CN114565149A - CGA fusion model-based time series data prediction method and device and computer equipment - Google Patents
CGA fusion model-based time series data prediction method and device and computer equipment Download PDFInfo
- Publication number
- CN114565149A CN114565149A CN202210169309.8A CN202210169309A CN114565149A CN 114565149 A CN114565149 A CN 114565149A CN 202210169309 A CN202210169309 A CN 202210169309A CN 114565149 A CN114565149 A CN 114565149A
- Authority
- CN
- China
- Prior art keywords
- target data
- data
- time sequence
- features
- feature
- 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
- 238000000034 method Methods 0.000 title claims abstract description 53
- 230000004927 fusion Effects 0.000 title claims abstract description 44
- 238000007781 pre-processing Methods 0.000 claims abstract description 16
- 238000013528 artificial neural network Methods 0.000 claims abstract description 14
- 238000000605 extraction Methods 0.000 claims description 23
- 238000013527 convolutional neural network Methods 0.000 claims description 21
- 238000012545 processing Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 14
- 238000005070 sampling Methods 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 6
- 230000015654 memory Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000012549 training Methods 0.000 abstract description 7
- 230000007246 mechanism Effects 0.000 description 5
- 125000004122 cyclic group Chemical group 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 3
- 238000013075 data extraction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 241001123248 Arma Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Development Economics (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Educational Administration (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of time sequence prediction, in particular to a time sequence data prediction method, a device and computer equipment based on a CGA fusion model; the method comprises the following steps: acquiring historical target data, and preprocessing the historical target data to obtain preprocessed historical target data; inputting the preprocessed historical target data into the constructed CGA fusion model to obtain a time sequence prediction result of the target data, and controlling future traffic flow according to the time sequence prediction result; the invention combines the stack type and the parallel type to fuse the multi-channel multi-scale convolution neural network and the gate control cycle unit network, and combines the multi-channel multi-scale convolution neural network, the gate control cycle unit network and the autoregressive model in a residual error mode, thereby reducing the training difficulty of the model, avoiding the loss of original information caused by a convolution structure and a GRU structure, and further improving the prediction precision.
Description
Technical Field
The invention belongs to the technical field of time sequence prediction, and particularly relates to a time sequence data prediction method and device based on a CGA fusion model, and computer equipment.
Background
The time series prediction task is a type of task ubiquitous in machine learning, and is widely applied to the fields of finance, industry, manufacturing, transportation and the like, such as power load prediction, traffic flow prediction, stock price prediction, weather data prediction and the like. With the rapid development of deep learning technology in recent years, the deep learning technology has proved to be an important modeling method in many fields. Meanwhile, some time series prediction methods have been developed, however, the existing time series prediction method based on the neural network has some problems, for example, the local features of the time series data on different time scales, the global time series feature of the whole data and the burst feature of the time series data are not considered at the same time, that is, the non-linear feature and the linear feature are not considered at the same time; or the single model has limited performance and limited applicable scenes, so that the prediction precision is not high. The following description will be made for a currently existing time series prediction method.
A multi-element time sequence multi-layer space-time dependency modeling method based on deep learning is disclosed in a patent with the application number of CN202010496285.8 and the name of 'multi-element time sequence multi-layer space-time dependency modeling method based on deep learning'. The method mainly utilizes a CNN-LSTM structure to preliminarily extract features and consists of a CNN space attention mechanism, a CNN channel attention mechanism, a CNN time attention mechanism and an autoregressive component. The method has the advantages that the concept of multi-layer space-time dependence is introduced, the CNN-based channel attention mechanism and the CNN-based space attention mechanism are used for respectively paying attention to space-time dependence characteristics of different layers, the redundant information is filtered, the characteristics with larger influence on the prediction result are effectively extracted, and the purpose of improving the prediction result is achieved. Although the method combines the nonlinear characteristic and the linear characteristic at the same time, the method does not consider the local characteristics of different time scales and the stacked CNN-LSTM structure, thereby increasing the training difficulty in model training.
A CNN-based time series prediction method and a model determination method are disclosed in a patent with an application number of CN201910460741.0 entitled "CNN-based time series prediction method and model determination method". The method comprises the following steps: acquiring historical time sequence data, and determining cycle parameters according to cycle characteristics of the historical time sequence data, wherein the cycle parameters comprise cycle types and corresponding cycle parameters; determining corresponding component data of the prediction time point in the historical time series data based on the prediction time point, the historical time series data, the period parameter and a preset cycle span, wherein the component data comprises closest time period data and period data; and predicting the component data by adopting the determined CNN model to obtain a prediction result corresponding to the prediction time point. The core of the method is to use the period parameters of the data, respectively use CNN to extract features through long, short periods and adjacent data, and predict after fusing the features. Although the method considers the characteristics of different time scales, the method has certain requirements on the periodicity of data, does not consider the non-periodic time-dependent characteristics and the short-term linear characteristics, and has certain limitations on the prediction effect and the applicable scene.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a time series data prediction method, a device and computer equipment based on a CGA fusion model, wherein the method comprises the following steps: acquiring historical data of target data, and preprocessing the historical target data to obtain preprocessed historical target data; the target data is traffic flow data; inputting the preprocessed historical target data into the constructed CGA fusion model to obtain a time sequence prediction result of the target data, and controlling future traffic flow according to the time sequence prediction result; the CGA fusion model comprises a first layer of multi-channel multi-scale convolution neural network, a gated circulation unit network, a second layer of multi-channel multi-scale convolution neural network and an autoregressive model;
the process of constructing the CGA fusion model comprises the following steps:
s1: inputting historical target data into a first layer of multi-channel multi-scale convolutional neural network MCMSCNN to obtain primary features;
s2: inputting the primary characteristics into a gated cycle unit network GRU to obtain time sequence characteristics;
s3: inputting the primary features into a second-layer multi-channel multi-scale convolutional neural network MCMSCNN to obtain local features;
s4: inputting historical target data into an autoregressive model AR to obtain linear characteristics;
s5: and adding the time sequence characteristics, the local characteristics and the linear characteristics to obtain a constructed CGA fusion model.
Preferably, the process of preprocessing the historical target data includes:
removing abnormal values from the historical target data and filling missing values to obtain complete target data;
carrying out normalization processing on the perfect target data to obtain normalized data;
and setting the length T of a sliding window, and sampling the normalized data by adopting the sliding window to obtain preprocessed historical target data.
Preferably, the process of obtaining the preliminary characteristics comprises: convolution cores with three different sizes are adopted to carry out convolution, activation and dropout processing on input data, and three feature tensors with different lengths, namely primary features, are obtained.
Preferably, the process of obtaining the time sequence characteristics includes: processing the primary characteristics, and inputting the processed primary characteristics into a gated cycle unit network GRU to obtain time sequence characteristics; processing the preliminary features includes:
filling two feature tensors with shorter lengths in the three feature tensors with different lengths to obtain three feature tensors with the same length;
and splicing the three feature tensors with the same length to obtain the processed primary features.
Preferably, the process of obtaining the local feature includes:
performing convolution, activation, maximum pooling and dropout processing on input data by adopting convolution kernels with three different sizes to obtain three feature tensors with different lengths;
splicing and flatten operations are carried out on the feature tensors with three different lengths, and the feature tensors are output through a linear layer to obtain local features;
preferably, the process of obtaining the time-series prediction result of the target data includes:
inputting the sampling data of the first sliding window into a CGA fusion model to obtain a predicted value at the current moment;
splicing the target data with the predicted value at the current moment, sliding a window, intercepting the next sliding window sampling data, and inputting the next sliding window sampling data into a CGA fusion model to obtain the predicted value at the next moment; and repeating the process to obtain a time sequence prediction result of the target data.
A CGA fusion model-based time series data prediction device comprises: the system comprises a data preprocessing module, a primary feature extraction module, a local feature extraction module, a time sequence feature extraction module, a linear feature capture module and a time sequence prediction module;
the data preprocessing module is used for preprocessing historical target data;
the primary feature extraction module is used for extracting features of the preprocessed historical target data and outputting primary features;
the local feature extraction module is used for extracting local features according to the primary features and outputting the local features;
the time sequence feature extraction module is used for acquiring time sequence features according to the primary features;
the linear feature capturing module is used for acquiring linear features according to historical target data;
the time sequence prediction module is used for obtaining a time sequence prediction result according to the local characteristic, the time sequence characteristic and the linear characteristic and outputting the time sequence prediction result.
A computer device comprising a processor and a memory, said memory storing computer readable instructions which, when executed by said processor, perform the steps of the above method.
The invention has the beneficial effects that: compared with the existing time sequence prediction method, the CGA fusion model provided by the invention comprehensively considers nonlinear features and linear features in the aspect of feature extraction, wherein the nonlinear features comprise local features of a plurality of time scales captured by MCMSCNN, convolution output vectors of the scales are spliced, on one hand, the time sequence features are captured globally by GRU, on the other hand, the MCMSCNN added with a pooling layer is continuously sent to extract more complex local features; the AR model can capture the linear characteristics of the short-time burst, and the prediction result is finally obtained. Because the characteristics of time sequence data extraction in the fusion model are rich, a more accurate prediction result can be obtained. In the aspect of model fusion, the invention is different from a typical stacked mode and a typical parallel mode, combines the two modes simultaneously to fuse the MCMSCNN and GRU structures, combines the MCMSCNN, the GRU and the AR in a residual error mode, reduces the training difficulty of the model, and avoids the loss of original information caused by a convolution structure and the GRU structure, thereby further improving the prediction precision.
Drawings
FIG. 1 is a schematic diagram of a CGA fusion model framework according to the present invention;
FIG. 2 is a flow chart of the construction of a CGA fusion model according to the present invention;
FIG. 3 is a diagram illustrating the multi-scale convolution according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention provides a time series data prediction method, a time series data prediction device and computer equipment based on a CGA fusion model, and as shown in figure 1, the method comprises the following steps: acquiring historical data of target data, and preprocessing the historical target data to obtain preprocessed historical target data; the target data is traffic flow data; inputting the preprocessed historical target data into the constructed CGA fusion model to obtain a time sequence prediction result of the target data, and controlling future traffic flow according to the time sequence prediction result; the CGA fusion model comprises a first layer of multi-channel multi-scale convolution neural network, a gate control cycle unit network, a second layer of multi-channel multi-scale convolution neural network and an autoregressive model;
as shown in fig. 2, the process of constructing the CGA fusion model includes:
s1: inputting historical target data into a first layer of multi-channel multi-scale convolutional neural network MCMSCNN to obtain primary features;
s2: inputting the primary characteristics into a gated cycle unit network GRU to obtain time sequence characteristics;
s3: inputting the primary features into a second-layer multi-channel multi-scale convolutional neural network MCMSCNN to obtain local features;
s4: inputting historical target data into an autoregressive model AR to obtain linear characteristics;
s5: and adding the time sequence characteristic, the local characteristic and the linear characteristic to obtain the constructed CGA fusion model.
Acquiring historical target data, and preprocessing the historical target data; specifically, removing abnormal values from the historical target data and filling missing values to obtain complete target data; carrying out normalization processing on the perfect target data to obtain normalized data; setting the length T of a sliding window, and sampling the normalized data by adopting the sliding window to obtain preprocessed historical target data; preferably, the dimensions of the historical target data may be multidimensional data, i.e., not limited to cell or multi-element time series data.
As shown in fig. 3, inputting historical target data into a first-layer multichannel multi-scale convolutional neural network MCMSCNN to obtain a primary feature; the first layer of multi-channel multi-scale convolution neural network MCMSCNN is composed of convolution kernels with various sizes, preferably, the number of the convolution kernels is 3, the sizes are respectively 1 × 2, 1 × 3 and 1 × 4, the number of the convolution kernels in each size is more than 2, and for example, the number of the convolution kernels in each size is 50; the first layer of MCMSCNN processes the historical target data as follows: performing convolution, activation and dropout processing on input data by adopting convolution cores with three different sizes to obtain three feature tensors with different lengths, wherein the three feature tensors with different lengths are primary features;
processing the primary features:
filling two feature tensors with shorter lengths in the three feature tensors with different lengths to obtain three feature tensors with the same length; splicing three feature tensors with the same length to obtain processed primary features; and inputting the processed primary characteristics into a gated cycle unit network GRU to obtain the time sequence characteristics.
After the historical target data is subjected to primary local feature extraction through the first layer of MCMSCNN, the historical target data still has a time sequence relation in a time dimension; the GRU is a recurrent neural network, is suitable for capturing time sequence characteristics, and has the advantages that the parameter quantity can be smaller under similar performance compared with an LSTM (long short term memory network); preferably, the primary signature is inputted into the gated loop unit network GRU corresponding to each time point thereof, and the timing signature can be captured. Optionally, in the gated cyclic unit neural network structure, the depth is not limited to one layer, and the gated cyclic unit structure may be replaced by other cyclic neural network structures (including, but not limited to, network structures such as RNN, LSTM, GRU, SRU, etc. of one or two directions).
Inputting the primary features into a second layer of multi-channel multi-scale convolutional neural network MCMSCNN to obtain local features; the second layer of multi-channel multi-scale convolutional neural network MCMSCNN is composed of convolution kernels with various sizes, preferably, the number of the convolution kernels is 3, the sizes are respectively 1 × 2, 1 × 3 and 1 × 4, the number of the convolution kernels in each size is more than 2, and for example, the number of the convolution kernels in each size is 50; the process of the second layer MCMSCNN for processing the primary features is as follows:
performing convolution, activation, maximum pooling and dropout processing on input data by adopting convolution kernels with three different sizes to obtain three feature tensors with different lengths;
splicing and flatten operations are carried out on the feature tensors with three different lengths, and the feature tensors are output through a linear layer to obtain local features;
preferably, the historical target data is input into an autoregressive model AR to obtain linear characteristics:
featureAR=wTx+b
x=(x1,x2,...,xn)
wherein featureARRepresenting a linear characteristic, xnThe value represented on the nth attribute of the historical target data corresponds to a time sequence and is the value of the nth time in a certain training sample, and w is equal to (w)1,w2,...,wn) And b are parameters that require training.
Alternatively, the regression model may be replaced with other linear models (including but not limited to ARIMA, ARMA, MA, etc. models).
Adding the time sequence characteristics, the local characteristics and the linear characteristics to obtain a constructed CGA fusion model:
featureall=featureCNN+featureGRU+featureAR
wherein featureCNNLocal features (local features at various time scales) representing MCMSCNN extraction; featureGRURepresenting the time sequence characteristics extracted by the GRU; featureARLinear features of AR extraction are represented.
Inputting the preprocessed historical target data into a constructed CGA fusion model, wherein the process of carrying out multi-step time sequence prediction on the CGA fusion model according to the historical target data comprises the following steps:
intercepting the first time window t by adopting a sliding windowN,…,tN+T-1Inputting the sampled data into the CGA fusion model, and performing inverse normalization on the output of the CGA fusion model to obtain the current time tN+TThe predicted value of (2);
splicing the target data with the predicted value of the current moment, and intercepting the next time window { t ] by a sliding windowN+1,…,tN+TInputting the sampling data into the CGA fusion model to obtain the next time tN+T+1The predicted value of (2); repeating the process to obtain the t of the target dataN+T+2,tN+T+3… } time series prediction results.
If only single-step time sequence prediction is needed, a sliding window is not needed, the current sampling data is input into the CGA fusion model, and the output of the CGA fusion model is subjected to inverse normalization to obtain a time sequence prediction result of the target data at the next moment.
The user can control the future traffic flow according to the time sequence prediction result.
The invention also provides a CGA fusion model-based time series data prediction device, which comprises: the system comprises a data preprocessing module, a primary feature extraction module, a local feature extraction module, a time sequence feature extraction module, a linear feature capture module and a time sequence prediction module;
the data preprocessing module is used for preprocessing historical target data;
the primary feature extraction module is used for extracting features of the preprocessed historical target data and outputting primary features;
the local feature extraction module is used for extracting local features according to the primary features and outputting the local features;
the time sequence feature extraction module is used for acquiring time sequence features according to the primary features;
the linear feature capturing module is used for acquiring linear features according to historical target data;
the time sequence prediction module is used for obtaining a time sequence prediction result according to the local characteristic, the time sequence characteristic and the linear characteristic and outputting the time sequence prediction result.
The invention also provides computer equipment comprising a processor and a memory, wherein the memory stores computer readable instructions, and the computer readable instructions are executed by the processor to execute the steps of the method.
Compared with the existing time sequence prediction method, the CGA fusion model comprehensively considers nonlinear features and linear features in the aspect of feature extraction, wherein the nonlinear features comprise local features of a plurality of time scales captured by MCMSCNN, convolution output vectors of the plurality of scales are spliced, on one hand, the convolution output vectors are sent to GRU to capture time sequence features globally, on the other hand, the MCMSCNN added with a pooling layer is continuously sent to the GRU, and more complex local features are extracted; the AR model can capture the linear characteristics of the short-time burst, and the prediction result is finally obtained. Because the characteristics of time sequence data extraction in the fusion model are rich, a more accurate prediction result can be obtained. In the aspect of model fusion, the method is different from a typical stacked type and a parallel type, a convolutional neural network and a gated cyclic unit neural network are fused by combining a stacking mode and a parallel mode, and an autoregressive model is combined in a residual error mode, so that the loss of original information caused by model stacking is avoided, the training difficulty is reduced, and the prediction precision is further improved.
The above-mentioned embodiments, which further illustrate the objects, technical solutions and advantages of the present invention, should be understood that the above-mentioned embodiments are only preferred embodiments of the present invention, and should not be construed as limiting the present invention, and any modifications, equivalents, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (8)
1. A CGA fusion model-based time series data prediction method is characterized by comprising the following steps: acquiring historical data of target data, and preprocessing the historical target data to obtain preprocessed historical target data; the target data is traffic flow data; inputting the preprocessed historical target data into the constructed CGA fusion model to obtain a time sequence prediction result of the target data, and controlling future traffic flow according to the time sequence prediction result; the CGA fusion model comprises a first layer of multi-channel multi-scale convolution neural network, a gate control cycle unit network, a second layer of multi-channel multi-scale convolution neural network and an autoregressive model;
the process of constructing the CGA fusion model comprises the following steps:
s1: inputting historical target data into a first layer of multi-channel multi-scale convolutional neural network MCMSCNN to obtain primary features;
s2: inputting the primary characteristics into a gated cycle unit network GRU to obtain time sequence characteristics;
s3: inputting the primary features into a second-layer multi-channel multi-scale convolutional neural network MCMSCNN to obtain local features;
s4: inputting historical target data into an autoregressive model AR to obtain linear characteristics;
s5: and adding the time sequence characteristic, the local characteristic and the linear characteristic to obtain the constructed CGA fusion model.
2. The CGA fusion model-based time series data prediction method of claim 1, wherein the pre-processing of the historical target data comprises:
removing abnormal values from the historical target data and filling missing values to obtain complete target data;
carrying out normalization processing on the perfect target data to obtain normalized data;
and setting the length T of a sliding window, and sampling the normalized data by adopting the sliding window to obtain the preprocessed historical target data.
3. The method of claim 1, wherein the step of obtaining the primary features comprises: convolution cores with three different sizes are adopted to carry out convolution, activation and dropout processing on input data, and three feature tensors with different lengths, namely primary features, are obtained.
4. The method of claim 1, wherein the step of obtaining the time series feature comprises: processing the primary characteristics, and inputting the processed primary characteristics into a gated cycle unit network GRU to obtain time sequence characteristics; processing the preliminary features includes:
filling two feature tensors with shorter lengths in the three feature tensors with different lengths to obtain three feature tensors with the same length;
and splicing the three feature tensors with the same length to obtain the processed primary features.
5. The method of claim 1, wherein the obtaining of the local features comprises:
performing convolution, activation, maximum pooling and dropout processing on input data by adopting convolution kernels with three different sizes to obtain three feature tensors with different lengths;
and splicing and flatten operations are carried out on the feature tensors with three different lengths, and the local features are obtained through linear layer output.
6. The method of claim 1, wherein obtaining the time-series prediction result of the target data comprises:
inputting the sampling data of the first sliding window into a CGA fusion model to obtain a predicted value at the current moment;
splicing the target data with the predicted value at the current moment, sliding a window, intercepting the next sliding window sampling data, and inputting the next sliding window sampling data into a CGA fusion model to obtain the predicted value at the next moment; and repeating the process to obtain a time sequence prediction result of the target data.
7. A time series data prediction device based on a CGA fusion model is characterized by comprising: the system comprises a data preprocessing module, a primary feature extraction module, a local feature extraction module, a time sequence feature extraction module, a linear feature capture module and a time sequence prediction module;
the data preprocessing module is used for preprocessing historical target data;
the primary feature extraction module is used for extracting features of the preprocessed historical target data and outputting primary features;
the local feature extraction module is used for extracting local features according to the primary features and outputting the local features;
the time sequence feature extraction module is used for acquiring time sequence features according to the primary features;
the linear feature capturing module is used for acquiring linear features according to historical target data;
the time sequence prediction module is used for obtaining a time sequence prediction result according to the local characteristic, the time sequence characteristic and the linear characteristic and outputting the time sequence prediction result.
8. A computer device comprising a processor and a memory, said memory storing computer readable instructions which, when executed by said processor, perform the steps of the method of any one of claims 1 to 6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210169309.8A CN114565149A (en) | 2022-02-23 | 2022-02-23 | CGA fusion model-based time series data prediction method and device and computer equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210169309.8A CN114565149A (en) | 2022-02-23 | 2022-02-23 | CGA fusion model-based time series data prediction method and device and computer equipment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114565149A true CN114565149A (en) | 2022-05-31 |
Family
ID=81713038
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210169309.8A Pending CN114565149A (en) | 2022-02-23 | 2022-02-23 | CGA fusion model-based time series data prediction method and device and computer equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114565149A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114971748A (en) * | 2022-07-27 | 2022-08-30 | 阿里健康科技(中国)有限公司 | Prediction data generation method, model training method, computer device, and storage medium |
CN116821673A (en) * | 2023-05-19 | 2023-09-29 | 中国科学院自动化研究所 | ENSO prediction method, ENSO prediction device, electronic equipment and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180053108A1 (en) * | 2016-08-16 | 2018-02-22 | Toyota Jidosha Kabushiki Kaisha | Efficient Driver Action Prediction System Based on Temporal Fusion of Sensor Data Using Deep (Bidirectional) Recurrent Neural Network |
CN111950810A (en) * | 2020-08-27 | 2020-11-17 | 南京大学 | Multivariable time sequence prediction method and device based on self-evolution pre-training |
AU2020104006A4 (en) * | 2020-12-10 | 2021-02-18 | Naval Aviation University | Radar target recognition method based on feature pyramid lightweight convolutional neural network |
CN112488392A (en) * | 2020-12-01 | 2021-03-12 | 重庆邮电大学 | Intelligent water affair daily water consumption prediction method based on machine learning |
US20210142161A1 (en) * | 2019-11-13 | 2021-05-13 | General Electric Company | Systems and methods for model-based time series analysis |
CN113486938A (en) * | 2021-06-28 | 2021-10-08 | 重庆大学 | Multi-branch time convolution network-based re-landing analysis method and device |
-
2022
- 2022-02-23 CN CN202210169309.8A patent/CN114565149A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180053108A1 (en) * | 2016-08-16 | 2018-02-22 | Toyota Jidosha Kabushiki Kaisha | Efficient Driver Action Prediction System Based on Temporal Fusion of Sensor Data Using Deep (Bidirectional) Recurrent Neural Network |
US20210142161A1 (en) * | 2019-11-13 | 2021-05-13 | General Electric Company | Systems and methods for model-based time series analysis |
CN111950810A (en) * | 2020-08-27 | 2020-11-17 | 南京大学 | Multivariable time sequence prediction method and device based on self-evolution pre-training |
CN112488392A (en) * | 2020-12-01 | 2021-03-12 | 重庆邮电大学 | Intelligent water affair daily water consumption prediction method based on machine learning |
AU2020104006A4 (en) * | 2020-12-10 | 2021-02-18 | Naval Aviation University | Radar target recognition method based on feature pyramid lightweight convolutional neural network |
CN113486938A (en) * | 2021-06-28 | 2021-10-08 | 重庆大学 | Multi-branch time convolution network-based re-landing analysis method and device |
Non-Patent Citations (2)
Title |
---|
万晨: "基于深度学习的多变量时间序列预测算法与应用研究", 《中国优秀硕士学位论文全文数据库基础科学辑》, vol. 2021, no. 02, 15 February 2021 (2021-02-15), pages 002 - 925 * |
邓磊: "基于融合模型的短期用电负荷预测研究", 《万方数据》, 6 July 2023 (2023-07-06), pages 1 - 60 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114971748A (en) * | 2022-07-27 | 2022-08-30 | 阿里健康科技(中国)有限公司 | Prediction data generation method, model training method, computer device, and storage medium |
CN114971748B (en) * | 2022-07-27 | 2022-11-01 | 阿里健康科技(中国)有限公司 | Prediction data generation method, model training method, computer device, and storage medium |
CN116821673A (en) * | 2023-05-19 | 2023-09-29 | 中国科学院自动化研究所 | ENSO prediction method, ENSO prediction device, electronic equipment and storage medium |
CN116821673B (en) * | 2023-05-19 | 2024-01-16 | 中国科学院自动化研究所 | ENSO prediction method, ENSO prediction device, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110084274B (en) | Real-time image semantic segmentation method and system, readable storage medium and terminal | |
US11538286B2 (en) | Method and apparatus for vehicle damage assessment, electronic device, and computer storage medium | |
CN114565149A (en) | CGA fusion model-based time series data prediction method and device and computer equipment | |
WO2023217163A1 (en) | Dam defect time-sequence image description method based on local self-attention mechanism | |
CN113905391A (en) | Ensemble learning network traffic prediction method, system, device, terminal, and medium | |
US20230162477A1 (en) | Method for training model based on knowledge distillation, and electronic device | |
CN111598860A (en) | Lithium battery defect detection method based on yolov3 network embedded in self-attention door module | |
CN113889076B (en) | Speech recognition and coding/decoding method, device, electronic equipment and storage medium | |
CN114816997B (en) | Defect prediction method based on graph neural network and bidirectional GRU feature extraction | |
US20230368500A1 (en) | Time-series image description method for dam defects based on local self-attention | |
CN117668563B (en) | Text recognition method, text recognition device, electronic equipment and readable storage medium | |
CN116051388A (en) | Automatic photo editing via language request | |
CN115100549A (en) | Transmission line hardware detection method based on improved YOLOv5 | |
CN115994317A (en) | Incomplete multi-view multi-label classification method and system based on depth contrast learning | |
CN115618269A (en) | Big data analysis method and system based on industrial sensor production | |
CN114385694A (en) | Data processing method and device, computer equipment and storage medium | |
WO2023185209A1 (en) | Model pruning | |
CN116825238A (en) | Crystal geometry and symmetry prediction method and device based on physical information fusion neural network | |
CN111476131B (en) | Video processing method and device | |
CN111382761B (en) | CNN-based detector, image detection method and terminal | |
CN117407754B (en) | Multi-mode large model training strategy determination method, electronic equipment and medium | |
CN114550036B (en) | Searching method and system for optimal cascade configuration of video target detection | |
CN113793604B (en) | Speech recognition system optimization method and device | |
CN114969359B (en) | Entity relation joint extraction method and system based on machine reading understanding | |
CN118113503A (en) | Intelligent operation and maintenance system fault prediction method, device, equipment and 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 |