CN117828456A - Method, system and electronic equipment for predicting interpretable runoff sequence - Google Patents
Method, system and electronic equipment for predicting interpretable runoff sequence Download PDFInfo
- Publication number
- CN117828456A CN117828456A CN202410026297.2A CN202410026297A CN117828456A CN 117828456 A CN117828456 A CN 117828456A CN 202410026297 A CN202410026297 A CN 202410026297A CN 117828456 A CN117828456 A CN 117828456A
- Authority
- CN
- China
- Prior art keywords
- training
- runoff
- sequence
- subsequence
- subsequences
- 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 35
- 238000012360 testing method Methods 0.000 claims abstract description 78
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 47
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims description 167
- 238000012795 verification Methods 0.000 claims description 50
- 238000005457 optimization Methods 0.000 claims description 13
- 238000005259 measurement Methods 0.000 claims description 12
- 238000012706 support-vector machine Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 6
- 238000010200 validation analysis Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010206 sensitivity analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method, a system and electronic equipment for predicting an interpretable runoff sequence, and relates to the technical field of runoff prediction. The method comprises the following steps: carrying out serialization decomposition on the test set by using a variation modal decomposition algorithm to obtain a plurality of test subsequences; performing modal reconstruction on the plurality of test subsequences according to the sample entropy value to obtain a plurality of test modal reconstruction subsequences so as to determine a test prediction factor; inputting the test prediction factor into a runoff sequence prediction model to obtain a runoff prediction sequence of the runoff to be measured in the next time period at the current moment. The invention completes the combination of the test subsequences through the modal reconstruction, can improve the accuracy and the efficiency of the radial flow sequence prediction, and reduces the calculated amount of the radial flow sequence prediction.
Description
Technical Field
The invention relates to the technical field of runoff prediction, in particular to a method, a system and electronic equipment for predicting an interpretable runoff sequence.
Background
The accurate and reliable medium-and-long-term runoff prediction has important significance for flood control, drought resistance and water resource planning and management of the runoff domain, but the runoff shows uncertainty in time and space scales, and has great difficulty in accurately predicting future runoffs. And the runoff sequence with high irregularity, complex nonlinearity and multi-scale variability characteristics is decomposed by adopting a signal processing technology, so that information hidden in the hydrologic sequence can be mined. Based on the characteristics, a radial flow sequence prediction model based on a machine learning algorithm is established, so that the prediction accuracy of the radial flow sequence can be improved, and compared with a physical prediction model, the radial flow sequence prediction model is easier to realize.
At present, the whole runoff sequence is firstly decomposed into a plurality of subsequences based on a decomposition model, and then the subsequences are divided into a training period and a verification period, and the decomposition-before-division method can lead researchers to use verification period data in advance before training the model, so that the actual forecast requirement is difficult to meet. In addition, the existing decomposition algorithm generally generates a large number of subsequences, and the subsequences are directly input as prediction factors, so that a large amount of calculation resources are consumed, an overfitting phenomenon is caused, and the generalization capability of the runoff prediction model is limited. Finally, the previous research mostly does not explain the cause of the model prediction result, and limits the reliability and application value of the runoff prediction model.
Disclosure of Invention
The invention aims to provide a method, a system and electronic equipment for predicting an interpretable runoff sequence, which can improve the accuracy of the runoff sequence prediction.
In order to achieve the above object, the present invention provides the following solutions:
an interpretable runoff sequence prediction method, comprising:
obtaining a plurality of runoff actual measurement values of runoff to be measured in a preset time period at the current moment as a test set;
carrying out serialization decomposition on the test set by using a variation modal decomposition algorithm to obtain a plurality of test subsequences;
performing modal reconstruction on the plurality of test subsequences according to the sample entropy value to obtain a plurality of test modal reconstruction subsequences; the sample entropy value is determined according to a training set of a runoff sequence prediction model; runoff measurement data in a historical time period of the runoff to be measured of the training set;
determining a test predictor according to the multiple test modality reconstruction subsequences and the lag subintervals; the lag subperiod is determined according to a training set of a runoff sequence prediction model;
inputting the test prediction factor into a runoff sequence prediction model to obtain a runoff prediction sequence of the runoff to be measured in the next time period at the current moment; the runoff sequence prediction model is obtained by training a support vector machine model by using a Bayesian optimization algorithm according to a historical runoff sequence of runoffs to be measured.
Optionally, before obtaining the plurality of measured runoffs of the runoffs to be measured in the preset time period at the current moment as the test set, the method further includes:
acquiring a historical runoff sequence of runoff to be measured;
splitting the historical runoff sequence into a training set and a verification set according to a splitting proportion;
parallelizing and decomposing the training set by using a variational modal decomposition algorithm to obtain a plurality of training subsequences;
respectively determining a sample entropy value of each training subsequence;
performing modal reconstruction on the training subsequences according to the sample entropy values to obtain a plurality of training modal reconstruction subsequences;
determining a partial autocorrelation coefficient of each training modality reconstruction sub-sequence;
determining a plurality of sub-periods of which the partial autocorrelation coefficients are larger than the partial autocorrelation coefficient threshold value and the training mode reconstruction sub-sequence is lagged as lagged sub-periods; the sub-periods are in one-to-one correspondence with the training modality reconstruction sub-sequences;
determining a union of training mode reconstruction subsequences corresponding to a plurality of lag subperiods as a training predictor;
and training the support vector machine model by using a Bayesian optimization algorithm with the training prediction factor as input and the historical runoff sequence as a prediction target to obtain a predicted model of the runoff sequence to be determined.
Optionally, after training the support vector machine model by using the training predictor as input and using the historical runoff sequence as a prediction target and using a bayesian optimization algorithm to obtain a predicted model of the runoff sequence, the method further comprises:
let iteration number i=1;
taking the ith element in the verification set as a last element, and taking the training set as a first element to construct an additional set of the ith iteration;
carrying out serialization decomposition on the additional set of the ith iteration by utilizing a variational modal decomposition algorithm to obtain a plurality of sub-sequences to be verified;
taking the additional set of the ith iteration as a training set, increasing the value of i by 1, and returning to the step of taking the ith element in the verification set as a last element, and taking the training set as a first element to construct the additional set of the ith iteration until the verification set is traversed, so that a plurality of to-be-determined verification subsequences are verification subsequences;
performing modal reconstruction on the verification subsequences according to the sample entropy values to obtain verification modal reconstruction subsequences;
determining a verification predictor from the plurality of verification modality reconstruction subsequences and the lag subperiod;
inputting the verification prediction factor into a predicted runoff sequence prediction model to obtain a runoff sequence history prediction result;
updating the predicted runoff sequence prediction model according to the historical runoff sequence prediction result and the historical runoff sequence until the error of the historical runoff sequence prediction result and the historical runoff sequence is within an error threshold value, and obtaining the runoff sequence prediction model.
Optionally, parallelizing the training set by using a variational modal decomposition algorithm to obtain a plurality of training subsequences, including:
determining any training subsequence as a current training subsequence;
determining any adjacent training subsequence of the current subsequence as the current adjacent training subsequence;
judging whether the absolute value of the sample entropy difference value of the current training subsequence and the current adjacent training subsequence is smaller than a difference value threshold value or not, and obtaining a first judging result;
if the first judgment result is yes, determining that the union of the current training subsequence and the current adjacent training subsequence is a modal reconstruction subsequence;
deleting the current training subsequence and the current adjacent training subsequence, taking the modal reconstruction subsequence as the current training subsequence, and returning to the step of determining any adjacent training subsequence of the current subsequence as the current adjacent training subsequence;
if the first judgment result is negative, judging whether to traverse all adjacent training subsequences of the current subsequence, and obtaining a second judgment result;
if the second judgment result is negative, updating the current adjacent training subsequence, and returning to the step of judging whether the absolute value of the sample entropy value difference value of the current training subsequence and the current adjacent training subsequence is smaller than a difference value threshold value, so as to obtain a first judgment result;
if the second judgment result is yes, judging whether all training subsequences are traversed or not to obtain a third judgment result;
if the third judgment result is negative, updating the current training subsequence, and returning to the step of determining any adjacent training subsequence of the current subsequence as the current adjacent training subsequence;
if the third judgment result is yes, completing the modal reconstruction, and determining a plurality of training subsequences as training modal reconstruction subsequences.
An interpretable runoff sequence prediction system, comprising:
the test set acquisition module is used for acquiring a plurality of runoff actual measurement values of the runoff to be tested in a preset time period at the current moment as a test set;
the test subsequence determining module is used for carrying out serialization decomposition on the test set by utilizing a variation modal decomposition algorithm to obtain a plurality of test subsequences;
the test mode reconstruction sub-sequence determining module is used for carrying out mode reconstruction on the plurality of test sub-sequences according to the sample entropy value to obtain a plurality of test mode reconstruction sub-sequences; the sample entropy value is determined according to a training set of a runoff sequence prediction model; runoff measurement data in a historical time period of the runoff to be measured of the training set;
the test prediction factor determining module is used for determining a test prediction factor according to the multiple test mode reconstruction subsequences and the lag subintervals; the lag subperiod is determined according to a training set of a runoff sequence prediction model;
the runoff prediction sequence determining module is used for inputting the test prediction factor into a runoff sequence prediction model to obtain a runoff prediction sequence of the runoff to be measured in the next time period at the current moment; the runoff sequence prediction model is obtained by training a support vector machine model by using a Bayesian optimization algorithm according to a historical runoff sequence of runoffs to be measured.
An electronic device comprising a memory for storing a computer program and a processor that runs the computer program to cause the electronic device to perform the one interpretable runoff sequence prediction method.
Optionally, the memory is a readable storage medium.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
according to the method, the system and the electronic equipment for predicting the interpretable runoff sequence, provided by the invention, a signal processing technology and a machine learning technology are introduced, the runoff sequence with high irregularity, complex nonlinearity and multi-scale variability characteristics is decomposed, and information hidden in the runoff sequence is mined; the training set and the verification set are respectively subjected to parallelization and serialization decomposition, future information is prevented from being used in the runoff prediction process, modeling calculation cost can be reduced, generalization capability of a model is improved through modal reconstruction based on a sample entropy value, a runoff sequence prediction model based on a machine learning algorithm is established based on the characteristics, and therefore runoff sequence prediction precision and generalization capability are improved; finally, the SHAP interpretable method is adopted to analyze the cause of the model prediction result, so that the reliability and the application value of the runoff prediction model are improved, and the accuracy of the runoff sequence prediction can be improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart illustrating a method for predicting runoff sequences according to embodiment 1 of the present invention;
FIG. 2 is a schematic diagram of a method for predicting runoff sequences according to embodiment 1 of the present invention;
FIG. 3 is a diagram showing the radial flow prediction result in embodiment 1 of the present invention;
FIG. 4 is a schematic diagram showing the result of the analysis in example 1 of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide a method, a system and electronic equipment for predicting an interpretable runoff sequence, which can improve the accuracy of the runoff sequence prediction.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Example 1
As shown in fig. 1, the present embodiment provides an interpretable runoff sequence prediction method, including:
step 101: obtaining a plurality of runoff actual measurement values of runoff to be measured in a preset time period at the current moment as a test set.
Step 102: and carrying out serialization decomposition on the test set by using a variation modal decomposition algorithm to obtain a plurality of test subsequences.
Step 103: performing modal reconstruction on the plurality of test subsequences according to the sample entropy value to obtain a plurality of test modal reconstruction subsequences; the sample entropy value is determined according to a training set of a runoff sequence prediction model; and the runoff measurement data in the historical time period of the runoff to be measured of the training set.
Step 104: determining a test predictor according to the multiple test modality reconstruction subsequences and the lag subintervals; the lag sub-period is determined from a training set of radial flow sequence prediction models.
Step 105: inputting the test prediction factor into a runoff sequence prediction model to obtain a runoff prediction sequence of the runoff to be measured in the next time period at the current moment; the runoff sequence prediction model is obtained by training a support vector machine model by using a Bayesian optimization algorithm according to the historical runoff sequence of runoffs to be measured.
Prior to step 101, further comprising:
step 106: and acquiring a historical runoff sequence of the runoff to be measured.
Step 107: and splitting the historical runoff sequence into a training set and a verification set according to the splitting proportion.
Step 108: and carrying out parallelization decomposition on the training set by using a variational modal decomposition algorithm to obtain a plurality of training subsequences.
Step 109: the sample entropy value of each training sub-sequence is determined separately.
Step 1010: and carrying out modal reconstruction on the training subsequences according to the sample entropy values to obtain a plurality of training modal reconstruction subsequences.
Step 1011: partial autocorrelation coefficients of each training modality reconstruction sub-sequence are determined.
Step 1012: and determining a plurality of sub-periods of the training mode reconstruction sub-sequence lag with the partial autocorrelation coefficient greater than the partial autocorrelation coefficient threshold as lag sub-periods. The subintervals correspond to the training modality reconstruction subsequences one-to-one.
Step 1013: and determining a union of training mode reconstruction subsequences corresponding to the hysteresis sub-periods as a training predictor.
Step 1014: and training the support vector machine model by using a Bayesian optimization algorithm with the training prediction factor as input and the historical runoff sequence as a prediction target to obtain a predicted model of the runoff sequence to be determined.
Step 1015: let iteration number i=1.
Step 1016: and constructing an additional set of the ith iteration by taking the ith element in the verification set as a last element and the training set as a first element.
Step 1017: and carrying out parallelization decomposition on the additional set of the ith iteration by using a variational modal decomposition algorithm to obtain a plurality of to-be-determined verification subsequences.
Step 1018: taking the additional set of the ith iteration as a training set, increasing the value of i by 1, and returning to the step 1016 until the verification set is traversed, so as to obtain a plurality of to-be-determined verification subsequences as verification subsequences.
Step 1017: the sample entropy value of each verification sub-sequence is determined separately.
Step 1018: and carrying out modal reconstruction on the verification sub-sequences according to the sample entropy values to obtain verification modal reconstruction sub-sequences.
Step 1019: partial autocorrelation coefficients of each verification modality reconstruction sub-sequence are determined.
Step 1020: and determining a plurality of sub-periods of the verification mode reconstruction sub-sequence lag with the partial autocorrelation coefficient greater than the partial autocorrelation coefficient threshold as lag sub-periods. The sub-periods correspond to the verification modality reconstruction sub-sequences one to one.
Step 1021: and determining the union of the verification modality reconstruction subsequences corresponding to the plurality of lag subperiods as a verification predictor.
Step 1022: and inputting the verification prediction factor into a predicted runoff sequence prediction model to obtain a runoff sequence history prediction result.
Step 1023: updating the predicted runoff sequence prediction model according to the historical runoff sequence prediction result and the historical runoff sequence until the error of the historical runoff sequence prediction result and the historical runoff sequence is within an error threshold value, and obtaining the runoff sequence prediction model.
Step 108, including:
step 108-1: and determining any training subsequence as the current training subsequence.
Step 108-2: and determining any adjacent training subsequence of the current subsequence as the current adjacent training subsequence.
Step 108-3: and judging whether the absolute value of the sample entropy value difference value of the current training subsequence and the current adjacent training subsequence is smaller than a difference value threshold value, and obtaining a first judging result.
Step 108-4: if the first judgment result is yes, determining that the union of the current training subsequence and the current adjacent training subsequence is a modal reconstruction subsequence.
Step 108-5: deleting the current training subsequence and the current adjacent training subsequence, and returning the mode reconstruction subsequence to the step 108-2 as the current training subsequence.
Step 108-6: if the first judgment result is negative, judging whether all adjacent training subsequences of the current subsequence are traversed or not, and obtaining a second judgment result.
Step 108-7: if the second determination result is no, the current adjacent training sub-sequence is updated, and step 108-3 is returned.
Step 108-8: if the second judgment result is yes, judging whether all training subsequences are traversed, and obtaining a third judgment result.
Step 108-9: if the third judgment result is negative, the current training sub-sequence is updated, and the step of 'determining any adjacent training sub-sequence of the current sub-sequence as the current adjacent training sub-sequence' is returned.
Step 108-10: if the third judgment result is yes, completing the modal reconstruction, and determining a plurality of training subsequences as training modal reconstruction subsequences.
As shown in fig. 2, the method for predicting an interpretable runoff sequence provided in the present embodiment mainly includes three stages: the decomposition phase, the reconstruction phase and the prediction phase, which are described in detail below in connection with the method of the invention.
And step 1, splitting data. Splitting the original hydrologic sequence into a training set and a verification set according to a certain proportion. The splitting ratio can be determined according to the length of the hydrologic sequence, the ratio of the general training set is 70%, the ratio of the verification set is 30%, the ratio of the training set can be increased when the length of the sequence is longer, the ratio of the test set can be reduced, and the ratio of the training set can be reduced when the length of the sequence is smaller, and the ratio of the verification set is increased, so that the model obtains better generalization performance.
And 2, parallelizing and decomposing the training set. And carrying out parallelization decomposition on the training set by adopting a variational modal decomposition algorithm (VMD), wherein the bandwidth of a model parameter is limited to 2000, the noise tolerance is set to 0, the convergence error is set to 1e-9, and the optimal decomposition level K of the VMD algorithm is determined by observing whether the center frequency of the last component in the subsequence obtained by decomposition is aliased. For example, the decomposition level is tested stepwise upward from k=2, and when the number of components of the test decomposition is k=m, the center frequency of the mth component is found to be aliased for the first time, and the decomposition level is selected to be m-1. In this way, the generation of redundant components can be effectively avoided.
And 3, carrying out serialization decomposition on the verification set. And moving the first sample in the verification set to the tail part of the training set to form an additional set, decomposing the additional set by adopting the same parameter setting as the training set, and serializing and circulating until all the verification set samples are added to the tail part of the training set one by one to be decomposed, and selecting the last additional decomposition result as the verification sample. The decomposition process accords with the habit of observing runoffs in reality time by time. For example, if the training set has 70 samples and the validation set has 30 samples, the serialization addition is to take the 1 st sample in the validation set out of the validation set, put it behind the 70 samples in the validation set, so that there are 71 additional samples of the 71 samples, decompose the 71 samples, take one validation set sample again next, put it behind the 71 samples, form 72 additional sets of samples, and decompose with the same parameters, and repeat this step until all 30 validation set samples are moved to the tail of the training set for decomposition.
Step 4, modal reconstruction based on sample entropy: calculating sample entropy values of all sub-sequences of the training set, merging sub-sequences with similar sample entropy as new sub-sequences, performing the same operation on the last sample in the additional samples to complete modal reconstruction, and giving a sample entropy calculation process:
in the formulae (1) - (2), n is the length of the radial flow sequence, m is the reconstruction dimension, r is the similarity tolerance, B m (r) is the probability that two sequences match m points with a similar tolerance r.
Step 5, sample generation: calculating Partial Autocorrelation Coefficients (PACFs) of the subsequences obtained after reconstruction, selecting a lag time period of PACFs larger than 0.5 in the subsequences as a prediction factor, and taking an original runoff sequence as a prediction daily mark to form a training sample; for the subsequence obtained by additional decomposition, the hysteresis time period which is the same as that of the subsequence corresponding to the training set is selected as a prediction factor, the original hydrologic sequence is used as a prediction target to form an additional sample, the last sample in the additional sample is taken as a verification sample, the verification sample is split into (50%) and a test sample (50%), wherein the development sample is used for selecting an optimal model from a plurality of models, and the test sample is used for testing the optimal model. For example, when 5 subsequences are generated by modal reconstruction, respectively calculating PACFs of the subsequences, finding out lag time periods, in which the PACFs are greater than 0.5, in each subsequence as a prediction factor, and assuming that the partial autocorrelation coefficient of the subsequence 1 in lag time periods of 2 time periods is greater than 0.5, selecting the 2 lag time periods as the prediction factor, and similarly, assuming that the optimal lag time periods of the subsequences 2 to 5 are 3,2,4,3 respectively, selecting the original runoff sequence of the current time period as a prediction target, and generating a learning sample, wherein the total of 2+3+2+4+3=14 lag time periods is 2+3+4.
And 6, normalizing the sequence. The magnitude or the value range of the predictive factor sequence is far different, so that the objective function optimization algorithm in the machine learning model cannot work normally. Sample normalization enables the influence of a prediction factor sequence on an objective function to be consistent, can accelerate optimal convergence, and can obtain higher prediction precision. Step 4 is obtained by adopting a formula (3)Training samples, development samples and test samples normalized to [ -1,1]. D in formula (1) norm And D represents normalized data and raw data, respectively, D max And D min Representing the maximum and minimum values in the original data, respectively. The invention normalizes the development sample and the test sample by adopting the maximum value and the minimum value of the training sample so that the whole sequence obeys the same distribution.
And 7, training a Support Vector Regression (SVR) model. Inputting the training sample and the development sample into the SVR model, optimizing the super parameters such as weight penalty parameter, radial basis function width, error tolerance and the like in the SVR model by adopting a Bayesian optimization algorithm, determining the super parameters of the optimal SVR model, and adopting the optimal super parameter model as the optimal SVR model.
And 8, verifying an optimal model. And inputting the predictive factors in the test samples into an optimal SVR model, predicting the original runoff sequence, and evaluating the original runoff sequence by adopting four indexes, namely Root Mean Square Error (RMSE), correlation coefficient (R), nash coefficient (NSE) and peak threshold percentage statistics (PPTS).
These four index calculations are explained below:
in the formulae (4) to (7), n is the length of the runoff sequence,is the original runoff sequence, < > 10 >>Is the average of the original runoff sequence,/->For the prediction result of the model, +.>Is the average value of the model prediction results. Before the formula (5) is calculated, firstly, the original runoff sequences are arranged in a descending order, the corresponding prediction results are arranged according to indexes corresponding to the original runoff sequences, the threshold value gamma represents the maximum runoff sequence of which the first gamma% is selected from the descending order, and G represents the length of the selected runoff sequence.
And 9, when the prediction model of the three-stage decomposition integration is constructed, and the runoff sequence of the next period is required to be predicted, adding the current measured value to the historical data set, replacing the original runoff sequence in the step 1, and repeating part of contents in the steps 1, 3, 4, 5 and 6, namely, carrying out the steps of data splitting, verification set serialization decomposition, modal reconstruction, additional sample generation and splitting and test sample input prediction on the data set formed after the addition, so as to obtain the runoff sequence of the next period, wherein the optimal SVR model is established by utilizing the training set in the steps, only carrying out the additional set decomposition in the subsequent prediction process, and not carrying out the training set decomposition, and the runoff prediction result is shown in figure 3.
Step 10, interpretation analysis of the prediction results: in order to enhance the interpretability of the model, a SHAP visual model interpretation tool is adopted to explore the contribution of each input feature to the predicted result, SHAP is a global sensitivity analysis method based on game theory, and can be used for analyzing the contribution of each input feature of the model to the predicted result so as to improve the reliability and application value of the model, calculating SHAP values of the input features corresponding to different subsequences, analyzing the contribution of each reconstructed subsequence to the predicted result so as to improve the reliability and application value of the model, and the calculation process of the SHAP values is given below:
f x (S)＝E[f(x)|x s ] (9)。
s in equations (8) - (9) is a subset of the input features used in the model, x is a vector of feature values for the instance to be interpreted, P is the feature dimension,for the weight of subset S, +.>Represents x j For marginal contribution of predicted value, E [ f (x) |x s ]Is the conditional expectation of the input feature subset S, I j For the global importance of feature j, n is the number of samples of feature j, and the result of the interpretation analysis is shown in FIG. 4.
In the above steps, steps 1 to 3 belong to the decomposition stage, steps 4 to 6 belong to the reconstruction stage, and steps 7 to 10 belong to the prediction stage.
Example 2
In order to perform the method corresponding to the above embodiment 1 to achieve the corresponding functions and technical effects, an interpretable runoff sequence prediction system is provided below, including:
the test set acquisition module is used for acquiring a plurality of runoff actual measurement values of the runoff to be tested in a preset time period at the current moment as a test set.
And the test subsequence determining module is used for carrying out serialization decomposition on the test set by utilizing a variation modal decomposition algorithm to obtain a plurality of test subsequences.
The test mode reconstruction sub-sequence determining module is used for carrying out mode reconstruction on the plurality of test sub-sequences according to the sample entropy value to obtain a plurality of test mode reconstruction sub-sequences; the sample entropy value is determined according to a training set of a runoff sequence prediction model; and the runoff measurement data in the historical time period of the runoff to be measured of the training set.
The test prediction factor determining module is used for determining a test prediction factor according to the multiple test mode reconstruction subsequences and the lag subintervals; the lag sub-period is determined from a training set of radial flow sequence prediction models.
The runoff prediction sequence determining module is used for inputting the test prediction factor into a runoff sequence prediction model to obtain a runoff prediction sequence of the runoff to be measured in the next time period at the current moment; the runoff sequence prediction model is obtained by training a support vector machine model by using a Bayesian optimization algorithm according to a historical runoff sequence of runoffs to be measured.
Example 3
The present embodiment provides an electronic device, including a memory for storing a computer program and a processor that runs the computer program to cause the electronic device to execute an interpretable runoff sequence prediction method as described in embodiment 1. Wherein the memory is a readable storage medium.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (7)
1. A method of interpretable runoff sequence prediction, comprising:
obtaining a plurality of runoff actual measurement values of runoff to be measured in a preset time period at the current moment as a test set;
carrying out serialization decomposition on the test set by using a variation modal decomposition algorithm to obtain a plurality of test subsequences;
performing modal reconstruction on the plurality of test subsequences according to the sample entropy value to obtain a plurality of test modal reconstruction subsequences; the sample entropy value is determined according to a training set of a runoff sequence prediction model; runoff measurement data in a historical time period of the runoff to be measured of the training set;
determining a test predictor according to the multiple test modality reconstruction subsequences and the lag subintervals; the lag subperiod is determined according to a training set of a runoff sequence prediction model;
inputting the test prediction factor into a runoff sequence prediction model to obtain a runoff prediction sequence of the runoff to be measured in the next time period at the current moment; the runoff sequence prediction model is obtained by training a support vector machine model by using a Bayesian optimization algorithm according to a historical runoff sequence of runoffs to be measured.
2. The method for predicting an interpretable runoff sequence of claim 1, further comprising, prior to obtaining a plurality of actual runoff values of the runoff to be measured within a preset time period at the current time as a test set:
acquiring a historical runoff sequence of runoff to be measured;
splitting the historical runoff sequence into a training set and a verification set according to a splitting proportion;
parallelizing and decomposing the training set by using a variational modal decomposition algorithm to obtain a plurality of training subsequences;
respectively determining a sample entropy value of each training subsequence;
performing modal reconstruction on the training subsequences according to the sample entropy values to obtain a plurality of training modal reconstruction subsequences;
determining a partial autocorrelation coefficient of each training modality reconstruction sub-sequence;
determining a plurality of sub-periods of which the partial autocorrelation coefficients are larger than the partial autocorrelation coefficient threshold value and the training mode reconstruction sub-sequence is lagged as lagged sub-periods; the sub-periods are in one-to-one correspondence with the training modality reconstruction sub-sequences;
determining a union of training mode reconstruction subsequences corresponding to a plurality of lag subperiods as a training predictor;
and training the support vector machine model by using a Bayesian optimization algorithm with the training prediction factor as input and the historical runoff sequence as a prediction target to obtain a predicted model of the runoff sequence to be determined.
3. The method of claim 2, wherein training the support vector machine model with the training predictor as input and the historical runoff sequence as the prediction target by using a bayesian optimization algorithm to obtain the predicted model of the runoff sequence, further comprises:
let iteration number i=1;
taking the ith element in the verification set as a last element, and taking the training set as a first element to construct an additional set of the ith iteration;
carrying out serialization decomposition on the additional set of the ith iteration by utilizing a variational modal decomposition algorithm to obtain a plurality of sub-sequences to be verified;
taking the additional set of the ith iteration as a training set, increasing the value of i by 1, and returning to the step of taking the ith element in the verification set as a last element, and taking the training set as a first element to construct the additional set of the ith iteration until the verification set is traversed, so that a plurality of to-be-determined verification subsequences are verification subsequences;
performing modal reconstruction on the verification subsequences according to the sample entropy values to obtain verification modal reconstruction subsequences;
determining a verification predictor from the plurality of verification modality reconstruction subsequences and the lag subperiod;
inputting the verification prediction factor into a predicted runoff sequence prediction model to obtain a runoff sequence history prediction result;
updating the predicted runoff sequence prediction model according to the historical runoff sequence prediction result and the historical runoff sequence until the error of the historical runoff sequence prediction result and the historical runoff sequence is within an error threshold value, and obtaining the runoff sequence prediction model.
4. The method of claim 2, wherein parallelizing the training set with a variational modal decomposition algorithm to obtain a plurality of training subsequences, comprises:
determining any training subsequence as a current training subsequence;
determining any adjacent training subsequence of the current subsequence as the current adjacent training subsequence;
judging whether the absolute value of the sample entropy difference value of the current training subsequence and the current adjacent training subsequence is smaller than a difference value threshold value or not, and obtaining a first judging result;
if the first judgment result is yes, determining that the union of the current training subsequence and the current adjacent training subsequence is a modal reconstruction subsequence;
deleting the current training subsequence and the current adjacent training subsequence, taking the modal reconstruction subsequence as the current training subsequence, and returning to the step of determining any adjacent training subsequence of the current subsequence as the current adjacent training subsequence;
if the first judgment result is negative, judging whether to traverse all adjacent training subsequences of the current subsequence, and obtaining a second judgment result;
if the second judgment result is negative, updating the current adjacent training subsequence, and returning to the step of judging whether the absolute value of the sample entropy value difference value of the current training subsequence and the current adjacent training subsequence is smaller than a difference value threshold value, so as to obtain a first judgment result;
if the second judgment result is yes, judging whether all training subsequences are traversed or not to obtain a third judgment result;
if the third judgment result is negative, updating the current training subsequence, and returning to the step of determining any adjacent training subsequence of the current subsequence as the current adjacent training subsequence;
if the third judgment result is yes, completing the modal reconstruction, and determining a plurality of training subsequences as training modal reconstruction subsequences.
5. An interpretable runoff sequence prediction system, comprising:
the test set acquisition module is used for acquiring a plurality of runoff actual measurement values of the runoff to be tested in a preset time period at the current moment as a test set;
the test subsequence determining module is used for carrying out serialization decomposition on the test set by utilizing a variation modal decomposition algorithm to obtain a plurality of test subsequences;
the test mode reconstruction sub-sequence determining module is used for carrying out mode reconstruction on the plurality of test sub-sequences according to the sample entropy value to obtain a plurality of test mode reconstruction sub-sequences; the sample entropy value is determined according to a training set of a runoff sequence prediction model; runoff measurement data in a historical time period of the runoff to be measured of the training set;
the test prediction factor determining module is used for determining a test prediction factor according to the multiple test mode reconstruction subsequences and the lag subintervals; the lag subperiod is determined according to a training set of a runoff sequence prediction model;
the runoff prediction sequence determining module is used for inputting the test prediction factor into a runoff sequence prediction model to obtain a runoff prediction sequence of the runoff to be measured in the next time period at the current moment; the runoff sequence prediction model is obtained by training a support vector machine model by using a Bayesian optimization algorithm according to a historical runoff sequence of runoffs to be measured.
6. An electronic device comprising a memory for storing a computer program and a processor that runs the computer program to cause the electronic device to perform an interpretable runoff sequence prediction method as claimed in any one of claims 1 to 4.
7. The electronic device of claim 6, wherein the memory is a readable storage medium.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410026297.2A CN117828456A (en) | 2024-01-08 | 2024-01-08 | Method, system and electronic equipment for predicting interpretable runoff sequence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410026297.2A CN117828456A (en) | 2024-01-08 | 2024-01-08 | Method, system and electronic equipment for predicting interpretable runoff sequence |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117828456A true CN117828456A (en) | 2024-04-05 |
Family
ID=90505793
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410026297.2A Pending CN117828456A (en) | 2024-01-08 | 2024-01-08 | Method, system and electronic equipment for predicting interpretable runoff sequence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117828456A (en) |
-
2024
- 2024-01-08 CN CN202410026297.2A patent/CN117828456A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| CN108985335B (en) | Integrated learning prediction method for irradiation swelling of nuclear reactor cladding material | |
| CN107463993B (en) | Medium-and-long-term runoff forecasting method based on mutual information-kernel principal component analysis-Elman network | |
| CN111143981B (en) | Virtual test model verification system and method | |
| CN110633859B (en) | Hydrologic sequence prediction method integrated by two-stage decomposition | |
| CN110083125B (en) | Machine tool thermal error modeling method based on deep learning | |
| CN109947652A (en) | A kind of improvement sequence learning method of software defect prediction | |
| CN115374995A (en) | Distributed photovoltaic and small wind power station power prediction method | |
| CN111797364A (en) | Landslide multilevel safety evaluation method based on composite cloud model | |
| CN112614021B (en) | Tunnel surrounding rock geological information prediction method based on built tunnel information intelligent identification | |
| CN115982141A (en) | Characteristic optimization method for time series data prediction | |
| CN114881343A (en) | Short-term load prediction method and device of power system based on feature selection | |
| CN117272202A (en) | Dam deformation abnormal value identification method and system | |
| CN117828456A (en) | Method, system and electronic equipment for predicting interpretable runoff sequence | |
| CN116613745A (en) | PSO-ELM electric vehicle charging load prediction method based on variation modal decomposition | |
| CN115081856A (en) | Enterprise knowledge management performance evaluation device and method | |
| Zou et al. | A deep learning approach for state-of-health estimation of lithium-ion batteries based on differential thermal voltammetry and attention mechanism | |
| CN116361731A (en) | Method, device and equipment for detecting corrosion of tank wall of oil storage tank and readable storage medium | |
| CN112906991A (en) | Method and device for predicting RAFM steel radiation hardening behavior | |
| CN112307670A (en) | Design method of pressurized water reactor core parameter prediction model based on bagging integrated neural network | |
| CN113076587A (en) | Short-term prediction method for micro-strain of large-span steel structure building | |
| CN111160715A (en) | BP neural network based new and old kinetic energy conversion performance evaluation method and device | |
| CN113256018B (en) | Wind power ultra-short term probability prediction method based on conditional quantile regression model | |
| CN113673015B (en) | Intelligent system construction and parameter identification method for beam column end plate connection node optimization design | |
| Guo et al. | A learning-based method for efficient large-scale sensitivity analysis and tuning of single column atmosphere model (SCAM) |
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 |