing step S204, and if not, repeatedly executing steps S201-S202 on the first wind pressure data sequence until the ith wind pressure data sequence obtained after m times of repetition meets the condition that the IMF component is met;
s204: taking the ith wind pressure data sequence as a first IMF component of the original wind pressure data sequence, and separating the first IMF component from the original wind pressure data sequence to obtain a first residual wind pressure data sequence;
s205: and judging whether the first residual wind pressure data sequence is a monotonic function, if so, completing the decomposition, otherwise, repeatedly executing the steps S201-S204 on the first residual wind pressure data sequence until the j residual wind pressure data sequence obtained by repeating p times is a monotonic function, so as to decompose and obtain p IMF components and one RES component.
3. The method for supplementing long-term wind pressure missing data for high-rise buildings according to claim 2, wherein in the step S203, the condition for the IMF component to be satisfied is:
in the whole time interval, the number of maximum value points and minimum value points on the first wind pressure data sequence is equal to or at most different from the number of zero crossing points;
and the average value of the upper envelope curve and the lower envelope curve at any point on the first wind pressure data sequence is zero in the whole time interval.
4. The method for supplementing long-term wind pressure missing data of high-rise building according to claim 1, wherein in step S302, the CNN network comprises a convolution layer and a pooling layer, the convolution layer comprehensively learns local features of each historical wind pressure data sequence in the wind pressure data sequence matrix through convolution operation to obtain global features, and the pooling layer eliminates redundant feature information to extract spatial correlation feature sequences between surrounding wind pressure data and missing data measurement points.
5. The method for supplementing long-term wind pressure missing data for high-rise buildings according to claim 1, wherein in step S303, the CNN-biglu neural network model includes two GRU sub-models with identical structures, and the two GRU sub-models process the input IMF components in positive sequence and reverse sequence in time respectively, and the training process is as follows:
s3031: the spatial correlation characteristic sequences extracted in the step S302 are respectively input into two GRU sub-models, and a reset gate and an update gate of the two GRU sub-models are calculated;
s3032: calculating candidate activation states of the two GRU sub-models respectively, and calculating to obtain hidden layer output of the corresponding GRU sub-model according to the candidate activation states, the reset gate and the update gate;
s3033: and fusing the calculated hidden layer outputs of the two GRU submodels to obtain the time front-back dependency characteristics between the EMD decomposition data sample and the missing data measuring point, and further outputting a wind pressure predicted value.
6. The method for supplementing long-term wind pressure missing data of high-rise building according to claim 5, wherein in step S3033, the wind pressure prediction value y t The method comprises the following steps:
wherein:outputting a hidden layer which propagates forward for the GRU submodel at the moment t; />Outputting a hidden layer which is transmitted backward for another GRU submodel at the moment t; alpha t ,β t Weights output by hidden layers of forward propagation and backward propagation of the two GRU submodels at the moment t respectively; b t And (5) hiding the bias quantity corresponding to the layer state for the BiGRU network at the moment t.
7. The method for supplementing long-term wind pressure missing data for high-rise buildings according to claim 1, wherein in step S304, the loss function is a mean square error function, and the expression is:
wherein: x is x t Inputting the spatial correlation characteristic of the BiGRU network at the moment t, y t The method is characterized in that wind pressure prediction data of the high-rise building to be predicted at the time t is obtained, and N is the number of iterations of model training.
8. An apparatus for implementing the high-rise building long-term wind pressure missing data complement method according to any one of claims 1 to 7, comprising:
the data acquisition module is used for acquiring wind pressure data of the high-rise building, wherein the wind pressure data comprises historical wind pressure data and real-time wind pressure data;
the data screening and integrating module is used for determining missing data measuring points, selecting historical wind pressure data at the moment before the missing data measuring points and surrounding wind pressure data of each historical wind pressure data, and integrating the historical wind pressure data and the surrounding wind pressure data thereof into a wind pressure data sequence matrix according to a time sequence;
the data decomposition module is used for performing empirical mode decomposition on the wind pressure data sequence matrix output by the data screening and integrating module to obtain a plurality of IMF components and one RES component, so as to form an EMD decomposition data sample;
the model training module is used for inputting the EMD decomposition data sample obtained by the data decomposition module and the historical wind pressure data measured by the missing data measuring points into the CNN-BiGRU neural network model for iterative training to obtain a trained CNN-BiGRU neural network model;
the prediction data output module is used for outputting wind pressure prediction data; and
and the missing data complement module is used for complementing the missing wind pressure data by utilizing the wind pressure prediction data output by the prediction data output module when the real-time wind pressure data is missing.
9. The apparatus of claim 8, further comprising a display module for displaying the wind pressure prediction data output by the prediction data output module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111026123.9A CN113688770B (en) | 2021-09-02 | 2021-09-02 | Method and device for supplementing long-term wind pressure missing data of high-rise building |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111026123.9A CN113688770B (en) | 2021-09-02 | 2021-09-02 | Method and device for supplementing long-term wind pressure missing data of high-rise building |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113688770A CN113688770A (en) | 2021-11-23 |
| CN113688770B true CN113688770B (en) | 2024-02-06 |
Family
ID=78585056
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111026123.9A Active CN113688770B (en) | 2021-09-02 | 2021-09-02 | Method and device for supplementing long-term wind pressure missing data of high-rise building |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113688770B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116861347B (en) * | 2023-05-22 | 2024-06-11 | 青岛海洋地质研究所 | Magnetic force abnormal data calculation method based on deep learning model |
| CN117708625B (en) * | 2024-01-09 | 2024-06-18 | 水利部交通运输部国家能源局南京水利科学研究院 | Dam monitoring historical data filling method under spent data background |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106126896A (en) * | 2016-06-20 | 2016-11-16 | 中国地质大学（武汉） | The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth and system |
| CN106372731A (en) * | 2016-11-14 | 2017-02-01 | 中南大学 | Strong-wind high-speed railway along-the-line wind speed space network structure prediction method |
| CN106557840A (en) * | 2016-11-14 | 2017-04-05 | 中南大学 | A kind of high wind line of high-speed railway wind speed adaptive decomposition Forecasting Methodology |
| JP2019179014A (en) * | 2018-03-30 | 2019-10-17 | 株式会社熊谷組 | Wind state prediction method |
| CN111738512A (en) * | 2020-06-22 | 2020-10-02 | 昆明理工大学 | Short-term power load prediction method based on CNN-IPSO-GRU hybrid model |
| CN111754025A (en) * | 2020-05-25 | 2020-10-09 | 苏州大学文正学院 | Public transport short-time passenger flow prediction method based on CNN + GRU |
| CN112000084A (en) * | 2020-09-07 | 2020-11-27 | 华北电力大学 | Intelligent BIT design method of controller module based on 1D-CNN and GRU-SVM |
| CN112022125A (en) * | 2020-09-28 | 2020-12-04 | 无锡博智芯科技有限公司 | Intelligent blood pressure prediction method based on CNN-BiGRU model and PPG |
| CN112465225A (en) * | 2020-11-27 | 2021-03-09 | 云南电网有限责任公司电力科学研究院 | Wind power prediction method based on secondary modal decomposition and cascade deep learning |
-
2021
- 2021-09-02 CN CN202111026123.9A patent/CN113688770B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106126896A (en) * | 2016-06-20 | 2016-11-16 | 中国地质大学（武汉） | The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth and system |
| CN106372731A (en) * | 2016-11-14 | 2017-02-01 | 中南大学 | Strong-wind high-speed railway along-the-line wind speed space network structure prediction method |
| CN106557840A (en) * | 2016-11-14 | 2017-04-05 | 中南大学 | A kind of high wind line of high-speed railway wind speed adaptive decomposition Forecasting Methodology |
| JP2019179014A (en) * | 2018-03-30 | 2019-10-17 | 株式会社熊谷組 | Wind state prediction method |
| CN111754025A (en) * | 2020-05-25 | 2020-10-09 | 苏州大学文正学院 | Public transport short-time passenger flow prediction method based on CNN + GRU |
| CN111738512A (en) * | 2020-06-22 | 2020-10-02 | 昆明理工大学 | Short-term power load prediction method based on CNN-IPSO-GRU hybrid model |
| CN112000084A (en) * | 2020-09-07 | 2020-11-27 | 华北电力大学 | Intelligent BIT design method of controller module based on 1D-CNN and GRU-SVM |
| CN112022125A (en) * | 2020-09-28 | 2020-12-04 | 无锡博智芯科技有限公司 | Intelligent blood pressure prediction method based on CNN-BiGRU model and PPG |
| CN112465225A (en) * | 2020-11-27 | 2021-03-09 | 云南电网有限责任公司电力科学研究院 | Wind power prediction method based on secondary modal decomposition and cascade deep learning |
Non-Patent Citations (5)
| Title |
|---|
| Multi-Model Fusion Short-Term Load Forecasting Based on Random Forest Feature Selection and Hybrid Neural Network;YI XUAN 等;《IEEE Access》;第9卷;69002-69009 * |
| 利用时空相关性的多位置多步风速预测模型;陈金富 等;《中国电机工程学报》;第39卷(第7期);第2093-2105、S25页，摘要，第2-5节 * |
| 基于二次模式分解和级联式深度学习的超短期风电功率预测;殷豪 等;《电网技术》;第44卷(第2期);445-453 * |
| 基于机器学习方法的风荷载预测研究;黄佳星;《中国优秀硕士学位论文全文数据库 工程科技II辑》;第C038-249页，正文第32-36、45页 * |
| 融合多源信息的短期风电功率预测方法研究;马吉科 等;《计算机仿真》;第37卷(第07期);137-143 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113688770A (en) | 2021-11-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111130839B (en) | Flow demand matrix prediction method and system | |
| CN109033450B (en) | Elevator equipment fault prediction method based on deep learning | |
| CN111832216B (en) | Rolling bearing residual service life prediction method based on EEMD-MCNN-GRU | |
| CN109934337B (en) | Spacecraft telemetry data anomaly detection method based on integrated LSTM | |
| CN113688770B (en) | Method and device for supplementing long-term wind pressure missing data of high-rise building | |
| CN109828552B (en) | Intermittent process fault monitoring and diagnosing method based on width learning system | |
| CN111461187B (en) | Intelligent building settlement detection system | |
| CN114358192B (en) | Multi-source heterogeneous landslide data monitoring and fusing method | |
| CN115673596B (en) | Welding abnormity real-time diagnosis method based on Actor-Critic reinforcement learning model | |
| CN114897277B (en) | Burst-type landslide displacement prediction method based on LSTM | |
| CN114266201A (en) | Self-attention elevator trapping prediction method based on deep learning | |
| CN113988210A (en) | Method and device for restoring distorted data of structure monitoring sensor network and storage medium | |
| CN113722308B (en) | Acceleration response data complement method and device based on EEMD-MultiCNN-LSTM | |
| CN113537638A (en) | Short-term wind pressure prediction method and abnormal data completion method and device for high-rise building | |
| CN115062764B (en) | Intelligent illuminance adjustment and environmental parameter Internet of things big data system | |
| CN113779724B (en) | Intelligent fault prediction method and system for filling packaging machine | |
| RU2600099C1 (en) | Method of neural network forecasting of change of values of function with its complementary wavelet processing and device for its implementation | |
| CN117010683A (en) | Operation safety risk prediction method based on hybrid neural network and multiple agents | |
| CN114358244B (en) | Big data intelligent detection system of pressure based on thing networking | |
| CN111473768B (en) | Building safety intelligent detection system | |
| CN114841063A (en) | Aero-engine residual life prediction method based on deep learning | |
| CN111104298A (en) | LSTM-based power grid server running state prediction device | |
| CN113688771B (en) | LNG storage tank acceleration response data supplementing method and device | |
| CN117349770B (en) | Structural health monitoring multi-strain sensor data anomaly detection and repair method | |
| CN117851802A (en) | Water quality prediction method and device and computer readable storage medium |
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 |