CN112699614B - XGboost-based sequence prediction model construction and precipitation trend prediction method and device - Google Patents
XGboost-based sequence prediction model construction and precipitation trend prediction method and device Download PDFInfo
- Publication number
- CN112699614B CN112699614B CN202110046489.6A CN202110046489A CN112699614B CN 112699614 B CN112699614 B CN 112699614B CN 202110046489 A CN202110046489 A CN 202110046489A CN 112699614 B CN112699614 B CN 112699614B
- Authority
- CN
- China
- Prior art keywords
- sequence
- stalagmite
- oxygen isotope
- data
- obtaining
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Other Investigation Or Analysis Of Materials By Electrical Means (AREA)
Abstract
The invention discloses a method and a device for constructing a sequence prediction model and predicting precipitation trend based on XGboost, wherein a precipitation amount is predicted by using a value of a stalagmite, a stalagmite isotope can simulate the change of the precipitation amount, and meanwhile, data of the isotope of the stalagmite can be collected in the past decades, so that a large amount of data is more convenient for the prediction research of the data.
Description
Technical Field
The invention relates to a precipitation prediction method and a precipitation prediction device, in particular to a XGboost-based sequence prediction model construction method and a precipitation trend prediction method and a device.
Background
Precipitation changes have direct effects on surface runoff and crop growth, and are also closely related to the stable development of social economy and human production life. Numerous studies have shown that: flood or drought events in the historical period have important influence on civilization aging and dynasty alternation. In recent years, with global warming, extreme climatic events are frequent, and the development of agriculture, population and economy is seriously influenced. In order to prevent drought and waterlogging disasters, the prediction and prediction research on future rainfall change needs to be enhanced urgently.
The traditional precipitation sequence analysis and prediction methods are applied to a plurality of fields with certain success, but once complex data are faced, the methods are not satisfactory and cannot meet the prediction precision requirement.
At present, the following methods are commonly used:
1. such as long-term trend fluctuation, periodic fluctuation, seasonal fluctuation, and the like. And then the three components are integrated to predict the result. However, in this method, although the overall trend of the prediction result is generally close, the local fluctuation effect of the prediction result is relatively poor.
2. By adopting an ARIMA method (autoregressive moving average model), the ARIMA model is established on the basis of a stable time sequence, so that the stationarity of the time sequence is an important premise for modeling. The method for checking the stability of the time series model generally adopts an ADF unit root checking model to check. Of course, if the time series is not stable, the time series can be stabilized by some operations (such as taking the logarithm and the difference), then the ARIMA model prediction is performed to obtain the prediction result of the stable time series, and then the inverse operation of the previous operation (taking the logarithm and the difference) for stabilizing the time series is performed on the prediction result, so that the prediction result of the original data can be obtained.
On the other hand, using ARIMA model prediction, the time series data is required to be stable (static), or stable after differentiation (differentiating). It essentially captures only linear relationships, not nonlinear relationships. If the time sequence data is unstable, the regularity cannot be captured.
In summary, most of the existing prediction methods adopt the traditional mathematical methods to perform prediction analysis on the sequence, which has high requirements on the quality of information data, and the incorrect value of the historical value can cause the predicted value to have larger deviation. Meanwhile, the predicted value is related to the near term and is not related to the data before the calculation period, which is not in accordance with the objective condition. In addition, most methods adopting machine learning all adopt a single learning model, and the methods cannot well solve the problems of nonlinearity and non-stationarity when predicting long-term periodic time sequences, only can roughly display the approximate trend of data, and have poor local detailed information change performance.
Disclosure of Invention
The invention aims to provide a method and a device for constructing a sequence prediction model based on XGboost and predicting a precipitation trend, which are used for solving the problem of inaccurate trend prediction of a precipitation trend prediction method in the prior art.
In order to realize the task, the invention adopts the following technical scheme:
a sequence prediction model construction method based on XGboost is used for obtaining a prediction model of a random item sequence in a stalagmite oxygen isotope sequence, and the method is implemented according to the following steps:
step 2, performing data quality improvement on the stalagmite oxygen isotope sequence y' to obtain a quality-improved stalagmite oxygen isotope sequence y * ；
Step 3, enabling the promoted oxygen isotope sequence y of the stalagmite bamboo shoot * Performing data reconstruction to obtain a reconstructed stalagmite oxygen isotope sequence y, wherein the reconstructed stalagmite oxygen isotope sequence y comprises a random item sequence and a periodic item sequence;
step 4, obtaining the t item x of the random item sequence by adopting the formula II t ：
Wherein y is t Represents the t value, p of the reconstructed oxygen isotope sequence of the stalagmite shoots obtained in the step 3 t Value representing the t-th periodic item sequence, a 0 Denotes the initial value of the period term, a k Coefficient representing the cosine function corresponding to the kth harmonic, b k Coefficients representing a sine function corresponding to the k-th harmonic, k =1,2, …M, M represents the number of harmonics, and M is an integer greater than 0;
taking the q +1 th column data of the random item sequence x corresponding to each sample data as tag data, obtaining N-1 tag data, and obtaining a tag set;
and 7, taking the sample set as input, taking the label set as reference output, training an XGboost model, and obtaining a random item sequence prediction model.
Further, the step 2 specifically includes:
2.1, repairing the oxygen isotope sequence y' of the stalagmite bamboo shoot by adopting a nearest neighbor interpolation method to obtain a repaired oxygen isotope sequence of the stalagmite bamboo shoot;
2.2, performing trend removing treatment on the repaired stalagmite oxygen isotope sequence by adopting a least square method to obtain a quality-improved stalagmite oxygen isotope sequence y * 。
3. The XGboost-based sequence prediction model construction method according to claim 1, wherein the step 3 specifically comprises:
step 3.1, improving the quality of the oxygen isotope sequence y of the stalagmite shoots * After conversion to the trajectory matrix X, a feature matrix X' is obtained using formula I:
X'＝X T * X formula I
Step 3.2, carrying out singular value decomposition on the characteristic matrix X' to obtain m characteristic values, wherein m is a positive integer;
and 3.3, after the M characteristic values are arranged from large to small, selecting the first M characteristic values to perform matrix reconstruction, and obtaining a reconstructed stalagmite oxygen isotope sequence y.
Further, M =28.
A precipitation tendency prediction method is used for predicting precipitation tendency of a to-be-predicted area moment j, wherein j is a positive integer, and the method is executed according to the following steps:
step A, obtaining a stalagmite oxygen isotope value from time 1 to time j-1 of an area to be predicted to obtain a stalagmite oxygen isotope sequence;
acquiring a stalagmite oxygen isotope value of a to-be-predicted region moment j;
step B, processing the oxygen isotope sequence of the stalagmite bamboo shoot obtained in the step A by adopting a method in steps 2-5 in a sequence prediction model construction method based on XGboost to obtain a random term sequence prediction model and a periodic term sequence p;
inputting the oxygen isotope value of the stalagmite bamboo shoot at the time j of the area to be predicted into a random item sequence prediction model to obtain a random item sequence x' at the time j;
step C, inputting the stalagmite oxygen isotope at the moment j into the periodic item sequence p to obtain a periodic item sequence p' at the moment j;
d, obtaining a precipitation trend prediction function l, l = p '+ x';
and E, judging the slope of the rainfall trend prediction function l, if the slope is smaller than 0, increasing the rainfall of the area j to be predicted at the moment, and otherwise, decreasing the rainfall of the area j to be predicted at the moment.
A sequence prediction model construction device based on XGboost comprises a data acquisition module, a data quality improvement module, a data reconstruction module, a sequence acquisition module, a data set acquisition module and a model training module;
the data acquisition module is used for acquiring N oxygen isotope values of the stalagmite shoots and acquiring a stalagmite shoot oxygen isotope sequence y' = { y = 1 '，y 2 '，…，y t '，…，y N '}，y t ' represents the t-th isotope value of the stalagmite oxygen isotope sequence, t =1,2, …, N is a positive integer;
the data quality promotion module is used for promoting the data quality of the stalagmite oxygen isotope sequence y' to obtain the quality-promoted stalagmite oxygen isotope sequence y * ；
The data reconstruction module is used for reconstructing the promoted oxygen isotope sequence y of the stalagmite bamboo shoots * Performing data reconstruction to obtain a reconstructed stalagmite oxygen isotope sequence y, wherein the reconstructed stalagmite oxygen isotope sequence y comprises a random item sequence and a periodic item sequence;
the sequence obtaining module is used for obtaining the t term x of the random term sequence by adopting a formula II t ：
Wherein y is t Denotes the t value, p of the obtained reconstructed oxygen isotope sequence of the stalagmite t Value representing the t-th periodic item sequence, a 0 Denotes the initial value of the period term, a k Coefficient representing the cosine function corresponding to the kth harmonic, b k Representing coefficients of a sine function corresponding to the kth harmonic, wherein k =1,2, …, M represents the number of harmonics, and M is an integer greater than 0;
let t = t +1 until t = N, obtaining a random term sequence x and a periodic term sequence p;
the data set obtaining module is used for selecting the first q columns of data of the random item sequence x as sample data, wherein q =1,2, … and N-1, obtaining N-1 sample data and obtaining a sample set;
taking the q +1 th column data of the random item sequence x corresponding to each sample data as tag data, obtaining N-1 tag data, and obtaining a tag set;
and the model training module is used for training the XGboost model by taking the sample set as input and the label set as reference output to obtain a random term sequence prediction model.
Further, the data quality improving module specifically comprises a repairing sub-module and a trend removing processing sub-module;
the repairing submodule is used for repairing the stalagmite oxygen isotope sequence y' by adopting a nearest neighbor interpolation method to obtain a repaired stalagmite oxygen isotope sequence;
the de-trend processing submodule is usedPerforming trend removing treatment on the repaired oxygen isotope sequence of the stalagmite bamboo shoot by adopting a least square method to obtain the oxygen isotope sequence y of the stalagmite bamboo shoot with improved quality * 。
Further, the data reconstruction module specifically comprises a track matrix conversion submodule, a singular value decomposition submodule and a matrix reconstruction submodule;
the track matrix conversion submodule is used for converting the mass-improved stalagmite oxygen isotope sequence y * After conversion to the trajectory matrix X, a feature matrix X' is obtained using formula I:
X'＝X T * X formula I
The singular value decomposition submodule is used for carrying out singular value decomposition on the characteristic matrix X' to obtain m characteristic values, wherein m is a positive integer;
and the matrix reconstruction submodule is used for arranging the M characteristic values in a descending order and then selecting the first M characteristic values to carry out matrix reconstruction to obtain a reconstructed stalagmite oxygen isotope sequence y.
Further, M =28.
A precipitation trend prediction device is used for predicting the precipitation trend of a time j of an area to be predicted, wherein j is a positive integer, and the device comprises a sequence acquisition module, a random item prediction sequence acquisition module, a periodic item prediction sequence acquisition module, a prediction function acquisition module and a prediction module;
the sequence acquisition module is used for acquiring the oxygen isotope value of the stalagmite bamboo shoot from the moment 1 to the moment j-1 of the area to be predicted and acquiring an oxygen isotope sequence of the stalagmite bamboo shoot;
the sequence acquisition module is also used for acquiring the oxygen isotope value of the stalagmite oxygen at the moment j of the area to be predicted;
the random term prediction sequence obtaining module is used for processing the obtained stalagmite oxygen isotope sequence by adopting the method in the XGboost-based sequence prediction model construction device of any one of claims 6 to 9 to obtain a random term sequence prediction model and a periodic term sequence p;
the method is also used for inputting the oxygen isotope value of the stalagmite bamboo shoot at the time j of the area to be predicted into a random item sequence prediction model to obtain a random item sequence x' at the time j;
the period item prediction sequence obtaining module is used for inputting the stalagmite oxygen isotope at the moment j into the period item sequence p to obtain a period item sequence p' at the moment j;
the prediction function obtaining module is used for obtaining a precipitation trend prediction function l, l = p '+ x';
the prediction module is used for judging the slope of the precipitation trend prediction function l, if the slope is smaller than 0, the precipitation quantity of the area j to be predicted rises at the moment, otherwise, the precipitation quantity of the area j to be predicted falls at the moment.
Compared with the prior art, the invention has the following technical effects:
1. the method and the device for constructing the sequence prediction model based on the XGboost and predicting the precipitation tendency provided by the invention have the advantages that the periodicity of the periodic time sequence can be well simulated, and the local burr change can be well predicted by using the random term, so that the prediction accuracy is improved.
2. The method for predicting precipitation by using the value of the stalagmite provided by the XGboost-based sequence prediction model construction and precipitation trend prediction method and device provided by the invention has the advantages that the stalagmite isotope can simulate the change of precipitation, meanwhile, data of the stalagmite isotope in the past decades can be acquired, and a large amount of data is more convenient for the prediction research of the data.
3. The method for constructing the sequence prediction model based on the XGboost and predicting the rainfall trend and the device for improving the data quality provided by the method have the advantages that the abnormal 0 value of the data is found, and then the abnormal data is filled by using a nearest neighbor method.
4. The method for constructing the sequence prediction model based on the XGboost and predicting the rainfall trend and the method for obtaining the random term provided by the device have the advantages that the method for decomposing the singular spectrum is used for obtaining the periodic function of the signal, and then the characteristic of the combination of the trigonometric functions is used for well expressing the periodicity of the signal, so that the accurate prediction method is obtained.
Drawings
FIG. 1 is a flow chart illustrating a prediction method according to the present invention;
FIG. 2 is a raw data graph provided in one embodiment of the present invention;
FIG. 3 is a schematic representation of a repaired sequence of oxygen isotopes of stalagmite provided in an embodiment of the present invention;
FIG. 4 is a schematic diagram of a sequence of oxygen isotopes of stalagmite provided in an embodiment of the present invention after mass enhancement;
FIG. 5 is a schematic diagram of variance contribution ratios provided in an embodiment of the present invention;
FIG. 6 is a graph of different q-valued scores provided in an embodiment of the present invention;
FIG. 7 is a diagram illustrating the prolongation result of the periodic item sequence provided in an embodiment of the present invention;
FIG. 8 is a diagram illustrating the results of random sequence prediction provided in one embodiment of the present invention;
FIG. 9 is a comparison of different model predictions provided in an embodiment of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and examples. So that those skilled in the art can better understand the present invention. It is to be expressly noted that in the following description, a detailed description of known functions and designs will be omitted when it may obscure the main content of the present invention.
The following definitions or conceptual connotations relating to the present invention are provided for illustration:
and (3) a stalagmite oxygen isotope sequence: stalagmite delta 18 O value sequence, commonly used at present, shisun delta 18 Fluctuations in the value of O predict climate change.
And (3) data quality improvement: various problems such as data omission, distortion, incompleteness, relevance and logical inconsistency and the like exist in data quality, and abnormal value detection and restoration are carried out on data through data quality improvement, so that the data have accuracy and authenticity.
Nearest neighbor interpolation: also called k-neighbor interpolation, i.e. searching k observation point samples nearest to the point to be estimated, and assigning the point to be estimated by the weighted sum of the observation values of these sample points.
Least square method: a mathematical optimization technique. It finds the best functional match of the data by minimizing the sum of the squares of the errors. Unknown data can be simply obtained by using a least square method, and the square sum of errors between the obtained data and actual data is minimum, so that the method can be used for curve fitting to realize optimization problems.
Example one
The embodiment discloses a sequence prediction model construction method based on XGboost, which is used for obtaining a prediction model of a random term sequence in a stalagmite oxygen isotope sequence.
As shown in fig. 1, in the present invention, a stalagmite oxygen isotope sequence is used to predict a precipitation variation trend, and the stalagmite oxygen isotope sequence is split into a periodic item sequence and a random item sequence; and predicting a random term sequence by using the model so as to predict the precipitation variation trend.
The method is executed according to the following steps:
in this embodiment, y' = { -5.5, -5.19, … -5.43}.
Step 2, performing data quality improvement on the stalagmite oxygen isotope sequence y' to obtain a quality-improved stalagmite oxygen isotope sequence y * ；
In the invention, when the oxygen isotope of the stalagmite bamboo shoot is collected, the condition that the stalagmite bamboo shoot sample is possibly stored improperly when an isotope mass spectrometer is adopted to measure the stalagmite bamboo shoot sample, so that the isotope data of the stalagmite bamboo shoot can be lost and the like, so that the data quality improvement method special for the isotope sequence of the stalagmite bamboo shoot is provided in the invention to improve the accuracy of the data.
Optionally, the step 2 specifically includes:
2.1, repairing the oxygen isotope sequence y' of the stalagmite bamboo shoot by adopting a nearest interpolation method to obtain a repaired oxygen isotope sequence of the stalagmite bamboo shoot;
in the present embodiment, the abnormal values and the missing values of the data are filled and processed. As shown in fig. 2, the raw data is visualized to find that there are few abnormal values with a value of 0, which are due to abnormal situations existing in the data acquisition stage.
Therefore, the method of nearest neighbor interpolation is utilized in the invention, abnormal data is repaired, and the data is filled by the last data. Data 0 from 1826 and 1925 was changed to data from 1825 and 1946, as shown in fig. 3.
2.2, performing trend removing treatment on the repaired stalagmite oxygen isotope sequence by adopting a least square method to obtain a quality-improved stalagmite oxygen isotope sequence y * 。
In the invention, the raw data of the isotope of the stalagmite bamboo shoot in 1733-2004 are processed by a least square method so as to eliminate the baseline deviation of the linear state and the trend term of a high-order polynomial in the data. The method comprises the following specific steps:
firstly, a solving equation of a trend polynomial is listed by using a least square principle, then a trend term coefficient is calculated by using a matrix method, a trend term fitting curve is obtained, and finally a useful signal can be obtained by subtracting the trend term from an original signal.
The patent takes 1733-2004 Klang hole in south Thailand of Klang for a long time delta 18 And O is recorded as that the original data is subjected to detrending preprocessing, and a least square method is adopted for processing so as to eliminate the baseline deviation of the linear state and the trend item of the high-order polynomial in the data.
The detrending graph of the raw data finally obtained by the method is shown in fig. 4, wherein the straight line in fig. 4 is a trend line, the trend curve formula is y =0.0031x-5.5, and y is delta 18 The value of O, x is year.
Step 3,The lifted oxygen isotope sequence y of the stalagmite * Performing data reconstruction to obtain a reconstructed stalagmite oxygen isotope sequence y, wherein the reconstructed stalagmite oxygen isotope sequence y comprises a random item sequence and a periodic item sequence;
optionally, the step 3 specifically includes:
step 3.1, improving the quality of the oxygen isotope sequence y of the stalagmite shoots * After formula I is converted into trajectory matrix X, a feature matrix X' is obtained using formula I:
X'＝X T *X
in the present embodiment, the time series y t ' converting to the trajectory matrix X is obtained by:
selecting an appropriate window length: m (2. Ltoreq. M. Ltoreq.T), converting the observed one-dimensional time series data into a multi-dimensional sequence: x1,. Xn, (Xi = (yi,.., yi + m-1), n = T-m + 1), resulting in a trajectory matrix:
further, a feature matrix X' = X is obtained T *X。
Step 3.2, carrying out singular value decomposition on the characteristic matrix X' to obtain m characteristic values, wherein m is a positive integer;
and 3.3, after the M characteristic values are arranged from large to small, selecting the first M characteristic values to perform matrix reconstruction, and obtaining a reconstructed stalagmite oxygen isotope sequence y.
In this embodiment, the first M eigenvalues are selected from the M eigenvalues for matrix reconstruction, M (M)<m) the sum of the eigenvalues accounts for 85 percent of the sum of the total eigenvalues, and a reconstructed time series y is obtained as shown in a small formula t M is a positive integer;
optionally, M =28.
In the present embodiment, the optimum value of M is 28.
If the first M typical waveform vectors of the sequence already account for 85% of the total variance contribution, they characterize the dominant oscillation mode of the time series variation, which characterizes the most dominant trend of the sequence. Therefore, when selecting significant harmonics of the model, the range of the number of harmonics M can be determined according to P (i) ≈ 85%.
In this example, the detrending data from 1733-2004 was divided into two parts, and data from 1733-1964 was used to predict and compare data from 1965-2004 with the real data from 1965-2004. The window length is selected to be not more than N/2 generally, the window lengths selected respectively when the front part data is subjected to the singular spectrum analysis are 1-116 (N/2), when the window length is 80, the prediction precision is highest, and the variance contribution rate can be obtained when the selected window length is 80, as shown in FIG. 5. The black curve in the graph is the variance contribution rate, when i is more than or equal to 28, P (i) is more than or equal to 85%, the data processing requirement of the invention is met, namely when the number M of the significant harmonics is 28, the corresponding trigonometric function signal can represent the most main trend of the sequence.
Step 4, obtaining the t item x of the random item sequence by adopting the formula II t ：
Wherein y is t Represents the t value, p of the reconstructed oxygen isotope sequence of the stalagmite shoots obtained in the step 3 t Value representing the t-th periodic item sequence, a 0 Denotes the initial value of the period term, a k Coefficient representing the cosine function corresponding to the kth harmonic, b k Representing coefficients of a sine function corresponding to the kth harmonic, wherein k =1,2, …, M represents the number of harmonics, and M is an integer greater than 0;
in the present embodiment, the meaning of the parameters is as follows:
in the present invention, δ is considered 18 The value of O has a certain periodicity, so that the decomposition of the periodic signal and the random signal is carried out on the value of O.
In formula I, M represents the number of harmonics of the periodic function, and has a value equal to the value of I at the time when the variance contribution ratio is greater than 0.85.
taking the q +1 th column data of the random item sequence x corresponding to each sample data as tag data, obtaining N-1 tag data, and obtaining a tag set;
in this embodiment, a random sequence of terms x t (t =1 … N) into q +1 columns of data, the first q columns are used as characteristic input, the q +1 column is used as a label input into the XGBoost model, and the model is trained.
The embodiment adopts the traversing idea to combine the whole time series in turn. And sequentially predicting data of the (q + 1) th year by using data of the previous q years (q =1 … N-q + 1), predicting the result by using an integrated model, and sequentially comparing the predicted result with a real value to obtain the corresponding RMSE. Finding out the q value corresponding to the minimum RMSE is the optimal q value, i.e. determining the magnitude of the optimal characteristic q value is as shown in fig. 6 below.
And 7, taking the sample set as input, taking the label set as reference output, training an XGboost model, and obtaining a random item sequence prediction model.
The oxygen isotope sequence of the stalagmite shoot is split through the embodiment, the XGboost model is used for training and predicting, and the prediction result is shown in figure 8.
Example two
A precipitation tendency prediction method is used for predicting precipitation tendency of a to-be-predicted area at time j, wherein j is a positive integer.
In the embodiment, the precipitation tendency is predicted by using the oxygen isotope value of the stalagmite.
The method is executed according to the following steps:
step A, obtaining a stalagmite oxygen isotope value from time 1 to time j-1 of an area to be predicted to obtain a stalagmite oxygen isotope sequence;
acquiring a stalagmite oxygen isotope value of a to-be-predicted region moment j;
in the present example, the obtained stalagmite oxygen isotope sequence was y' = -5.5, -5.193, … -5.315}.
Step B, processing the oxygen isotope sequence of the stalagmite bamboo shoot obtained in the step A by adopting the method in the step 2-5 in the XGboost-based sequence prediction model construction method of the embodiment to obtain a random term sequence prediction model and a periodic term sequence p;
inputting the oxygen isotope value of the stalagmite bamboo shoot at the time j of the area to be predicted into a random term sequence prediction model to obtain a random term sequence x' at the time j;
and (5) carrying out sample set reconstruction in the step (6) on the random item x to obtain a sample set and a label set:
and training the XGboost model by using the sample set and the label set to obtain a random item sequence prediction model and a periodic item sequence p.
Inputting the oxygen isotope value of the stalagmite bamboo shoot at the time j of the area to be predicted into a random term sequence prediction model to obtain a random term sequence x' at the time j.
In the present embodiment, x' = {0.123,0.285, … 0.121.121 }, for a total of 232 random data.
p = { -5.623, -5.478, … -5.436}, total 232 periods of data.
Where p and x are both variables in step 5.
Step C, inputting the stalagmite oxygen isotope at the moment j into the periodic item sequence p to obtain a periodic item sequence p' at the moment j;
the periodic function is known from formula II
In the formula a 0 = 5.623, M =28, there is data y t ＝{-5.5,-5.193,…-5.315},t＝1,2…N， N＝232。
According to y t T, N and M, using the formula
Can find that 1 ,a 2 ,…a 28 Same principle of
B can be determined in the same way 1 ,b 2 ,…b 28 。
According to p t The expression of (c) has only one time variable: the data corresponding to the periodic function at the time t, j is:
d, obtaining a precipitation trend prediction function l, l = p '+ x';
in this embodiment, the period term is extended as shown in fig. 7 below.
The random term is predicted as shown in fig. 8.
The extension value of the periodic function in 1965 to 2004 obtained by the SSA-XGboost model and the prediction value of the random term are superposed to obtain the delta in 1965 to 2004 18 And O, predicting the value. In order to verify the prediction effect of the SSA-XGboost model, the patent also predicts the data by using a traditional method ARIMA, an SSA-ARIMA and a machine learning method XGboost and LightGBM respectively. The prediction results obtained by the different prediction models are compared with the original data to obtain fig. 9. The figure includes original numbers from 1965 to 2004According to the prediction results of the SSA-XGboost model, the prediction results of the ARIMA, XGboost and LightGBM model and the prediction results of the SSA-ARIMA model. Because the requirement of the ARIMA model on the smoothness of the time sequence is too high, the phenomenon that the whole prediction effect tends to the average value in the graph occurs when the smoothness performance of the time sequence is poor. Although the SSA-ARIMA model adopts a periodic term, the final prediction result is still not ideal because the ARIMA prediction effect is poorer than the XGboost prediction effect. Due to the fact that the data volume is limited, the curve obtained by using the XGboost model and the LightGBM model independently is smooth in whole, and delta cannot be represented completely 18 Local characteristics of O. Only the prediction result of the SSA-XGboost model can realize high-precision stable prediction of the oxygen isotope change trend of the stalagmite in 1965-2004.
And E, judging the slope of the rainfall trend prediction function l, if the slope is smaller than 0, increasing the rainfall of the area j to be predicted at the moment, and otherwise, decreasing the rainfall of the area j to be predicted at the moment.
In this embodiment, the precipitation trend prediction function l may be converted into an image form, the trend of the prediction function l at the time j is viewed from the image, and if the trend is a descending trend, the precipitation amount is increased; otherwise, the precipitation amount is reduced.
In this example, the accuracy of the prediction results of the different models was also evaluated using Mean Absolute Error (MAE), root Mean Square Error (RMSE) and correlation coefficient (R2) analysis (table 1). From table 1 it can be found that: the MAE of the SSA-XGboost model is minimal (0.1334). In terms of RMSE, SSA-XGboost, ARIMA, SSA-ARIMA and LightGBM are 0.1687, 0.2520, 0.2632, 0.2224 and 0.3641, respectively, with the RMSE of the SSA-XGboost model still being minimal. In addition, R2 of the SSA-XGboost model is closest to 1, and R2 of other models is negative or smaller. Since the smaller the MAE and the RMSE are, the closer the R2 is to 1, the better the prediction effect is, the comparative data in the table 1 show that the prediction effect of the SSA-XGboost is the best, and the prediction effects of the other four models are poor.
TABLE 1 evaluation index of each model
EXAMPLE III
The embodiment provides a sequence prediction model construction device based on XGboost, and the device comprises a data acquisition module, a data quality improvement module, a data reconstruction module, a sequence acquisition module, a data set acquisition module and a model training module;
the data acquisition module is used for acquiring N oxygen isotope values of the stalagmite shoots and acquiring a stalagmite shoot oxygen isotope sequence y' = { y = 1 '，y 2 '，…，y t '，…，y N '}，y t ' represents the t-th isotope value of the oxygen isotope sequence of the stalagmite bamboo shoots, t =1,2, …, N and N are positive integers;
the data quality promotion module is used for promoting the data quality of the stalagmite oxygen isotope sequence y' to obtain the quality-promoted stalagmite oxygen isotope sequence y * ；
The data reconstruction module is used for reconstructing the promoted oxygen isotope sequence y of the stalagmite bamboo shoots * Performing data reconstruction to obtain a reconstructed stalagmite oxygen isotope sequence y, wherein the reconstructed stalagmite oxygen isotope sequence y comprises a random item sequence and a periodic item sequence;
the sequence obtaining module is used for obtaining the t term x of the random term sequence by adopting a formula II t ：
Wherein y is t Denotes the t value, p, of the obtained reconstructed oxygen isotope sequence of stalagmite t Value representing the t-th periodic item sequence, a 0 Denotes the initial value of the period term, a k Coefficient representing the cosine function corresponding to the kth harmonic, b k Representing coefficients of a sine function corresponding to the kth harmonic, wherein k =1,2, …, M represents the number of harmonics, and M is an integer greater than 0;
let t = t +1 until t = N, obtaining a random term sequence x and a periodic term sequence p;
the data set obtaining module is used for selecting the first q columns of data of the random item sequence x as sample data, wherein q =1,2, … and N-1, obtaining N-1 sample data and obtaining a sample set;
taking the q +1 th column data of the random item sequence x corresponding to each sample data as tag data, obtaining N-1 tag data, and obtaining a tag set;
and the model training module is used for training the XGboost model by taking the sample set as input and the label set as reference output to obtain a random term sequence prediction model.
Optionally, the data quality improving module specifically includes a repairing sub-module and a trending removing sub-module;
the repairing submodule is used for repairing the stalagmite oxygen isotope sequence y' by adopting a nearest interpolation method to obtain a repaired stalagmite oxygen isotope sequence;
the trend removing processing submodule is used for performing trend removing processing on the repaired stalagmite oxygen isotope sequence by adopting a least square method to obtain a quality-improved stalagmite oxygen isotope sequence y * 。
Optionally, the data reconstruction module specifically includes a trajectory matrix conversion sub-module, a singular value decomposition sub-module, and a matrix reconstruction sub-module;
the track matrix conversion submodule is used for converting the mass-improved stalagmite oxygen isotope sequence y * After conversion to the trajectory matrix X, a feature matrix X' is obtained using formula I:
X'＝X T * X formula I
The singular value decomposition submodule is used for carrying out singular value decomposition on the characteristic matrix X' to obtain m characteristic values, wherein m is a positive integer;
and the matrix reconstruction submodule is used for arranging the M characteristic values in a descending order and then selecting the first M characteristic values to carry out matrix reconstruction to obtain a reconstructed stalagmite oxygen isotope sequence y.
Optionally, M =28.
Example four
A rainfall trend prediction device is used for predicting the rainfall trend of a time j of an area to be predicted, wherein j is a positive integer, and the rainfall trend prediction device comprises a sequence acquisition module, a random item prediction sequence acquisition module, a periodic item prediction sequence acquisition module, a prediction function acquisition module and a prediction module;
the sequence acquisition module is used for acquiring the oxygen isotope value of the stalagmite bamboo shoot from the moment 1 to the moment j-1 of the area to be predicted and acquiring an oxygen isotope sequence of the stalagmite bamboo shoot;
the random term prediction sequence obtaining module is used for processing the oxygen isotope sequence of the stalagmite bamboo shoot obtained in the step A by adopting the method in the XGboost-based sequence prediction model construction device of any one of claims 6 to 9 to obtain a random term prediction sequence x' and a periodic term sequence p at the moment j;
the periodic item prediction sequence obtaining module is used for inputting a moment j into the periodic item sequence p to obtain a periodic item prediction sequence p' of the moment j;
the prediction function obtaining module is used for obtaining a precipitation trend prediction function l, l = p '+ x';
the prediction module is used for judging the slope of the precipitation trend prediction function l, if the slope is smaller than 0, the precipitation quantity of the area j to be predicted rises at the moment, otherwise, the precipitation quantity of the area j to be predicted falls at the moment.
Through the above description of the embodiments, those skilled in the art will clearly understand that the present invention may be implemented by software plus necessary general hardware, and certainly may also be implemented by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present invention or portions thereof contributing to the prior art may be embodied in the form of a software product, which is stored in a readable storage medium, such as a floppy disk, a hard disk, or an optical disk of a computer, and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute the method according to the embodiments of the present invention.
Claims (10)
1. A sequence prediction model construction method based on XGboost is used for obtaining a prediction model of a random item sequence in a stalagmite oxygen isotope sequence, and is characterized by comprising the following steps:
step 1, obtaining N oxygen isotope values of the stalagmite shoots, and obtaining an oxygen isotope sequence y' = { y } of the stalagmite shoots 1 '，y 2 '，…，y t '，…，y N '}，y t ' represents the t-th isotope value of the oxygen isotope sequence of the stalagmite bamboo shoots, t =1,2, …, N and N are positive integers;
step 2, performing data quality improvement on the oxygen isotope sequence y' of the stalagmite bamboo shoot to obtain the oxygen isotope sequence y of the stalagmite bamboo shoot after the quality improvement * ；
Step 3, the promoted oxygen isotope sequence y of the stalagmite bamboo shoots * Carrying out data reconstruction to obtain a reconstructed stalagmite oxygen isotope sequence y, wherein the reconstructed stalagmite oxygen isotope sequence y comprises a random item sequence and a periodic item sequence;
step 4, obtaining the t item x of the random item sequence by adopting the formula II t ：
Wherein y is t Represents the t value, p of the reconstructed oxygen isotope sequence of the stalagmite shoots obtained in the step 3 t Value representing the t-th periodic item sequence, a 0 Denotes the initial value of the period term, a k Coefficient representing the cosine function corresponding to the kth harmonic, b k Representing coefficients of a sine function corresponding to the kth harmonic, wherein k =1,2, …, M represents the number of harmonics, and M is an integer greater than 0;
step 5, letting t = t +1, returning to the step 4 until t = N, and obtaining a random item sequence x and a periodic item sequence p;
step 6, repeating the step, selecting front q columns of data of the random item sequence x as sample data, wherein q =1,2, … and N-1, obtaining N-1 sample data, and obtaining a sample set;
taking the q +1 th column data of the random item sequence x corresponding to each sample data as tag data, obtaining N-1 tag data, and obtaining a tag set;
and 7, taking the sample set as input, taking the label set as reference output, training an XGboost model, and obtaining a random item sequence prediction model.
2. The XGboost-based sequence prediction model construction method according to claim 1, wherein the step 2 specifically comprises:
2.1, repairing the oxygen isotope sequence y' of the stalagmite bamboo shoot by adopting a nearest neighbor interpolation method to obtain a repaired oxygen isotope sequence of the stalagmite bamboo shoot;
2.2, performing trend removing treatment on the repaired stalagmite oxygen isotope sequence by adopting a least square method to obtain a quality-improved stalagmite oxygen isotope sequence y * 。
3. The XGboost-based sequence prediction model construction method according to claim 1, wherein the step 3 specifically comprises:
step 3.1, improving the quality of the oxygen isotope sequence y of the stalagmite shoots * After conversion to the trajectory matrix X, a feature matrix X' is obtained using formula I:
X'＝X T * X formula I
Step 3.2, carrying out singular value decomposition on the characteristic matrix X' to obtain m characteristic values, wherein m is a positive integer;
and 3.3, after the M characteristic values are arranged from large to small, selecting the first M characteristic values to perform matrix reconstruction, and obtaining a reconstructed stalagmite oxygen isotope sequence y.
4. The method for constructing an XGBoost-based sequence prediction model of claim 1, wherein M =28.
5. A precipitation tendency prediction method is used for predicting precipitation tendency of a to-be-predicted area moment j, wherein j is a positive integer, and the method is implemented according to the following steps:
step A, obtaining a stalagmite oxygen isotope value from time 1 to time j-1 of an area to be predicted to obtain a stalagmite oxygen isotope sequence;
acquiring a stalagmite oxygen isotope value of a to-be-predicted region moment j;
step B, processing the oxygen isotope sequence of the stalagmite bamboo shoot obtained in the step A by adopting the method in the steps 2 to 5 in the XGboost-based sequence prediction model construction method of any one of claims 1 to 4 to obtain a random term sequence prediction model and a periodic term sequence p;
inputting the oxygen isotope value of the stalagmite bamboo shoot at the time j of the area to be predicted into a random term sequence prediction model to obtain a random term sequence x' at the time j;
step C, inputting the stalagmite oxygen isotope at the moment j into the periodic item sequence p to obtain a periodic item sequence p' at the moment j;
d, obtaining a precipitation trend prediction function l, l = p '+ x';
and E, judging the slope of the rainfall trend prediction function l, if the slope is smaller than 0, increasing the rainfall of the area j to be predicted at the moment, and otherwise, decreasing the rainfall of the area j to be predicted at the moment.
6. A sequence prediction model construction device based on XGboost is characterized by comprising a data acquisition module, a data quality improvement module, a data reconstruction module, a sequence acquisition module, a data set acquisition module and a model training module;
the data acquisition module is used for acquiring N oxygen isotope values of the stalagmite shoots and acquiring a stalagmite shoot oxygen isotope sequence y' = { y = 1 '，y 2 '，…，y t '，…，y N '}，y t ' represents the t-th isotope value of the stalagmite oxygen isotope sequence, t =1,2, …, N is a positive integer;
the data quality promotion module is used for promoting the data quality of the stalagmite oxygen isotope sequence y' to obtain the quality-promoted stalagmite oxygen isotope sequence y * ；
The data reconstruction moduleFor the raised oxygen isotope sequence y of the stalagmite * Performing data reconstruction to obtain a reconstructed stalagmite oxygen isotope sequence y, wherein the reconstructed stalagmite oxygen isotope sequence y comprises a random item sequence and a periodic item sequence;
the sequence obtaining module is used for obtaining the t term x of the random term sequence by adopting a formula II t ：
Wherein y is t Denotes the t value, p of the obtained reconstructed oxygen isotope sequence of the stalagmite t Value representing the t-th periodic item sequence, a 0 Denotes the initial value of the period term, a k Coefficient representing the cosine function corresponding to the kth harmonic, b k Representing coefficients of a sine function corresponding to the kth harmonic, wherein k =1,2, …, M represents the number of harmonics, and M is an integer greater than 0;
let t = t +1 until t = N, obtaining a random term sequence x and a periodic term sequence p;
the data set obtaining module is used for selecting the first q columns of data of the random item sequence x as sample data, wherein q =1,2, … and N-1, obtaining N-1 sample data and obtaining a sample set;
taking the q +1 th column data of the random item sequence x corresponding to each sample data as tag data to obtain N-1 tag data and obtain a tag set;
and the model training module is used for training the XGboost model by taking the sample set as input and the label set as reference output to obtain a random term sequence prediction model.
7. The XGboost-based sequence prediction model construction device according to claim 6, wherein the data quality improvement module specifically comprises a repair sub-module and a de-trend processing sub-module;
the repairing submodule is used for repairing the stalagmite oxygen isotope sequence y' by adopting a nearest neighbor interpolation method to obtain a repaired stalagmite oxygen isotope sequence;
the trend removing processing submodule is used for performing trend removing processing on the repaired stalagmite oxygen isotope sequence by adopting a least square method to obtain a quality-improved stalagmite oxygen isotope sequence y * 。
8. The XGboost-based sequence prediction model construction device according to claim 6, wherein the data reconstruction module specifically comprises a trajectory matrix conversion submodule, a singular value decomposition submodule and a matrix reconstruction submodule;
the track matrix conversion submodule is used for converting the mass-improved stalagmite oxygen isotope sequence y * After conversion to the trajectory matrix X, a feature matrix X' is obtained using formula I:
X'＝X T * X formula I
The singular value decomposition submodule is used for carrying out singular value decomposition on the characteristic matrix X' to obtain m characteristic values, wherein m is a positive integer;
and the matrix reconstruction submodule is used for arranging the M characteristic values in a descending order and then selecting the first M characteristic values to carry out matrix reconstruction to obtain a reconstructed stalagmite oxygen isotope sequence y.
9. The XGBoost-based sequence prediction model construction apparatus of claim 6, wherein M =28.
10. A rainfall trend prediction device is used for predicting the rainfall trend of a time j of an area to be predicted, wherein j is a positive integer, and the rainfall trend prediction device is characterized by comprising a sequence acquisition module, a random item prediction sequence acquisition module, a periodic item prediction sequence acquisition module, a prediction function acquisition module and a prediction module;
the sequence acquisition module is used for acquiring a stalagmite oxygen isotope value of the area to be predicted from time 1 to time j-1 to obtain a stalagmite oxygen isotope sequence;
the sequence acquisition module is also used for acquiring a stalagmite oxygen isotope value of the area to be predicted at the moment j;
the random term prediction sequence obtaining module is used for processing the obtained stalagmite oxygen isotope sequence by adopting the method in the XGboost-based sequence prediction model construction device of any one of claims 6 to 9 to obtain a random term sequence prediction model and a periodic term sequence p;
the method is also used for inputting the oxygen isotope value of the stalagmite bamboo shoot at the time j of the area to be predicted into a random item sequence prediction model to obtain a random item sequence x' at the time j;
the period item prediction sequence obtaining module is used for inputting the stalagmite oxygen isotope at the moment j into the period item sequence p to obtain a period item sequence p' at the moment j;
the prediction function obtaining module is used for obtaining a precipitation trend prediction function l, l = p '+ x';
the prediction module is used for judging the slope of the precipitation trend prediction function l, if the slope is smaller than 0, the precipitation quantity of the area j to be predicted rises at the moment, otherwise, the precipitation quantity of the area j to be predicted falls at the moment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110046489.6A CN112699614B (en) | 2021-01-14 | 2021-01-14 | XGboost-based sequence prediction model construction and precipitation trend prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110046489.6A CN112699614B (en) | 2021-01-14 | 2021-01-14 | XGboost-based sequence prediction model construction and precipitation trend prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112699614A CN112699614A (en) | 2021-04-23 |
| CN112699614B true CN112699614B (en) | 2023-01-31 |
Family
ID=75514638
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110046489.6A Active CN112699614B (en) | 2021-01-14 | 2021-01-14 | XGboost-based sequence prediction model construction and precipitation trend prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112699614B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115860268B (en) * | 2023-02-20 | 2023-07-14 | 南京信息工程大学 | Short-time strong precipitation prediction method based on CNN-XGBOOST model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014002974A1 (en) * | 2012-06-27 | 2014-01-03 | ヤンマー株式会社 | Prediction device, prediction method, and computer programme |
| CN106093175A (en) * | 2016-06-06 | 2016-11-09 | 中国科学院地球化学研究所 | A kind of method utilizing stalagmite inverting earthquake that the age occurs |
| CN111178585A (en) * | 2019-12-05 | 2020-05-19 | 国网浙江省电力有限公司杭州供电公司 | Fault reporting amount prediction method based on multi-algorithm model fusion |
| CN111241629A (en) * | 2020-01-08 | 2020-06-05 | 沈阳航空航天大学 | Intelligent prediction method for performance change trend of airplane hydraulic pump based on data driving |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI662422B (en) * | 2018-04-23 | 2019-06-11 | 國家中山科學研究院 | Air quality prediction method based on machine learning model |
| CN109009179B (en) * | 2018-08-02 | 2020-09-18 | 浙江大学 | Same isotope labeling double-tracer PET separation method based on deep belief network |
-
2021
- 2021-01-14 CN CN202110046489.6A patent/CN112699614B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014002974A1 (en) * | 2012-06-27 | 2014-01-03 | ヤンマー株式会社 | Prediction device, prediction method, and computer programme |
| CN106093175A (en) * | 2016-06-06 | 2016-11-09 | 中国科学院地球化学研究所 | A kind of method utilizing stalagmite inverting earthquake that the age occurs |
| CN111178585A (en) * | 2019-12-05 | 2020-05-19 | 国网浙江省电力有限公司杭州供电公司 | Fault reporting amount prediction method based on multi-algorithm model fusion |
| CN111241629A (en) * | 2020-01-08 | 2020-06-05 | 沈阳航空航天大学 | Intelligent prediction method for performance change trend of airplane hydraulic pump based on data driving |
Non-Patent Citations (3)
| Title |
|---|
| 一种融合历史均值与提升树的客流量预测模型;白智远等;《计算机技术与发展》;20181220(第04期);218-221 * |
| 时间序列分析及Monte Carlo方法在降水预报中的应用;杨磊磊等;《环境科学与技术》;20110515(第05期);114-118 * |
| 末次间冰期东亚季风气候不稳定的神农架洞穴石笋记录;姜修洋等;《沉积学报》;20080215(第01期);143-147 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112699614A (en) | 2021-04-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107292446B (en) | Hybrid wind speed prediction method based on component relevance wavelet decomposition | |
| CN111079998B (en) | Flow prediction method based on long and short time sequence correlation attention mechanism model | |
| US20100070441A1 (en) | Method, apparatus, and program for generating prediction model based on multiple regression analysis | |
| CN113411216B (en) | Network flow prediction method based on discrete wavelet transform and FA-ELM | |
| CN111160651B (en) | STL-LSTM-based subway passenger flow prediction method | |
| CN110210660B (en) | Ultra-short-term wind speed prediction method | |
| CN112699614B (en) | XGboost-based sequence prediction model construction and precipitation trend prediction method and device | |
| CN115800245A (en) | Short-term load prediction method based on SARIMA-random forest combined model | |
| CN113935535A (en) | Principal component analysis method for medium-and-long-term prediction model | |
| CN115062272A (en) | Water quality monitoring data abnormity identification and early warning method | |
| CN110909928A (en) | Energy load short-term prediction method and device, computer equipment and storage medium | |
| Arous et al. | Recovdb: Accurate and efficient missing blocks recovery for large time series | |
| CN113409072A (en) | Empirical mode decomposition and distributed GRU neural network and price prediction method | |
| CN116362823A (en) | Recommendation model training method, recommendation method and recommendation device for behavior sparse scene | |
| CN116342172A (en) | Oil price prediction method, device and equipment based on combination of linear regression and decision tree | |
| CN116720080A (en) | Homologous meteorological element fusion inspection method | |
| CN114037143A (en) | Short-term wind power combination prediction method | |
| CN118137479A (en) | Short-term wind power prediction method based on combined prediction model | |
| CN108509692A (en) | A kind of high sulfur content natural gas desulfurization process modeling method based on MiUKFNN algorithms | |
| Zhang | Forecasting with Bayesian Grouped random effects in panel data | |
| CN116739168A (en) | Runoff prediction method based on gray theory and codec | |
| CN114820074A (en) | Target user group prediction model construction method based on machine learning | |
| CN114141316A (en) | Method and system for predicting biological toxicity of organic matters based on spectrogram analysis | |
| CN118035110B (en) | Informationized model evaluation method and system based on big data | |
| CN117216081B (en) | Automatic question bank updating method and device, electronic 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |