CN114239990A - Time series data prediction method based on time series decomposition and LSTM - Google Patents
Time series data prediction method based on time series decomposition and LSTM Download PDFInfo
- Publication number
- CN114239990A CN114239990A CN202111602708.0A CN202111602708A CN114239990A CN 114239990 A CN114239990 A CN 114239990A CN 202111602708 A CN202111602708 A CN 202111602708A CN 114239990 A CN114239990 A CN 114239990A
- Authority
- CN
- China
- Prior art keywords
- neural network
- time series
- data
- prediction
- training set
- 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 33
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 65
- 238000013528 artificial neural network Methods 0.000 claims abstract description 55
- 238000012360 testing method Methods 0.000 claims abstract description 37
- 238000003062 neural network model Methods 0.000 claims abstract description 30
- 238000012545 processing Methods 0.000 claims abstract description 8
- 230000000737 periodic effect Effects 0.000 claims description 16
- 238000010606 normalization Methods 0.000 claims description 6
- 239000000654 additive Substances 0.000 claims 2
- 230000000996 additive effect Effects 0.000 claims 2
- 238000007781 pre-processing Methods 0.000 abstract description 2
- 210000002569 neuron Anatomy 0.000 description 6
- 238000012544 monitoring process Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 208000025721 COVID-19 Diseases 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000005265 energy consumption Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 208000025174 PANDAS Diseases 0.000 description 1
- 208000021155 Paediatric autoimmune neuropsychiatric disorders associated with streptococcal infection Diseases 0.000 description 1
- 240000000220 Panda oleosa Species 0.000 description 1
- 235000016496 Panda oleosa Nutrition 0.000 description 1
- 238000003915 air pollution Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000012528 membrane Substances 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012731 temporal analysis Methods 0.000 description 1
- 238000000700 time series analysis Methods 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Tourism & Hospitality (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)
- General Business, Economics & Management (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Primary Health Care (AREA)
- Educational Administration (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a time series data prediction method based on time series decomposition and LSTM, which comprises the following steps: acquiring time sequence data, preprocessing the time sequence data to obtain a time sequence sample set meeting the requirement of establishing prediction model data, and dividing a training set and a test set to obtain a first training set and a first test set; establishing a first neural network for predicting trend components and residual items based on the LSTM, training and adjusting parameters through a first training set, predicting the first training set through a trained first neural network model, obtaining the prediction results of the trend components and the residual items of the first training set, and further processing the prediction results into a second training set; step three, establishing a second neural network based on the ANN, and training and parameter adjustment are carried out through a second training set; and step four, performing combined prediction on the test set by the trained first neural network model and the trained second neural network model to obtain a fitting time series data prediction result.
Description
Technical Field
The invention relates to a time series data prediction method based on time series decomposition and LSTM, and belongs to the technical field of time series data prediction.
Background
The time series data prediction is to simulate the change rule of data through the analysis of historical data and then predict the data at a future time point. Time series prediction has important significance in stock analysis, air pollution monitoring, traffic flow scheduling and other applications.
Common time series data prediction methods include statistical methods and machine learning methods. In recent years, with the continuous development of deep learning, a Recurrent Neural Network (RNN) and a Long Short Term Memory (LSTM) Network have shown good prediction performance in data prediction. However, the conventional method cannot well predict a complex nonlinear time sequence, the performance of the deep learning method is improved, but the training time is long and the phenomenon of overfitting exists. In addition, the problem of prediction hysteresis still exists, the problem cannot be adapted to the change characteristics of time sequence data, and particularly the prediction of non-stationary burst data is not accurate.
Time series data decomposition plays an important role in time series analysis. Time series data is typically broken down into 3 parts: a trend component, a periodic component, and a remainder component.
Disclosure of Invention
The invention designs and develops a time sequence data prediction method based on time sequence decomposition and LSTM, which separates the trend and the period of time data by adopting the time sequence decomposition and combines the time sequence decomposition with the LSTM, thereby effectively improving the accuracy of prediction, supporting the pre-scheduling of traffic flow through the real-time prediction of traffic road conditions in a traffic flow monitoring system and solving the traffic jam phenomenon.
The technical scheme provided by the invention is as follows:
a method for predicting time series data based on time series decomposition and LSTM, comprising:
acquiring time sequence data, establishing a time sequence sample set of prediction model data requirements, and dividing a training set and a test set to obtain a first training set and a first test set;
establishing a first neural network for predicting trend components and residual items based on the LSTM, training and adjusting parameters through a first training set, predicting the first training set through a trained first neural network model, obtaining the prediction results of the trend components and the residual items of the first training set, and further processing the prediction results into a second training set;
step three, establishing a second neural network based on the ANN, and training and parameter adjustment are carried out through a second training set to obtain a trained second neural network;
and step four, performing combined prediction on the first test set by the trained first neural network and second neural network models to obtain a fitted time series data prediction result.
Preferably, the first step includes:
performing first-order difference and second-order difference on the original time sequence data, and decomposing the original time sequence data to obtain a periodic component, a trend component and a remainder;
determining the input and the output of the first neural network by a window sliding method, setting the window size as n, sliding the window 1 every time, and inputting as follows: the output of n-1 six kinds of data in front of the window is as follows: the nth trend component and the nth remainder of the window;
wherein, six kinds of data include: the method comprises the steps of obtaining an original sequence, a first-order difference sequence, a second-order difference sequence, a periodic component sequence, a trend component sequence and a remainder sequence;
dividing input data samples and output data samples of the first neural network according to the ratio of 0.95:0.05, and respectively carrying out normalization to obtain a first training set and a first testing set.
Preferably, the second step includes:
training and parameter adjusting are carried out on the first neural network model by using a first training set, so that the trained first neural network model is obtained;
and predicting trend components and remainder of the first training set and the first testing set through the trained first neural network.
Preferably, the third step includes:
the trend component, the remainder and the corresponding periodic component which are obtained by the prediction of the first neural network are used as the input of a second neural network, the corresponding original time sequence data is used as the output of the second neural network, and the second neural network is constructed based on the ANN;
and training and parameter adjustment are carried out on the second neural network through the second training set, so that the trained second neural network is obtained.
Preferably, the fourth step includes:
and processing the first test set by using the first neural network, taking the predicted trend component, the predicted remainder and the corresponding periodic component as the input of the second neural network, and taking the corresponding original time sequence data as the output of the second neural network to obtain a second test set.
And predicting the second test set through the trained second neural network, and performing inverse normalization on the prediction result to obtain a final fitted time series data prediction result.
Preferably, in the first step, the original time series data is decomposed by an addition model of a decomposition method, where the addition model is:
f(t)＝S(t)+T(t)+R(t)；
wherein S (t) is a period component, T (t) is a trend component, and R (t) is a remainder.
Preferably, in the first step, the period is selected to be 60 when decomposing the time series, and the moving average is used when decomposing the time trend.
The invention has the following beneficial effects:
1) the invention separates the trend component, the cycle component and the remainder of the traffic flow/network flow time sequence data by utilizing time sequence decomposition, predicts the trend component and the remainder, and then synthesizes the predicted trend component, the remainder and the cycle component into time sequence data. The neural network is combined with time sequence difference and decomposition, so that the overall data prediction is changed into 3 independent partial predictions, the traffic flow/network flow prediction difficulty is reduced, and the traffic flow/network flow prediction precision is effectively improved.
2) The invention utilizes time sequence decomposition to obtain definite periodicity of the traffic flow/network flow, has better prediction effect under the condition that the time sequence data of the traffic flow/network flow is periodically and suddenly increased, and improves the performance of traffic scheduling/network monitoring.
3) In a streaming computing platform, the application-oriented active elastic resource scheduling is based on the prediction of input stream data, and accurate prediction is beneficial to improving the performance of resource scheduling, so that the resource can be saved, and the energy consumption of a system can be further saved.
4) In a traffic flow monitoring system, the traffic flow is pre-scheduled by predicting the traffic road conditions in real time, so that the traffic jam phenomenon can be solved
Drawings
FIG. 1 is a schematic diagram of the prediction method of the present invention
FIG. 2 is a schematic flow chart of a time series data prediction method according to an embodiment of the present invention
FIG. 3 is a time-series decomposition result chart of the first 500 data according to the embodiment of the present invention
FIG. 4 is a comparison graph of predicted results according to the embodiment of the present invention
FIG. 5 is a graph comparing results predicted by conventional LSTM network
Detailed Description
The present invention is further described in detail below with reference to the attached drawings so that those skilled in the art can implement the invention by referring to the description text.
As shown in fig. 1 to 5, the present invention provides a time series data prediction method based on time series decomposition and LSTM, which separates the trend of time data from the period by using time series decomposition and combines with LSTM, thereby effectively improving the accuracy of prediction.
Adopting python software, the version number is: python3.7, tensoflow 1.14, running software environment: windows10, hardware configuration: the processor AMD Ryzen 54600H with radio Graphics (12CPUs), 3.0GHz, the memory 16G RAM, the video card NVIDIA GeForce GTX 1650;
the first neural network model: as shown in the LSTM network in fig. 1: layer 1 is the LSTM layer, 100 neurons. The layer 2 is a Dense layer and has 50 neurons, the layer 3 is divided into two parts, one part is used for predicting trend components, the part of the embodiment of the invention is to fit the trend components by using two Dense layers, the layer 1 has 24 neurons, the layer 2 has 1 neuron, and the prediction result of the trend components is output; and the other part is used for predicting the residual term, the part is used for fitting the residual term by using 1 Dense layer, and only one neuron is used for outputting the prediction result of the residual term.
Inputting a format: [ None, 60, 6], where 6 represents 6 input data: an original sequence, a first order difference sequence, a second order difference sequence, a periodic component sequence, a trend component sequence, and a remainder sequence. And 60 refers to the first 60 data in each sliding window. None indicates the number of sets per input.
And (3) output format: [ None,2], where 2 denotes the prediction of the 61 st trend component and the remainder in the sliding window by the neural network, and None denotes the number of groups per output.
A second neural network model: as shown in the ANN network in fig. 2, in the embodiment of the present invention, 3 layers are provided, which are all a sense layer, and the number of neurons is: 50, 24,1. The last layer is an output layer for outputting the prediction result.
Inputting a format: [ None, 3], where 3 denotes 3 input data: two outputs of the first neural network model and the 61 st periodic component in each sliding window. None indicates the number of sets per input.
And (3) output format: [ None, 1], where 1 denotes the prediction of the 61 st raw time series data in the sliding window by the second neural network, and None denotes the number of groups output at a time.
The prediction model is composed of two neural network models, a first neural network model for predicting the trend component and the remainder, and a second neural network model for fitting the trend component, the periodic component, and the remainder to the corresponding time series data. When experimental data is processed, two sets of training sets and test sets are used, namely a first training set and a first test set for training and testing a first neural network model and a second training set and a second test set for training and testing a second neural network model.
Because the input of the second neural network model needs to use the output of the first neural network model, the second training set and the second test set can be obtained only after the training of the first neural network model is completed, and the method specifically comprises the following steps:
acquiring time sequence data, preprocessing the time sequence data to obtain a time sequence sample set meeting the requirement of establishing prediction model data, and dividing a training set and a test set;
the experimental data are from text texts in the COVID19_ twitter data set, the contents of the text texts are tweets related to COVID-19 on tweets, the time range is 31 days from 2021-08-01 days, and the experimental data adopt English text data in order to enrich data samples considering that more countries and regions use English. The daily data is about 15-20 ten thousand. When a specific tweet is acquired, a tweet developer account needs to be applied, and in the application process, a plurality of mails are sent to indicate the purpose of the acquired data and other information. COVID-19 _ dataset _ Tutorial in this GitHub project (https:// GitHub. com/the panda/COVID 19_ widget) can be used to download quickly to specific data.
The original data comprises praise number, platform, text, comment, forwarding information, number and creation time of the tweet.
Processing the original data, sequencing the original data according to time, and calculating the number of tweets per minute to obtain original time sequence data;
performing first-order difference and second-order difference on the original time sequence data;
decomposing original time sequence data by using an addition model of a classical decomposition method to obtain a periodic component, a trend component and a remainder;
determining the input and the output of the first neural network by a window sliding method, setting the window size as n, sliding the window 1 every time, and inputting as follows: the output of n-1 six kinds of data in front of the window is as follows: the nth trend component and the nth remainder of the window, n > 1;
wherein, the addition model is as follows: (t) ═ s (t) + t (t) + r (t);
wherein S (t) is a period component, T (t) is a trend component, and R (t) is a remainder;
selecting the period as 60 when decomposing the time sequence, namely taking one hour as one period;
when decomposing the trend component, the moving average value is used, because the known data is needed to predict the unknown data, when decomposing the time series, the predicted data can not be used as the input of the decomposition, and when calculating the moving average value, only the past value is used for calculation;
when the moving average value is calculated, the average value can be calculated by using the forward k values and the backward k values of the current data, if the decomposition is carried out in this way, the decomposed three components contain the information of the future time sequence data, and k is more than or equal to 1.
The time-series decomposition result of the first 500 data is shown in fig. 3, since the calculation is performed only by using the past values when calculating the moving average, the first 60 data trend components and the remainder are null values and are filled with 0;
obtaining input and output of a first neural network by adopting six data of an original sequence, a first-order difference sequence, a second-order difference sequence, a periodic component sequence, a trend component sequence and a remainder sequence by adopting a sliding window method, wherein the window size is 61, the sliding is 1 every time, the first 60 data of the window are used as the input of the first neural network, and the trend component and the remainder of the 61 st data of the window are used as the output of the first neural network;
dividing input and output data samples of the first neural network according to the ratio of 0.95:0.05 to obtain a first training set and a first testing set;
deleting the 0 value of the abnormal data and 60 data before and after the abnormal data, and normalizing the training set and the test set by using a Min-Max Normalization method to obtain a first training set and a first test set;
step two, establishing a first neural network for trend component and remainder prediction based on LSTM, wherein the model structure is shown as LSTM in figure 1, the first layer is an LSTM layer, the second layer is a Dense layer, and the third layer is divided into two parts, wherein one part is used for predicting the trend component, and the part is used for fitting the trend component by using two Dense layers in the embodiment of the invention; another part is used to predict the residuals, which is to fit the residuals with a sense layer. Training and parameter adjustment are carried out through a first training set, and the trained first neural network model is used for predicting the first training set to obtain the prediction results of trend components and remainder items of the first training set;
training and parameter adjustment are carried out on the first neural network model by utilizing a first training set, the learning rate is set to be 0.001, and the training times are 20 times;
using trend components and remainder items in a first training set and a first test set obtained by predicting the first neural network model and corresponding periodic components as the input of a second neural network model, and using corresponding original time sequence data as the output of a second neural network membrane to obtain a second training set;
step three, establishing a second neural network based on the ANN, wherein the model structure is shown as the ANN in the figure 1, and because the fitting of the original time sequence data to the three decomposition quantities is simple, three Dense layers are connected into the ANN, and training and parameter adjustment are carried out through a second training set;
the input of the second neural network model is 3, which are respectively a trend component, a remainder and a period component of 61 th data of a sliding window, wherein the trend component and the remainder are prediction results of the first neural network, the period component has periodicity, prediction is not needed, and one output is the 61 st original time sequence data in the sliding window, namely the data needing to be predicted;
training and parameter adjusting are carried out on the second neural network model through a second training set, the learning rate is set to be 0.001, and the training times are 20 times;
step four, performing combined prediction on the first test set by the trained first neural network and second neural network models to obtain a fitted time series data prediction result;
processing the first test set by using a first neural network, taking the predicted trend component, remainder and corresponding periodic component as the input of a second neural network, and taking corresponding original time sequence data as the output of a second neural network to obtain a second test set;
and predicting the second test set by using the second neural network model, carrying out inverse normalization on the prediction, and comparing the result with a real result.
And after the first test set is input into the first neural network, processing the output result and part of data in the first step to obtain a second test set.
As shown in FIG. 4, the prediction results of the first 500 data of the test set show that the peak position is well predicted, and the problem of prediction hysteresis of the traditional neural network is well solved.
As shown in fig. 5, in contrast to the conventional model using LSTM alone:
both prediction methods are measured from four indicators of Mean Square Error (MSE), mean square error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R2).
Evaluation results of the conventional LSTM model: MSE:343.57, RMSE:18.53, MAE:13.41, R2:0.717, model evaluation index value of the invention: the MSE is 192.19, the RMSE is 13.86, the MAE is 10.54, and the R2 is 0.842, wherein the smaller the numerical values of the first three of the four measurement indexes are, the smaller the error between the obtained predicted value and the observed value is, the closer the last R2 is to 1, the better the prediction result is, and the accuracy of the final prediction result can be effectively improved by the time series decomposition and LSTM-based time series data prediction method through the comparison data of the four measurement indexes.
In summary, the present invention provides a time series data prediction method based on time series decomposition and LSTM, in which, when time series data is predicted, time series data is decomposed first, the decomposition amount is predicted, and finally, the predicted decomposition amount is synthesized to obtain a prediction of the time series data. Compared with the existing LSTM, the method has better performance in predicting the flow data. In a streaming computing platform, the application-oriented active elastic resource scheduling is based on the prediction of input stream data, and accurate prediction is beneficial to improving the performance of resource scheduling, so that the resource can be saved, and the energy consumption of a system can be further saved.
While embodiments of the invention have been described above, it is not limited to the applications set forth in the description and the embodiments, which are fully applicable in various fields of endeavor to which the invention pertains, and further modifications may readily be made by those skilled in the art, it being understood that the invention is not limited to the details shown and described herein without departing from the general concept defined by the appended claims and their equivalents.
Claims (7)
1. A time series data prediction method based on time series decomposition and LSTM is characterized by comprising the following steps:
acquiring time sequence data, establishing a time sequence sample set of prediction model data requirements, and dividing a training set and a test set to obtain a first training set and a first test set;
establishing a first neural network for predicting trend components and residual items based on the LSTM, training and adjusting parameters through a first training set, predicting the first training set through a trained first neural network model, obtaining the prediction results of the trend components and the residual items of the first training set, and further processing the prediction results into a second training set;
step three, establishing a second neural network based on the ANN, and training and parameter adjustment are carried out through a second training set to obtain a trained second neural network;
and step four, performing combined prediction on the first test set by the trained first neural network and second neural network models to obtain a fitted time series data prediction result.
2. The method for predicting time series data based on time series decomposition and LSTM according to claim 1, wherein said first step comprises:
performing first-order difference and second-order difference on the original time sequence data, and decomposing the original time sequence data to obtain a periodic component, a trend component and a remainder;
determining the input and the output of the first neural network by a window sliding method, setting the window size as n, sliding the window 1 every time, and inputting as follows: the output of n-1 six kinds of data in front of the window is as follows: the nth trend component and the nth remainder of the window;
wherein, six kinds of data include: the method comprises the steps of obtaining an original sequence, a first-order difference sequence, a second-order difference sequence, a periodic component sequence, a trend component sequence and a remainder sequence;
dividing input data samples and output data samples of the first neural network according to the ratio of 0.95:0.05, and respectively carrying out normalization to obtain a first training set and a first testing set.
3. The method for predicting time series data based on time series decomposition and LSTM according to claim 2, wherein said second step comprises:
training and parameter adjusting are carried out on the first neural network model by using a first training set, so that the trained first neural network model is obtained;
and predicting trend components and remainder of the first training set and the first testing set through the trained first neural network.
4. The method of predicting time series data based on time series decomposition and LSTM according to claim 3, wherein the third step comprises:
the trend component, the remainder and the corresponding periodic component which are obtained by the prediction of the first neural network are used as the input of the second neural network, the corresponding original time series data are used as the output of the second neural network, and the second neural network is constructed based on the ANN;
and training and parameter adjustment are carried out on the second neural network through the second training set, so that the trained second neural network is obtained.
5. The method of predicting time series data based on time series decomposition and LSTM according to claim 4, wherein the fourth step comprises:
and processing the first test set by using the first neural network, taking the predicted trend component, the predicted remainder and the corresponding periodic component as the input of the second neural network, and taking the corresponding original time sequence data as the output of the second neural network to obtain a second test set.
And predicting the second test set through the trained second neural network, and performing inverse normalization on the prediction result to obtain a final fitted time series data prediction result.
6. The method of predicting time series data according to claim 5, wherein in the first step, the original time series data is decomposed by an additive model of decomposition method, the additive model is:
f(t)＝S(t)+T(t)+R(t)；
wherein S (t) is a period component, T (t) is a trend component, and R (t) is a remainder.
7. The method of claim 6, wherein in the first step, the period is 60 when time series are decomposed, and the moving average is used when time trend is decomposed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111602708.0A CN114239990A (en) | 2021-12-24 | 2021-12-24 | Time series data prediction method based on time series decomposition and LSTM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111602708.0A CN114239990A (en) | 2021-12-24 | 2021-12-24 | Time series data prediction method based on time series decomposition and LSTM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114239990A true CN114239990A (en) | 2022-03-25 |
Family
ID=80762812
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111602708.0A Pending CN114239990A (en) | 2021-12-24 | 2021-12-24 | Time series data prediction method based on time series decomposition and LSTM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114239990A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115118511A (en) * | 2022-06-29 | 2022-09-27 | 北京奇艺世纪科技有限公司 | Abnormal flow identification method and device, electronic equipment and storage medium |
| CN116502774A (en) * | 2023-06-26 | 2023-07-28 | 南京信息工程大学 | Time sequence prediction method based on time sequence decomposition and Legend projection |
-
2021
- 2021-12-24 CN CN202111602708.0A patent/CN114239990A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115118511A (en) * | 2022-06-29 | 2022-09-27 | 北京奇艺世纪科技有限公司 | Abnormal flow identification method and device, electronic equipment and storage medium |
| CN116502774A (en) * | 2023-06-26 | 2023-07-28 | 南京信息工程大学 | Time sequence prediction method based on time sequence decomposition and Legend projection |
| CN116502774B (en) * | 2023-06-26 | 2023-09-12 | 南京信息工程大学 | Time sequence prediction method based on time sequence decomposition and Legend projection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109472110B (en) | Method for predicting residual service life of aeroengine based on LSTM network and ARIMA model | |
| Heck et al. | Simplified minimal gated unit variations for recurrent neural networks | |
| Hao et al. | A hybrid EMD-LSTM model for non-stationary wave prediction in offshore China | |
| CN114239990A (en) | Time series data prediction method based on time series decomposition and LSTM | |
| CN108268935A (en) | A kind of PM2.5 concentration values Forecasting Methodology and system based on sequential Recognition with Recurrent Neural Network | |
| CN112364975A (en) | Terminal operation state prediction method and system based on graph neural network | |
| CN111814956B (en) | Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction | |
| CN112036075A (en) | Abnormal data judgment method based on environmental monitoring data association relation | |
| CN110633859B (en) | Hydrologic sequence prediction method integrated by two-stage decomposition | |
| CN110443419A (en) | Medium-term and long-term Runoff Forecast method based on ICEEMDAN and extreme learning machine | |
| CN114117355A (en) | Optimization method, system, equipment and readable storage medium of time-varying-resistance model | |
| CN115098999A (en) | Multi-mode fusion fuel cell system performance attenuation prediction method | |
| Radev et al. | Bayesflow: Amortized bayesian workflows with neural networks | |
| Zhou et al. | Functional networks and applications: A survey | |
| CN115456245A (en) | Prediction method for dissolved oxygen in tidal river network area | |
| CN111141879A (en) | Deep learning air quality monitoring method, device and equipment | |
| Yang et al. | A prediction model of aquaculture water quality based on multiscale decomposition | |
| Huang et al. | A decomposition‐based multi‐time dimension long short‐term memory model for short‐term electric load forecasting | |
| CN114298299A (en) | Model training method, device, equipment and storage medium based on course learning | |
| CN112949932A (en) | Ship traffic flow prediction method and device, computer equipment and storage medium | |
| Varuna et al. | Trend prediction of GitHub using time series analysis | |
| Yarlagadda et al. | Power system state estimation and forecasting using cnn based hybrid deep learning models | |
| CN115389743A (en) | Method, medium and system for predicting content interval of dissolved gas in transformer oil | |
| WO2022222230A1 (en) | Indicator prediction method and apparatus based on machine learning, and device and storage medium | |
| CN112616160B (en) | Intelligent short-wave frequency cross-frequency-band real-time prediction method and system |
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 |