CN115659609A - DTW-DCRNN-based chemical industry park noise prediction method - Google Patents
DTW-DCRNN-based chemical industry park noise prediction method Download PDFInfo
- Publication number
- CN115659609A CN115659609A CN202211238200.1A CN202211238200A CN115659609A CN 115659609 A CN115659609 A CN 115659609A CN 202211238200 A CN202211238200 A CN 202211238200A CN 115659609 A CN115659609 A CN 115659609A
- Authority
- CN
- China
- Prior art keywords
- matrix
- noise
- sequence
- prediction
- time
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a DTW-DCRNN-based chemical industry park noise prediction method, belongs to the field of signal and information processing, and solves the problem of long-term global noise prediction of a chemical industry park. And (3) introducing a time dynamic regularization theory into the diffusion convolution recurrent neural network, and establishing a space-time noise prediction network model consisting of a time dynamic regularization method, the diffusion convolution recurrent neural network and Kalman filtering. Improving a DTW algorithm by using a penalty coefficient, and reconstructing the spatial relationship of the neural network through a DTW model; the output noise prediction of the neural network is dynamically adjusted using a Kalman method in conjunction with the traffic flow characteristics. According to the method, a spatial relationship is reconstructed based on time sequence similarity, and then data of each station are sent into a model for prediction and dynamic correction, so that multi-station noise prediction, long-term noise prediction and global noise level prediction are realized, and risks of noise disturbing residents and health damage can be avoided in advance.
Description
Technical Field
The invention belongs to the field of signal and information processing, and particularly relates to a DTW-DCRNN-based chemical industry park noise prediction method.
Background
With the continuous development of industry, the environmental noise pollution problem becomes more severe along with the rapid development of urban society and economy, and the main problems of noise pollution prevention and control work are more and more obvious. However, the research of the neural network is continuously advanced, so a method based on the neural network is needed to predict the environmental noise for a long time.
The rise of the concept of the intelligent chemical industry park lays a good foundation for the accurate and effective collection of various data. The construction of the intelligent garden of the Dongming engineering plastic industrial park is relatively perfect, various monitoring data are complete, the data volume is sufficient, and a good data basis is provided for developing noise prediction research of the chemical industry park. The data are modeled, analyzed and predicted based on a deep learning method, and the noise data rule is mined, so that a park manager can be helped to take preventive measures in advance in a period with high noise decibel, such as: the park operation workers wear noise reduction earphones in a specified time period, set up noise baffles in advance or place equipment with overhigh production noise away from residential areas and the like. On one hand, the work can provide a relatively safe working environment for operators through prediction reminding and early warning analysis, and on the other hand, the influence of noise on the life of residents nearby a park can be reduced. In addition, the noise levels of different positions in the park are different, and the method has very important significance for the overall prediction of environmental noise on the assessment of the professional risks and the prediction of the environmental risks in the park.
Disclosure of Invention
Based on the problems, the invention provides a DTW-DCRNN-based chemical industry park noise prediction method, and a space-time noise prediction network model formed by an improved time dynamic regularization algorithm, a diffusion convolution recurrent neural network and Kalman filtering is constructed from the perspective of space-time prediction so as to carry out long-term and global prediction on the noise of the chemical industry park.
In order to achieve the purpose, the invention adopts the following technical scheme:
a DTW-DCRNN-based chemical industry park noise prediction method adopts an improved time dynamic warping algorithm and reconstructs a spatial relationship through a graph structure of a monitoring station to realize long-term space-time prediction of the chemical industry park noise, and specifically comprises the following steps:
and 4, acquiring information of the monitoring station in real time and predicting noise in real time based on the trained model.
Further, in the step 1, a gateway and a plurality of monitoring sites are arranged in the chemical industry park for collecting data in real time and synchronously recording position information data of each monitoring site; sensors are arranged at the gateway and each monitoring station; the gateway collects information data of vehicles entering and leaving, including license plate number, vehicle entering and leaving time and vehicle flow characteristics of vehicle type; the monitoring station collects noise and natural environment data in a chemical industry park in real time;
and the station position data, the noise and natural environment data and the in-out vehicle information data are transmitted to the gateway equipment and then are interactively transmitted with the database server, and when noise prediction is carried out, data extraction is carried out in real time and 3 sigma criterion is adopted for processing.
Further, the specific process of step 2 is as follows:
step 2.1, a penalty coefficient is introduced to improve an original time dynamic warping algorithm, and a sequence similarity distance is calculated;
the principle of the improved time dynamic regularization algorithm is as follows: obtaining the number of the optimal paths and the public subsequences to calculate a penalty coefficient, and calculating the similarity distance between stations by combining the penalty coefficient; the specific process is as follows:
step 2.1.1, calculating a distance matrix;
two noise sequences S and T are assumed, as in equation (1):
the length of the sequence S is o, the length of the sequence T is m, o, m belongs to Z + Constructing a distance matrix d [ s ] based on the noise sequence mxo][t]Wherein the ith sequence point s i (i =1,2, \8230;, o) and the jth sequence point t j (j =1,2, \8230;, m) is a matrix(s) i ,t j ) Value of element, euclidean distance: dis(s) i ,t j )=(s i -t j ) 2 The matrix distance calculation criterion is shown in formula (2):
step 2.1.2, solving a matrix optimal path;
the optimal path is the path with the minimum accumulation distance meeting the lower boundary condition, and the lower boundary condition is as follows: starting from the d [ o ] [ m ] point at the upper right corner of the matrix, finding the minimum point in the three lower left points as the next node until the d [0] [0] point at the lower left corner is finished, ensuring continuity and monotonicity, not crossing or omitting a certain point for matching, only aligning or matching with adjacent points, and unchangeable time sequence;
step 2.1.3, calculating the number of public subsequences and the optimal path sequence length, and further obtaining weight to obtain a penalty coefficient;
wherein, the optimal path sequence length is the length of the optimal path of the matrix finally obtained in the step 2.1.2;
the calculation process of the number of the common subsequences is as follows: firstly, constructing a recording empty list record, circularly traversing sequences s1 and s2, recording the length ls1 or ls2 of a common sequence and the sequence position g of the common sequence by using the record list after the common sequence appears, and extracting the common subsequence from the s1 sequence and storing the common subsequence in an s _ sum array; the length of s _ sum after traversal is the number of the public subsequences; when calculating the number of the public subsequences, setting a threshold value as 1, calculating the subsequences with the length greater than 1 as the public subsequences, and finally counting the total number of the public subsequences to obtain the required number of the public noise subsequences;
the calculation formula of the weight w is as follows (3):
wherein, subseq is the length of the public noise subsequence, and seq is the sequence length of the optimal path of the matrix;
the calculation process for setting the penalty coefficient alpha is shown as formula (4), wherein x represents the number of public subsequences, and w i Represents the weight of the ith sequence point,
step 2.1.4, express the optimal path as (r) 1 ,r 2 ,…,r seq ),r i (i =1,2, \8230;, seq) represents the value of the ith sequence point in the optimal path sequence, and finally the sum of the values in the optimal path calculated by the original DTW is multiplied by a penalty coefficient to obtain the improved sequence similarity distance dis DTW ;
Step 2.2, forming a matrix by the noise sequence similarity distance among all the monitored sites, calculating the matrix into an adjacent matrix, and constructing a graph relation topological structure; the specific process is as follows:
step 2.2.1, according to the noise sequence similarity distance information of each monitored site, calculating a similarity distance matrix M between the sites, wherein the matrix M is as follows:
wherein a is the number of monitoring stations;
step 2.2.2, calculating the standard deviation lambda, M of the similarity distance matrix M c Is the c-th element value in the matrix, N is the number of elements in the M matrix, mu M For the matrix mean, the calculation process is as follows:
step 2.2.3, analyzing the similarity between the monitoring stations according to the similar distance matrix M, and constructing an adjacent matrix M by using the standard deviation lambda of all non-infinite numbers in the similar distance matrix d Of a contiguous matrix M d Each element in (1)The calculation process of (2) is as follows:
wherein c represents the sequence number of the matrix element, and the larger the similarity distance is, the larger the sequence number isThe smaller the threshold is set to 0.1, ifIf the similarity between the two sites is lower than the threshold value, the two sites have no mutual influence relationship, the two sites are regarded as non-adjacent sites, and the adjacent relationship is not formed when the weight value in the adjacent matrix is zero;
step 2.3, introducing a graph convolution neural network GCN to construct an adjacent matrix graph structure, and inputting the adjacent matrix graph structure into a diffusion convolution recurrent neural network for noise prediction to obtain a preliminary prediction result; the specific process is as follows:
the propagation mode between the GCN network layers of the graph convolution neural network is shown as formula (12), wherein L represents the Lth network layer:
wherein σ is a nonlinear activation function; w L Is a matrix of trainable weights that is trained to,a degree matrix, which is composed of:i denotes the I th row of the adjacency matrix A, J denotes the J th column of the adjacency matrix A,the characteristic information of the station is kept, H is the characteristic extracted by the current layer, and if the characteristic is the input layer, X = H;
on the basis of the GCN graph convolution neural network, a time-space relation of a noise sequence is modeled by using a diffusion convolution recursive neural network;
the smooth distribution of the diffusion process is represented as a weighted combination of infinite random walk on the graph, the diffusion process is represented by equation (13), and is calculated in closed form:
wherein W is a node similarity matrix,is the inverse of the rectangular output matrix, with beta ∈ [0,1 ]]Representing restart probability, k is diffusion degree, epsilon represents the possibility of diffusion from nodes, and the diffusion process in the DCRNN model is bidirectional;
thus, in a spatial relationship, based on the graph signal feature matrixAnd filter f θ The bi-diffusion convolution operation is defined as equation (14):
where θ is the filter parameter, G represents graph G, P represents the pth feature dimension, P represents the total number of feature dimensions, K represents the finite K step truncation of the diffusion process, and: (X:), p representing the convolution of all nodes with the p-th feature, θ k,1 The parameter of the convolution kernel, theta, representing the degree calculation k,2 Represents the parameters of the convolution kernel for the in-degree calculation,andrespectively representing transition matrixes of diffusion and inverse diffusion, and constructing a diffusion convolution layer with a mapping relation after defining convolution operation as shown in a formula (15):
wherein Q represents the qth feature, Q represents the total number of mapped features, H, q representing the q characteristic diffusion convolution operation of all the node pairs and a characteristic matrixFeatures extracted from the current layer as inputIn order to be output, the output is,representing a filter, wherein theta is a parameter tensor, and sigma is an activation function;
in the time relation, a diffusion convolution gating recursion unit DCGRU is constructed by combining GRU and diffusion convolution, and is expressed as an expression (16):
≧ G denotes a diffusion convolution of graph G, r (t) Indicating the reset gate at time t, [ theta ] r★G Tensor of parameter, X, representing reset gate (t) Indicating an input at time t, H (t-1) Representing the output at time t-1, b r A bias vector representing a reset gate; u. of (t) Indicating the update Gate at time t, Θ u★G Tensor of parameters representing the update gate, b u A bias vector representing an update gate; c (t) Indicates the hidden state of the next cell at time t, [ theta ] C★G Tensor of parameters representing hidden states, b C A bias vector representing a hidden state; h (t) Represents the output at time t;
step 2.4, extracting traffic flow characteristics at a road gate corresponding to the current time in real time, and dynamically adjusting a DCRNN noise predicted value by using a Kalman filtering method;
kalman filtering uses the estimate of the last state to make a prediction of the current state; finally, correcting the predicted value obtained in the prediction stage by utilizing the observed value of the current state to obtain a new estimated value which is closer to the true value; the method specifically comprises the steps of calculating Kalman gains of noise and drift by using normalized traffic flow characteristics and noise sequences through prior estimation and prediction of a covariance matrix, updating the covariance matrix by using Kalman gain correction, and then correcting and predicting the noise value at the current moment.
Further, the specific process of step 3 is as follows:
extracting historical data of each monitoring station for reconstructing the spatio-temporal relationship of data by an improved DTW algorithm, training a noise prediction network model DTW-DCRNN, selecting 70% of a data set as a training set, 10% of the data set as a verification set, and finally 20% of the data set as a test set; in addition, the training results are not recorded in the first 60 epochs, the training results are recorded once every 10 epochs in the 60-100 epochs and the model parameters are output, and the training results are recorded once every 5 epochs after 100 epochs and the model parameters are output; and finally stopping training in advance according to the measured verification loss to ensure that the model is captured when the model is about to be overfitting.
The invention has the following beneficial technical effects:
a penalty coefficient is introduced to improve an original time dynamic rule algorithm, and the limitation of an original physical distance method can be well made up by constructing an adjacent matrix relation through a similar distance matrix to represent a spatial distance relation; the prediction result adjusted by the Kalman filter is closer to a true value, and the compensation property is realized, so that the 3 sigma noise processing method effectively avoids the condition of large deviation of the model training prediction result. The time-space noise prediction network model established by the invention and composed of a time dynamic warping method, a diffusion convolution recurrent neural network and Kalman filtering can realize multi-site noise prediction, long-term noise prediction and global noise level prediction, thereby avoiding the risks of noise disturbing residents and health damage in advance.
Drawings
FIG. 1 is a flow chart of a DTW-DCRNN-based chemical industry park noise prediction method according to the present invention;
FIG. 2 is a schematic view of a data collection process of the present invention;
FIG. 3 is a schematic diagram of a noise prediction deep learning network model of the present invention;
FIG. 4 is a schematic diagram of the improved DTW algorithm of the present invention;
FIG. 5 is a flow chart of the number of common subsequences calculation of the present invention;
FIG. 6 is a schematic structural diagram of a diffusion convolution recurrent neural network model of the present invention;
FIG. 7 is a graph showing a comparison of the change in the adjacency matrix in experiment 1 according to the present invention; wherein, (a) shows the adjacency matrix constructed by the original DTW algorithm, and (b) shows the adjacency matrix constructed by the improved DTW algorithm;
FIG. 8 is a graph of the index variation trend based on time step in experiment 1 according to the present invention;
FIG. 9 is a graph comparing the performance indexes MAE and RMSE of the prediction model and the DCRNN model in experiment 2;
FIG. 10 is a graph comparing the long-term performance index MAPE of the prediction model and the DCRNN model in experiment 2.
Detailed Description
The invention is described in further detail below with reference to the following figures and detailed description:
the invention mainly solves the problems existing in the noise prediction of the chemical industrial park from the following four aspects: the noise data has zero values, which greatly interfere with the prediction precision, so that for the problem of processing the noise sequence monitoring data, a 3 sigma criterion is selected to remove the zero values, the zero values are filtered, the standard deviation of the sample unbiased ratio is reduced, and the prediction error is further reduced. The noise prediction of a single monitoring station is not associated with other stations, and the weight cannot be shared, so that the problem that data cannot be communicated between single station predictions and the consumption of the individual prediction calculation performance of each station is high is solved, and the data information prediction weight sharing between stations is realized through a graph neural network. At present, the noise similarity between non-adjacent monitoring stations does not participate in calculation, so that the problem of station similarity calculation between adjacent noise monitoring stations and between non-adjacent stations is faced, and the time dynamic warping DTW algorithm is used for reconstructing the spatial relationship from the time sequence similarity. In the face of noise additive property, the factors with high influence correlation on noise can not play a role in space-time prediction, and Kalman filtering can be combined with other influence factors to adjust and correct the prediction result, so the prediction result is dynamically adjusted according to the characteristics of the correlation factors by adopting a Kalman method.
The concrete expression is as follows: and introducing a time dynamic regularization theory into the DCRNN, and establishing a space-time noise prediction network model formed by a time dynamic regularization method, the DCRNN and the Kalman filtering. Improving a DTW algorithm by using a penalty coefficient, and reconstructing the spatial relationship of the neural network through a DTW model; and (3) dynamically adjusting the output noise prediction of the neural network by using a Kalman method in combination with the traffic flow characteristics. According to the method, a spatial relationship is reconstructed based on time series similarity, and then data of each station are sent into a model for prediction and dynamic correction, so that multi-station noise prediction, long-term noise prediction and global noise level prediction are realized, and the risks of noise disturbing residents and health damage can be avoided in advance.
Based on the characteristics of the DTW algorithm and the DCRNN model, the space-time noise prediction network model provided by the invention makes up the limitation of building a spatial relationship based on distance, can effectively utilize the advantages of park position information and two methods, and realizes high-precision noise prediction.
As shown in fig. 1, a DTW-DCRNN-based chemical industry park noise prediction method specifically includes the following steps:
A road gate and a plurality of monitoring stations are arranged in the chemical industry park for acquiring data in real time and synchronously recording position information data of each monitoring station; sensors are arranged at the gateway and each monitoring station; the gateway collects information data of vehicles entering and leaving, including license plate number, vehicle entering and leaving time, vehicle type and other traffic flow characteristics; the monitoring station collects noise and natural environment data in a chemical industry park in real time;
as shown in fig. 2, the station position data, the noise and natural environment data, and the in-out vehicle information data are transmitted to the gateway device, and then are interactively transmitted with the database server, and when noise prediction is performed, data extraction is performed in real time, and processing is performed by adopting a 3 σ criterion.
The 3 sigma criterion is that a group of data is supposed to only contain random error, standard deviation is calculated, an interval is determined according to probability, the error exceeding the interval is considered not to belong to the random error but to be coarse, the data containing the error is removed or replaced, and the noise value noise belongs to the element of the U-3 sigma Noise reduction ,u+3σ Noise making device ) The ratio of the intervals is about 99.74%, u represents the noise mean, σ Noise reduction Noise is the standard deviation of the noise, noise is the noise value. The noise range is more than or equal to 0 and less than u-3 sigma Noise reduction (dB) is replaced by the mean.
The original noise sequence collected by the sensor contains a sparse mutation point zero value. The noise value processed by the 3 sigma criterion has the advantages that the sample has no deviation standard deviation reduction and mutation value reduction, and the noise prediction accuracy of the neural network can be further improved.
And 2, introducing a penalty coefficient to improve an original time dynamic warping algorithm, introducing the original time dynamic warping algorithm into a diffusion convolution recurrent neural network, and constructing a noise prediction network model by combining Kalman filtering.
As shown in fig. 3, the principle of the noise prediction network model is as follows: firstly, calculating an optimal path by adopting an original time dynamic warping algorithm based on extracted data of each monitoring station, then improving the original time dynamic warping algorithm by introducing a penalty coefficient, calculating sequence similarity distance by the improved time dynamic warping algorithm, forming each monitoring station into a matrix, and further constructing an adjacent matrix graph structure; then inputting the adjacency matrix into a diffusion convolution recurrent neural network for noise preliminary prediction, and finally combining the traffic flow characteristics and Kalman filtering for dynamic adjustment of a noise prediction result to obtain a final noise prediction result; the specific process is as follows:
step 2.1, introducing a penalty coefficient to improve an original time dynamic warping algorithm, and calculating a sequence similarity distance;
the principle of the improved time dynamic warping algorithm is as follows: and solving the number of the optimal paths and the public subsequences to calculate a penalty coefficient, and calculating the similarity distance between stations by combining the penalty coefficient.
The main work flow is as shown in fig. 4, and the specific process is as follows:
step 2.1.1, calculating a distance matrix;
two noise sequences S and T are assumed, as in equation (1):
the length of the sequence S is o, the length of the sequence T is m, o, m belongs to Z + The distance can be directly calculated under the condition that the coincidence degree of wave crests and wave troughs of the traditional Euclidean distance is high, but the Euclidean distance cannot be used for calculating periodic sequences of different phases, so that the distance matrix d [ s ] based on the noise sequence m multiplied by o is constructed by the method][t]Wherein the ith sequence point s i (i =1,2, \8230;, o) and the jth sequence point t j (j =1,2, \8230;, m) is a matrix(s) i ,t j ) Value of the element, i.e. Euclidean distanceSeparation: dis(s) i ,t j )=(s i -t j ) 2 The matrix distance calculation criterion is shown in equation (2):
step 2.1.2, solving a matrix optimal path;
the rows of the matrix in the distance matrix calculated in 2.1.1 above represent the sequence S, and the columns of the matrix are the sequence T. The rows and columns correspond to the sequence one by one, but a certain time point of the S sequence corresponds to the points of a plurality of times in the T sequence, and the optimal path of the matrix is obtained through dynamic regulation. The selection of the path needs to satisfy the following boundary conditions: it must meet the requirement that starting from the d [ o ] [ m ] point at the upper right corner of the matrix, the minimum point of the three lower left points is found as the next node until the d [0] [0] point at the lower left corner is finished, and besides, the continuity and monotonicity are ensured, the matching can not be carried out by crossing or omitting a certain point, the matching can only be aligned or matched with the adjacent point, and the sequence of the time sequence can not be changed. The optimal path is a path in which the cumulative distance satisfying the above conditions is the smallest.
The single DTW algorithm for solving the distance is not friendly to the non-aligned periodic sequence, and the calculated distance is not ideal. Therefore, improvement is needed on the basis of the DTW algorithm. The improvement idea is that firstly, a distance matrix is generated by using a DTW algorithm and an optimal path is solved: taking the point (i, j) as an example to find the minimum value of the three lower left points, and if the value is the lower left diagonal value, (i-1, j-1) is the next node; if the value is the left neighbor, then the next node is (i-1, j); if the value is the next node, then (i, j-1) is the next node until there is no next node. The obtained optimal path provides a verification basis for solving a public subsequence of the noise sequence later.
Step 2.1.3, calculating the number of public subsequences and the optimal path sequence length, and further obtaining weight to obtain a penalty coefficient;
and (3) calculating the length of the optimal path of the matrix finally obtained in the step (2.1.2) to be the required optimal path sequence length.
The noise common subsequence number calculation implementation steps are as shown in fig. 5, firstly, a record empty list (record) is constructed, sequences s1 and s2 are traversed circularly, after a common sequence appears, the record list is used for recording the length ls1 or ls2 of the common sequence and the sequence position g, and the common subsequence is extracted from the s1 sequence and is placed in an s _ sum array for storage. The length of s _ sum after the traversal is completed is the number of the common subsequences. When the number of the public subsequences is calculated, the threshold value is set to be 1, namely the subsequences with the length larger than 1 can be calculated as the public subsequences, and finally the total number of the counted public subsequences is the number of the required public noise subsequences.
The calculation method is shown in an algorithm 1:
using all the public noise subsequence numbers to calculate a penalty coefficient;
first, the weight w is calculated, and the calculation formula is as follows (3):
wherein, subseq is the length of the public noise subsequence, and seq is the sequence length of the optimal path of the matrix;
then, the calculation process of setting the penalty coefficient alpha is as formula (4), wherein x represents the number of common subsequences, and w is i And the weight of the ith sequence point is represented, and the standard that the longer the optimal path sequence length is, the greater the number of public noise subsequences is, and the longer the length is, the smaller the penalty coefficient is satisfied:
step 2.1.4, express the optimal path as (r) 1 ,r 2 ,…,r seq ),r i (i =1,2, \8230;, seq) represents the value of the ith sequence point in the optimal path sequence, and the optimal path number obtained by the calculation of a penalty coefficient and the original DTW is finally usedMultiplying the sum of the values to obtain the improved sequence similarity distance dis DTW 。
And 2.2, forming a matrix by the noise sequence similarity distance among the monitoring stations, calculating the matrix as an adjacent matrix, and constructing a graph relation topological structure. The specific process is as follows:
step 2.2.1, according to the noise sequence similarity distance information of each monitored station, calculating a similarity distance matrix M (numerical unit: km) between stations, wherein the matrix M is as follows:
wherein a is the number of monitoring stations;
for example, 11 monitoring sites are arranged in the chemical industry park according to the embodiment of the present invention, and a matrix formed by each site is as follows.
Step 2.2.2, calculating the standard deviation lambda, M of the similarity distance matrix M c Is the c-th element value in the matrix, N is the number of elements in the M matrix, mu M For matrix mean, the calculation process is as follows:
step 2.2.3, analyzing the similarity between the monitoring stations according to the similar distance matrix M, and constructing an adjacent matrix M by using the standard deviation lambda of all non-infinite numbers in the similar distance matrix d Of a contiguous matrix M d Each element in (1)The calculation process of (2) is as follows:
wherein c represents the sequence number of the matrix element, and the larger the similarity distance is, the larger the similarity distance isThe smaller the threshold is set to 0.1, ifIf the similarity between the two sites is less than the threshold value, the similarity between the two sites is considered to be too low, no mutual influence relation exists, the two sites are considered to be non-adjacent sites, and the adjacent relation is not formed when the weight value in the adjacent matrix is zero.
The steps of the method meet the requirement of similarity calculation of the periodic noise time sequence. The method is used for reconstructing the spatial relationship, and lays a foundation for providing a more complete graph representation relationship for the subsequent graph convolution neural network work.
Step 2.3, introducing a graph convolution neural network GCN to construct an adjacent matrix graph structure, and inputting the adjacent matrix graph structure into a diffusion convolution recurrent neural network for noise prediction to obtain a preliminary prediction result;
first, the convolution is understood from the perspective of signal processing, the word "Convolve" is intended to mean inversion, and is translated into "convolution" in the convolution, and the "convolution" itself is an operation manner, meaning "product", and noise can be classified as discrete noise or continuous noise, the convolution generally refers to a continuous signal, the monitoring means of sound is realized through a sensor, and the noise collected by the sensor is a discrete value at an interval of 30s, so that only the discrete noise convolution sum is discussed here.
Defining the pulse signal as:n denotes the number of bits shifted, l denotes the discrete time, the discrete noise x n]Can be represented by formula (9):
wherein x [ l ] represents the discrete noise at time l;
the discrete noise convolution process can be understood as inverting the noise sequence, shifting, and finally multiplying and summing. Meaning the unit impulse response h n through the system]To characterize the discrete noise sequence x n of the input pair of the system]In response to the convolution sum y of the discrete noise convolution dis [n]Is represented by the formula (10).
During convolution, firstly inverting the unit impulse response to obtain h < -l >, then shifting n to obtain a shifted function h < n < -l >;
and the definition of convolution in the Convolutional Neural Network (CNN) is compared with the above-mentioned convolution principle and lacks the process of inversion, and the direct displacement of n can obtain h [ n + l [ ]]Convolution sum y of convolutional neural network cnn [n]Represented by the formula (11),
it is not the convolution sum in full meaning but the cross-correlation in signal processing, so the convolution kernel is also a filter. The fundamental purpose of using convolution in a convolutional neural network is to weight the sums and extract features, and flipping is not necessary.
The noise sequences of a plurality of monitored sites can form a topological structure diagram which is irregular and has no translation invariance, so that a diagram convolution neural network GCN is required to be introduced to solve the diagram relation problem. In the embodiment of the invention, eleven monitoring sites are used as eleven nodes, each site has characteristics, all the nodes form an adjacent matrix A with 11 × 11 dimensions, the noise sequence length is Y, the characteristics of the nodes are a characteristic matrix X with 11 × Y dimensions, and the X and A matrixes are input into the graph convolution network. The propagation mode between the GCN network layers of the graph convolution neural network is shown as formula (12), wherein L represents the Lth network layer:
wherein σ is a nonlinear activation function; w is a group of L Is a matrix of trainable weights that is trained to,a degree matrix, which is composed of:i denotes the I th row of the adjacency matrix A, J denotes the J th column of the adjacency matrix A,indicating characteristic information of the reservation station itself, I 11 And an identity matrix of 11 dimensions is represented, λ is a site noise feature weight, λ =1 represents that the noise feature of the site is as important as the noise feature of an adjacent site, H is a feature extracted by the current layer, and X = H if the input layer is used.
The features extracted by the GCN using the random initialization parameters are very excellent compared with the CNN model if the features extracted by the CNN model without training are very limited. The use of GCN to extract spatial relationships is the most preferred of the current research methods.
In order to formalize the space-time noise sequence prediction problem, on the basis of a GCN graph convolution neural network, the invention uses a diffusion convolution recursive neural network to model the space-time relation of a noise sequence. By associating noise monitoring sites with the diffusion process, modeling the spatial relationship that exists between the sites, it is also possible to capture the stochastic nature of the dynamics of the noise. The structure of a DCRNN model of a diffusion convolution recurrent neural network is shown in figure 6, the DCRNN is a recurrent neural network based on diffusion graph convolution, the original input of the model is a space-time noise data structure based on combination of an adjacent matrix of physical distances distributed by monitored sites and time series noise data, the improved DTW algorithm is provided for calculating the similar distances of the sequences, and the space-time relationship between the sites is reconstructed based on the similar distances. The model has the advantages that the diffusion convolution can take the wandering relation among noises into consideration, an encoder is added on a historical noise sequence, and the actual value or the predicted value at the previous moment is delayed by one time step to be used as the input of a current decoder for noise prediction at the current time. Where ReLU is the activation function, it will output directly if the input is positive, otherwise it will output zero. The smooth distribution of the diffusion process can be represented as a weighted combination of infinite random walks on the graph, the diffusion process can be represented as equation (13) and calculated in a closed form:
w is the node similarity matrix and,is the inverse of the rectangular output matrix, with beta ∈ [0,1 ]]And representing restart probability, wherein k is diffusion degree, epsilon represents the possibility of diffusion from the node, and the diffusion process in the DCRNN model is bidirectional, so that the environmental influence of the two sides of the noise monitoring station can be considered.
Thus, in the spatial relationship, based on the map signal feature matrixAnd filter f θ (θ is a filter parameter) the bi-diffusion convolution operation can be defined as equation (14):
wherein G represents graph G, P represents the pth feature dimension, P represents the total number of feature dimensions, K represents the finite K step truncation of the diffusion process,:, p represents the convolution operation of all nodes on the p-th feature, theta k,1 The parameter of the convolution kernel, theta, representing the degree calculation k,2 Represents the parameters of the convolution kernel for the in-degree calculation,and withRespectively representing transition matrixes of diffusion and inverse diffusion, defining convolution operation, and constructing a diffusion convolution layer (P dimension to Q dimension) with a mapping relation as shown in a formula (15):
wherein Q represents the qth feature, Q represents the total number of mapped features, H, q representing the q characteristic diffusion convolution operation of all the node pairs and a characteristic matrixFeatures extracted from the current layer as inputIn order to be output, the output is,the filter is represented (Θ is the parameter tensor) and σ is the activation function. And completing the construction of the spatial relationship based on the bidirectional diffusion convolution.
In the time relation, a diffusion convolution gate control recursive unit (DCGRU) is constructed by combining GRU and diffusion convolution, the essential principle of the DCGRU is to replace matrix multiplication in the GRU by the diffusion convolution, and the DCGRU can be expressed as a formula (16) by combining the GRU and the diffusion convolution on the basis of an LSTM principle:
∑ G represents a diffusion convolution of graph G, r is a radical of hydrogen (t) Denotes the reset gate at time t, theta r★G Tensor of parameter, X, representing reset gate (t) Indicating input at time t, H (t-1) Representing the output at time t-1, b r A bias vector representing a reset gate; u. of (t) Represents the update Gate at time t, Θ u★G Tensor of parameters representing the update gate, b u A bias vector representing an update gate; c (t) Indicates the hidden state of the next cell at time t, [ theta ] C★G Tensor of parameters representing hidden states, b C A bias vector representing a hidden state; h (t) Indicating the output at time t.
Through space-time modeling, the space-time correlation of a noise sequence is captured by utilizing two-way propagation to the maximum extent, but the current work is based on the space correlation of distances, and the space characteristics constructed based on physical distances are difficult to completely express the correlation of the noise sequence between stations, so a better method is needed to solve the problem. In order to improve the prediction accuracy of the spatial relationship, the spatial relationship needs to be deeply predicted to solve the problem.
And 2.4, extracting traffic flow characteristics at the road gate corresponding to the current time in real time, and dynamically adjusting the DCRNN noise predicted value by using a Kalman filtering method.
Kalman filtering uses the estimate of the last state to make a prediction of the current state. And finally, correcting the predicted value obtained in the prediction stage by using the observed value of the current state to obtain a new estimated value closer to the true value. The method specifically comprises the steps of calculating Kalman gains of noise and drift by using normalized traffic flow characteristics and noise sequences through priori estimation and covariance matrix prediction, updating a covariance matrix by using Kalman gain correction, and then correcting and predicting a noise value at the current moment.
And 3, training the constructed noise prediction network model and outputting the trained model. The specific process is as follows:
extracting historical data of each monitoring station for reconstructing a time-space relationship of data by an improved DTW algorithm, training a noise prediction network model DTW-DCRNN, selecting 70% of a data set as a training set, 10% as a verification set, and finally 20% as a test set. In addition, the training results are not recorded in the first 60 epochs, the training results are recorded once every 10 epochs in the 60-100 epochs and the model parameters are output, and the training results are recorded once every 5 epochs after 100 epochs and the model parameters are output. And finally stopping training in advance according to the measured verification loss to ensure that the model is captured when the model is about to be overfitting.
And 4, collecting information of the monitoring station in real time and predicting noise in real time based on the trained model.
According to the method, a similarity distance matrix based on an improved DTW algorithm is established according to the position of each station, the traditional method for constructing the matrix based on physical distance is abandoned, the similarity distance matrix is constructed into a representation form of an adjacent matrix, a time-space data set based on the similarity distance is sent into a DCRNN model to obtain a DTW-DCRNN model based on the spatial relation of the similarity distance, and finally, the prediction output result of the DCRNN is dynamically updated through a Kalman filter in combination with real-time traffic flow data.
The similarity distance matrix is used for constructing the adjacency matrix relationship to represent the spatial distance relationship, so that the limitation of the original physical distance method can be well made up. The prediction result adjusted by the Kalman filter is closer to the true value, and the compensation property is provided, so that the 3 sigma noise processing method effectively avoids the condition that the model training prediction result is greatly deviated.
The following experiments were conducted in order to demonstrate the feasibility and superiority of the present invention.
Experiment 1: comparison experiment before and after DTW algorithm improvement
Fig. 7 (a) shows an example of an adjacency matrix based on the original DTW algorithm, and the improved DTW algorithm, such as fig. 7 (b), increases the association between monitoring station No. 1 and monitoring station No. 3 compared to the original method.
The average absolute error MAE, the average absolute percentage error MAPE and the root mean square error RMSE are used as evaluation indexes in the experiment. The DTW algorithm and the improved DTW algorithm are combined with a DCRNN model respectively for pre-training, the average prediction result of twelve time steps is obtained, the indexes are shown in table 1, MAE noise is reduced by 0.06 decibel, accuracy is improved by 2.8%, MAPE is reduced by 0.11%, accuracy is improved by 1.9%, RMSE is improved by 0.02 decibel, and accuracy is reduced by 0.5% in the pre-training model result. The improved MAE and MAPE are lower than the original method, the RMSE index is slightly higher than the original method, namely the data outlier and the abnormal value prediction accuracy of the improved method are lower than the original method, but the prediction accuracy of the overall index new DTW algorithm is higher than that of the original DTW algorithm.
TABLE 1 DTW Algorithm Pre-and post-improvement contrast
Drawing an index graph according to the time step, as shown in FIG. 8, further observing the improved DTW algorithm and the original algorithm, wherein the abscissa is the time step, and the ordinate on the right side of the image is the MAPE index. The ordinate on the left side of the image is an RMSE index and an MAE index, the unit is decibel (dB), and it can be seen from the graph that the root mean square error of the DWT algorithm improved from the sixth time step is smaller, the new method is better in performance on the RMSE index, the average absolute error of the DWT algorithm improved from the fourth time step is smaller, and the new method is better in performance on the MAE index.
In conclusion, the improved DTW algorithm not only improves the overall prediction accuracy, but also does not increase the time cost of algorithm calculation, and can mine deeper sequence correlation, so that the improved DTW algorithm can be innovatively applied to graph convolution work to improve the prediction accuracy.
Experiment 2: comparison experiment of prediction model provided by the invention and other models
In order to further verify the prediction performance of the method, three evaluation indexes of RMSE, MAE and MAPE are used for evaluating HA historical average prediction models, VAR vector autoregressive models, STGCN space-time convolution neural networks, DCRNN diffusion convolution recursive neural networks, DTW-STGCN, DTW-DCRNN (N) and other methods or models, wherein the DTW-DCRNN (N) method is a method without using Kalman filtering and dynamic regulation of traffic flow data. The above experiments are all based on 12 time step prediction, and the average value of the prediction results of 12 time steps is taken as an index to measure the long-term prediction effect of the model, and according to the table 2, the prediction results of the invention are superior to other prediction methods.
HA: because the noise periodicity is strong and the mutation points are more, the model has low accuracy of the prediction result and high training speed.
STGCN: due to the randomness of the noise sequence, the result accuracy rate is low on noise prediction.
DCRNN: the model effect is stronger than that of STGCN, the time-series noise data has strong time dependence, and the side surface reflects that GRU has more advantages than CNN in time-series modeling.
VAR: because the prediction result of the vector autoregressive model with strong periodicity of the noise sequence is superior to STGCN, the neural network does not have better performance under any condition.
DTW-STGCN: the DTW algorithm is applied to the STGCN model to construct the DTW-STGCN method, and the prediction performance is still stronger than that of the STGCN model. It was also verified from the side that the spatial relationship is meaningful in view of similarity in time series.
DTW-DCRNN (N): compared with DTW-DCRNN (N), the method adds Kalman filtering, has correction and adjustment effects on data, and has the best effect. Compared with the original DCRNN model, the RMSE and MAE precision of the method are respectively improved by 11.1 percent and 5 percent, and the MAPE is improved by 5.26 percent.
TABLE 2 Baseline model prediction results
Besides, the method is found to perform well in long-term prediction in experiments, and specific data are shown in table 3, so that the prediction accuracy of the method is lower than that of DCRNN in the first time step, but the performance tends to be more stable in the following 11 time steps, and the accuracy is higher.
TABLE 3 Long-term prediction results analysis
In order to facilitate observation of specific prediction trends, the above table can be visualized as an image, fig. 9 is a graph of an average absolute error MAE and a root mean square error RMSE, a vertical coordinate on the right side of the image is an RMSE index, a unit is decibel (dB), and as can be seen from the graph, prediction of 2 nd, 3 rd and 4 th time steps is more accurate. The ordinate on the left side of the image is an MAE index, the unit is decibel (dB), and under the condition that the first time is not considered, the errors of the two methods are increased along with the increase of the time step, but the DTW-DCRNN method has smaller errors and higher precision. FIG. 10 is a graph of the mean absolute percentage error MAPE with a trend similar to the MAE index, but with a smaller mean absolute percentage error for the DTW-DCRNN model of the present invention. The three indexes begin from the second time step (including the second time step), and the method of the invention has better performance.
In conclusion, the improved method not only improves the overall prediction precision, but also has more advantages in long-term noise time series prediction work compared with the original method, and can provide a new thought and method for a graph neural network and related researches thereof in space-time prediction work.
It is to be understood that the above description is not intended to limit the present invention, and the present invention is not limited to the above examples, and those skilled in the art may make modifications, alterations, additions or substitutions within the spirit and scope of the present invention.
Claims (4)
1. A DTW-DCRNN-based chemical industry park noise prediction method is characterized in that an improved time dynamic warping algorithm is adopted, a space relation is reconstructed through a graph structure of a monitoring station, and long-term space-time prediction of the chemical industry park noise is achieved, and the method specifically comprises the following steps:
step 1, collecting and processing monitoring site information and vehicle information data in a chemical industry park;
step 2, introducing a penalty coefficient to improve an original time dynamic warping algorithm, introducing the penalty coefficient into a diffusion convolution recurrent neural network, and constructing a noise prediction network model by combining Kalman filtering;
step 3, training the constructed noise prediction network model and outputting the trained model;
and 4, acquiring information of the monitoring station in real time and predicting noise in real time based on the trained model.
2. The DTW-DCRNN-based chemical industry park noise prediction method according to claim 1, wherein in step 1, a gateway and a plurality of monitoring sites are arranged in the chemical industry park for collecting data in real time and synchronously recording position information data of each monitoring site; sensors are arranged at the gateway and each monitoring station; the gateway collects information data of vehicles entering and leaving, including license plate number, vehicle entering and leaving time and vehicle flow characteristics of vehicle type; the monitoring station collects noise and natural environment data in a chemical industry park in real time;
and the station position data, the noise and natural environment data and the in-out vehicle information data are transmitted to the gateway equipment and then are interactively transmitted with the database server, and when noise prediction is carried out, data extraction is carried out in real time and 3 sigma criterion is adopted for processing.
3. The DTW-DCRNN-based chemical industry park noise prediction method according to claim 1, wherein the specific process of the step 2 is as follows:
step 2.1, introducing a penalty coefficient to improve an original time dynamic warping algorithm, and calculating a sequence similarity distance;
the principle of the improved time dynamic regularization algorithm is as follows: obtaining the number of the optimal paths and the number of the public subsequences to calculate a penalty coefficient, and calculating the similarity distance between stations by combining the penalty coefficient; the specific process is as follows:
step 2.1.1, calculating a distance matrix;
two noise sequences S and T are assumed, as in equation (1):
the length of the sequence S is o, the length of the sequence T is m, o, m belongs to Z + Constructing a distance matrix d [ s ] based on the noise sequence mxo][t]Wherein the ith sequence point s i (i =1,2, \8230;, o) and the jth sequence point t j (j =1,2, \8230;, m) is a matrix(s) i ,t j ) The value of the element(s) is,euclidean distance: dis(s) i ,t j )=(s i -t j ) 2 The matrix distance calculation criterion is shown in equation (2):
step 2.1.2, solving a matrix optimal path;
the optimal path is a path with the minimum accumulation distance meeting the lower boundary condition, and the lower boundary condition is as follows: starting from the d [ o ] [ m ] point at the upper right corner of the matrix, finding the minimum point in the three lower left points as the next node until the d [0] [0] point at the lower left corner is finished, ensuring continuity and monotonicity, not crossing or omitting a certain point for matching, only aligning or matching with adjacent points, and unchangeable time sequence;
step 2.1.3, calculating the number of public subsequences and the optimal path sequence length, and further obtaining weight to obtain a penalty coefficient;
wherein, the optimal path sequence length is the length of the optimal path of the matrix finally obtained in the step 2.1.2;
the calculation process of the number of the common subsequences is as follows: firstly, constructing a recording empty list record, circularly traversing sequences s1 and s2, recording the length ls1 or ls2 of a common sequence and the sequence position g of the common sequence by using the record list after the common sequence appears, and extracting the common subsequence from the s1 sequence and storing the common subsequence in an s _ sum array; the length of s _ sum after traversal is the number of the public subsequences; when calculating the number of the public subsequences, setting a threshold value as 1, calculating the subsequences with the length greater than 1 as the public subsequences, and finally counting the total number of the public subsequences to obtain the required number of the public noise subsequences;
the calculation formula of the weight w is as follows (3):
wherein, the subseq is the length of the public noise subsequence, and the seq is the sequence length of the optimal path of the matrix;
the calculation process for setting the penalty coefficient alpha is shown as formula (4), wherein x represents the number of public subsequences, and w i Represents the weight of the ith sequence point,
step 2.1.4, express the optimal path as (r) 1 ,r 2 ,…,r seq ),r i (i =1,2, \8230;, seq) represents the value of the ith sequence point in the optimal path sequence, and finally, the sum of the values in the optimal path calculated by the original DTW is multiplied by a penalty coefficient to obtain the improved sequence similarity distance dis DTW ;
Step 2.2, forming a matrix by the noise sequence similarity distance among all the monitored sites, calculating the matrix as an adjacent matrix, and constructing a graph relation topological structure; the specific process is as follows:
step 2.2.1, according to the noise sequence similarity distance information of each monitored site, calculating a similarity distance matrix M between the sites, wherein the matrix M is as follows:
wherein a is the number of monitoring stations;
step 2.2.2, calculating the standard deviation lambda, M of the similarity distance matrix M c Is the c-th element value in the matrix, N is the number of elements in the M matrix, mu M For matrix mean, the calculation process is as follows:
step 2.2.3, analyzing the similarity between the monitoring stations according to the similar distance matrix M, and constructing an adjacent matrix M by using the standard deviation lambda of all non-infinite numbers in the similar distance matrix d Of a contiguous matrix M d Each element in (1)The calculation process of (c) is as follows:
wherein c represents the sequence number of the matrix element, and the larger the similarity distance is, the larger the similarity distance isThe smaller the threshold is set to 0.1, ifIf the similarity between the two sites is too low, no mutual influence relationship exists, the two sites are regarded as non-adjacent sites, and the adjacent relationship is not formed when the weight in the adjacent matrix is zero;
step 2.3, introducing a graph convolution neural network GCN to construct an adjacent matrix graph structure, and inputting the adjacent matrix graph structure into a diffusion convolution recurrent neural network for noise prediction to obtain a preliminary prediction result; the specific process is as follows:
the propagation mode between the GCN network layers of the graph convolution neural network is shown as formula (12), wherein L represents the Lth network layer:
wherein σ is a nonlinear activation function; w L Is a matrix of trainable weights to produce a desired weight,a degree matrix, which is composed of:i denotes the I th row of the adjacency matrix A, J denotes the J th column of the adjacency matrix A,the characteristic information of the station is kept, H is the characteristic extracted by the current layer, and if the characteristic is the input layer, X = H;
on the basis of the GCN graph convolution neural network, a time-space relation of a noise sequence is modeled by using a diffusion convolution recursive neural network;
the smooth distribution of the diffusion process is represented as a weighted combination of infinite random walk on the graph, the diffusion process is represented by equation (13), and is calculated in closed form:
wherein W is a node similarity matrix,is the inverse of the rectangular output matrix, with beta ∈ [0,1 ]]Representing restart probability, k is diffusion degree, epsilon represents the possibility of diffusion from nodes, and the diffusion process in the DCRNN model is bidirectional;
thus, in the spatial relationship, based on the map signal feature matrixAnd filter f θ The bi-diffusion convolution operation is defined as equation (14):
where θ is the filter parameter, G represents graph G, P represents the pth feature dimension, P represents the total number of feature dimensions, K represents the finite K-step truncation of the diffusion process, and G represents the graph GDiffusion convolution, X:, p representing the convolution of all nodes with the p-th feature, θ k,1 A convolution kernel parameter, θ, representing a degree calculation k,2 Represents the parameters of the convolution kernel of the in-degree calculation,and withRespectively representing transition matrixes of diffusion and inverse diffusion, and constructing a diffusion convolution layer with a mapping relation after defining convolution operation as shown in a formula (15):
wherein Q represents the qth feature, Q represents the total number of mapped features, H, q representing the q characteristic diffusion convolution operation of all the node pairs and a characteristic matrixFeatures extracted from the current layer as inputIn order to be output, the output is,representing a filter, wherein theta is a parameter tensor, and sigma is an activation function;
in the time relation, a diffusion convolution gating recursion unit DCGRU is constructed by combining GRU and diffusion convolution, and is expressed as an expression (16):
∑ G represents a diffusion convolution of graph G, r is a radical of hydrogen (t) Indicating the reset gate at time t, [ theta ] r★G Door with indication of resetTensor of parameter (c), X (t) Indicating input at time t, H (t-1) Representing the output at time t-1, b r A bias vector representing a reset gate; u. of (t) Indicating the update Gate at time t, Θ u★G Tensor of parameters representing the update gate, b u A bias vector representing an update gate; c (t) Indicates the hidden state of the next cell at time t, [ theta ] C★G Tensor of parameters representing hidden states, b C A bias vector representing a hidden state; h (t) Represents the output at time t;
step 2.4, extracting the traffic flow characteristics at the road gate corresponding to the current time in real time, and dynamically adjusting the DCRNN noise predicted value by using a Kalman filtering method;
kalman filtering uses the estimate of the previous state to make a prediction of the current state; finally, correcting the predicted value obtained in the prediction stage by utilizing the observed value of the current state to obtain a new estimated value which is closer to the true value; the method specifically comprises the steps of calculating Kalman gains of noise and drift by using normalized traffic flow characteristics and noise sequences through priori estimation and covariance matrix prediction, updating a covariance matrix by using Kalman gain correction, and then correcting and predicting a noise value at the current moment.
4. The DTW-DCRNN-based chemical industry park noise prediction method according to claim 1, wherein the specific process of the step 3 is as follows:
extracting historical data of each monitoring station for reconstructing a spatio-temporal relationship of data by an improved DTW algorithm, training a noise prediction network model DTW-DCRNN, selecting 70% of a data set as a training set, 10% as a verification set and finally 20% as a test set; in addition, the training results are not recorded in the first 60 epochs, the training results are recorded once every 10 epochs in the 60-100 epochs and the model parameters are output, and the training results are recorded once every 5 epochs after 100 epochs and the model parameters are output; and finally stopping training in advance according to the measured verification loss to ensure that the model is captured when the model is about to be overfitting.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211238200.1A CN115659609A (en) | 2022-10-11 | 2022-10-11 | DTW-DCRNN-based chemical industry park noise prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211238200.1A CN115659609A (en) | 2022-10-11 | 2022-10-11 | DTW-DCRNN-based chemical industry park noise prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115659609A true CN115659609A (en) | 2023-01-31 |
Family
ID=84987374
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211238200.1A Pending CN115659609A (en) | 2022-10-11 | 2022-10-11 | DTW-DCRNN-based chemical industry park noise prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115659609A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116206117A (en) * | 2023-03-03 | 2023-06-02 | 朱桂湘 | Signal processing optimization system and method based on number traversal |
CN117155707A (en) * | 2023-10-30 | 2023-12-01 | 广东省通信产业服务有限公司 | Harmful domain name detection method based on passive network flow measurement |
-
2022
- 2022-10-11 CN CN202211238200.1A patent/CN115659609A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116206117A (en) * | 2023-03-03 | 2023-06-02 | 朱桂湘 | Signal processing optimization system and method based on number traversal |
CN116206117B (en) * | 2023-03-03 | 2023-12-01 | 北京全网智数科技有限公司 | Signal processing optimization system and method based on number traversal |
CN117155707A (en) * | 2023-10-30 | 2023-12-01 | 广东省通信产业服务有限公司 | Harmful domain name detection method based on passive network flow measurement |
CN117155707B (en) * | 2023-10-30 | 2023-12-29 | 广东省通信产业服务有限公司 | Harmful domain name detection method based on passive network flow measurement |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111222698B (en) | Internet of things-oriented ponding water level prediction method based on long-time and short-time memory network | |
CN109754605B (en) | Traffic prediction method based on attention temporal graph convolution network | |
CN108304668B (en) | Flood prediction method combining hydrologic process data and historical prior data | |
CN115659609A (en) | DTW-DCRNN-based chemical industry park noise prediction method | |
Abbasi et al. | A hybrid of Random Forest and Deep Auto-Encoder with support vector regression methods for accuracy improvement and uncertainty reduction of long-term streamflow prediction | |
Kumar et al. | Artificial neural network model for rainfall-runoff-A case study | |
CN110648014B (en) | Regional wind power prediction method and system based on space-time quantile regression | |
CN114220271A (en) | Traffic flow prediction method, equipment and storage medium based on dynamic space-time graph convolution cycle network | |
CN111882869B (en) | Deep learning traffic flow prediction method considering adverse weather | |
CN111767517B (en) | BiGRU multi-step prediction method, system and storage medium applied to flood prediction | |
CN111612243A (en) | Traffic speed prediction method, system and storage medium | |
CN115578852B (en) | DSTGCN-based traffic prediction method | |
CN111275255A (en) | Construction method of concrete dam deformation monitoring and forecasting model | |
CN113705877A (en) | Real-time monthly runoff forecasting method based on deep learning model | |
CN115495991A (en) | Rainfall interval prediction method based on time convolution network | |
CN116933621A (en) | Urban waterlogging simulation method based on terrain feature deep learning | |
CN112163669A (en) | Pavement subsidence prediction method based on BP neural network | |
CN114611814A (en) | Urban traffic flow prediction method for aggregating multi-scale space-time similar information | |
Zhou et al. | Integrated dynamic framework for predicting urban flooding and providing early warning | |
CN116913098B (en) | Short-time traffic flow prediction method integrating air quality and vehicle flow data | |
CN111815075B (en) | Prediction method for transportation travel demand under major public health incident | |
CN110516792A (en) | Non-stable time series forecasting method based on wavelet decomposition and shallow-layer neural network | |
CN117131991A (en) | Urban rainfall prediction method and platform based on hybrid neural network | |
CN116050460A (en) | Air temperature data spatial interpolation method based on attention neural network | |
CN112115754A (en) | Short-term traffic flow prediction model based on firework differential evolution hybrid algorithm-extreme learning machine |
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 |