Runoff prediction method, device, equipment and medium

Abstract

The invention discloses a runoff prediction method, a runoff prediction device, runoff prediction equipment and a medium, and relates to the technical field of hydrological forecasting, original runoff sequence data are decomposed by using variational mode decomposition (VMD), and signal complexity of subsequences obtained after the original runoff sequence data are decomposed is evaluated by using fuzzy entropy FE, so that the runoff prediction accuracy is improved. A VFM method is used for setting runoff subsequence recombination components with low signal complexity, medium signal complexity and high signal complexity as a trend term, a periodic term and a random term, rich feature information in an original runoff sequence is extracted in the process, and meanwhile, runoff short-time and long-time periodic change trends are decomposed and displayed; and then a BiLSTM prediction model is adopted to predict the trend term, the periodic term and the random term to obtain various predicted values, and the various predicted values are added to forecast the runoff change trend of the hydrological station, so that the accuracy of runoff prediction is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological forecasting, and in particular to a runoff prediction method, device, equipment and medium. Background Art

[0002] Surface runoff is an important component of the water cycle and is of great significance for maintaining ecological and environmental stability and supporting human social development. Accurate runoff forecasting and research on the inherent variation characteristics of runoff are crucial for preventing flood disasters, scientifically managing water resources, and socioeconomic development. Due to the influence of climate change and human activities, runoff often exhibits highly nonlinear and non-steady-state variation characteristics, especially in river basins with large seasonal changes, where flow changes are often very large. Therefore, obtaining high-precision runoff forecast results and analyzing and studying the accurate variation trends and periodic variation patterns of runoff are very challenging.

[0003] At present, runoff prediction models are often divided into process-driven models and data-driven models; process-driven models can reflect the specific physical processes and formation mechanisms of the water cycle, but the calculation process is relatively complex and requires more meteorological data. In addition, parameter debugging also takes a lot of time and has certain limitations; data-driven models require less data, have fast calculation speeds, and can still show satisfactory results when facing complex influencing factors. With the continuous improvement of computer performance and the advancement of prediction models, data-driven models represented by machine learning have become a trend in runoff prediction. Artificial neural networks have better adaptability when dealing with nonlinear complex systems; researchers have used long short-term memory neural networks (LSTM), gated recurrent networks (GRU) and temporal convolutional networks (TCN) to predict runoff, extract runoff and process its long time series data.

[0004] Due to the influence of multiple complex meteorological factors, runoff series presents highly nonlinear characteristics. It is difficult to achieve satisfactory prediction results if only a single neural network model is used. To address this phenomenon, the "decomposition-integration" model that combines data preprocessing methods and neural network models has been widely used in the field of runoff prediction. The core of this type of model is to decompose a complex sequence into several subsequences with lower prediction difficulty through mathematical methods, and improve the prediction accuracy by predicting the subsequences and then summing them up. For the processing of subsequences, some researchers use empirical mode decomposition (EMD), but the EMD method is affected by the modal aliasing effect; some researchers apply ensemble empirical mode decomposition (EEMD) and variational mode decomposition (VMD). VMD can decompose the signal into a series of intrinsic mode components with different center frequencies. After decomposition, subsequences with different center frequencies can be regarded as having multiple time scales, and these multi-time scale components contain some characteristics of the original sequence.

[0005] However, because the extracted runoff sequence is affected by multiple complex meteorological factors, the runoff sequence is highly complex. The current analysis methods can often only identify some shallow features contained in the runoff sequence to analyze the runoff changes, ignoring the rich information in the runoff subsequence after signal decomposition and the relationship between the subsequence and the original runoff, so the runoff forecast accuracy is low. Summary of the Invention

[0006] The embodiments of the present invention provide a runoff prediction method, device, equipment and medium, which can solve the problem in the prior art that the current analysis methods can often only identify some shallow features contained in the runoff sequence to analyze the runoff changes, ignoring the rich information in the runoff subsequence after signal decomposition and the relationship between the subsequence and the original runoff, thereby causing low runoff prediction accuracy.

[0007] An embodiment of the present invention provides a runoff prediction method, comprising the following steps:

[0008] Obtain historical runoff data from hydrological stations and form runoff series data;

[0009] The runoff sequence data is decomposed by using variational mode decomposition (VMD) to obtain intrinsic mode component (IMF) subsequences with different center frequencies. The signal complexity of the intrinsic mode component (IMF) subsequences is evaluated using fuzzy entropy (FE) to obtain reconstructed runoff components with different signal complexities.

[0010] The VFM runoff component extraction method is used to set the recombined runoff components with low complexity as the runoff trend item, the recombined runoff components with medium complexity as the runoff period item, and the recombined runoff components with high complexity as the runoff random item. The low, medium and high complexity of the recombined runoff components are determined by pre-set thresholds.

[0011] The BiLSTM model is used to predict the runoff trend item, runoff period item and runoff random item to obtain the trend item prediction value, period item prediction value and random item prediction value. The trend item prediction value, period item prediction value and random item prediction value are added together, and the runoff change trend of the hydrological station is predicted based on the addition result.

[0012] Preferably, when using variational mode decomposition VMD to decompose the runoff sequence data, the fast Fourier transform spectrum of the last IMFs subsequence of the decomposed runoff sequence is observed to determine whether the decomposed intrinsic mode component IMFs subsequence has modal aliasing, and to determine the number of decomposition layers when decomposing the runoff sequence data.

[0013] Preferably, the using fuzzy entropy FE to evaluate the complexity of the intrinsic mode component IMFs subsequence to obtain the complexity of the intrinsic mode component IMFs subsequences of different center frequencies includes:

[0014] Variational mode decomposition (VMD) defines each IMF as a frequency-amplitude modulated signal, and represents the intrinsic mode component IMFs subsequences of different center frequencies as {u(i):1≤i≤N}. For a given N-dimensional time series [u(1),u(2),…,u(N)], the phase space dimension m (m≤N-2) and the similarity tolerance r are defined, and the phase space is reconstructed; it is quantized into a formal vector as follows:

[0015]

[0016]

[0017] Introduce fuzzy membership function:

[0018]

[0019] The r-dimensional similarity tolerance, for i=1,2…,N-m+1, is:

[0020]

[0021] The maximum absolute distance between window vectors X(i) and X(j) is:

[0022]

[0023] For each i, find the average value and get:

[0024]

[0025] For each vector Average of its adjacent vectors All similarities of , we get:

[0026]

[0027] And according to each vector after averaging Adjacent vectors of All similarities of and Then the fuzzy entropy FE of the intrinsic mode component IMFs subsequence with different center frequencies is:

[0028]

[0029] Preferably, the acquisition of the runoff trend term, the runoff period term and the runoff random term includes:

[0030] After obtaining the intrinsic mode component IMFs subsequences with different center frequencies, the fuzzy entropy FE is used to obtain the complexity of the intrinsic mode component IMFs subsequences with different center frequencies, and at the same time obtain the average complexity of all intrinsic mode component IMFs subsequences

[0031] According to the average The IMFs subsequence of the center frequency is divided into runoff trend term, runoff period term and runoff random term according to the difference of , including:

[0032] The runoff trend term is: The runoff period term is: The runoff random term is: (1-r) and (1+r) are the ranges of the complexity of the intrinsic mode component IMFs subsequence, r = 0.15 × SD is the boundary width, and SD is the standard deviation of the time series.

[0033] Preferably, when the runoff trend term, the runoff period term and the runoff random term are predicted using the BiLSTM model, the runoff trend term, the runoff period term and the runoff random term are normalized, and the normalization method is:

[0034]

[0035] Where: x norm represents the normalized value; x i represents the observed value; x max Indicates the maximum value; x min Indicates the minimum value.

[0036] An embodiment of the present invention further provides a runoff prediction device, comprising:

[0037] Data processing module, used to obtain historical runoff data of hydrological sites and form runoff series data;

[0038] The decomposition module is used to decompose the runoff sequence data using variational mode decomposition (VMD) to obtain intrinsic mode component (IMF) subsequences with different center frequencies; the fuzzy entropy (FE) is used to evaluate the signal complexity of the intrinsic mode component (IMF) subsequences to obtain reconstructed runoff components with different signal complexities;

[0039] The VFM runoff component extraction method is used to set the recombined runoff components with low complexity as the runoff trend item, the recombined runoff components with medium complexity as the runoff period item, and the recombined runoff components with high complexity as the runoff random item. The low, medium and high complexity of the recombined runoff components are determined by pre-set thresholds.

[0040] The prediction module is used to predict the runoff trend item, runoff period item and runoff random item using the BiLSTM model, obtain the trend item prediction value, period item prediction value and random item prediction value, and add the trend item prediction value, period item prediction value and random item prediction value to predict the runoff change trend of the hydrological station based on the addition result.

[0041] An embodiment of the present invention further provides an electronic device, including a memory and a processor;

[0042] The memory is used to store computer programs;

[0043] The processor is configured to implement the steps of the above-mentioned runoff prediction method when executing the computer program stored in the memory.

[0044] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the steps of the runoff prediction method described above.

[0045] The embodiments of the present invention provide a runoff prediction method, device, equipment, and medium. Compared with the prior art, the methods and devices have the following beneficial effects:

[0046] The present invention uses variational mode decomposition (VMD) to decompose the original runoff sequence data, uses fuzzy entropy (FE) to evaluate the signal complexity of the subsequences obtained after the decomposition of the original runoff sequence data, and uses the VFM method to set the recombined components of the runoff subsequences with low, medium and high signal complexity as trend terms, periodic terms and random terms respectively. This process extracts rich feature information from the original runoff sequence, and at the same time decomposes and displays the short-term and long-term periodic change trends of the runoff. Then, the BiLSTM prediction model is used to predict the trend term, periodic term and random term respectively to obtain the predicted values ​​of each item, and the predicted values ​​are added together to predict the runoff change trend of the hydrological station, which greatly improves the accuracy of runoff prediction.

[0047] Moreover, the trend term, periodic term and random term decomposed by the present invention have obvious periodic characteristics, and this characteristic is consistent with the changing trend of the original runoff sequence, which can better explain the changing law and intrinsic characteristics of runoff. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of the overall flow of a runoff prediction method provided by an embodiment of the present invention;

[0049] Figure 2 A schematic diagram of a specific flow chart of a runoff prediction VFMB model of a runoff prediction method provided in an embodiment of the present invention;

[0050] Figure 3A schematic diagram of the BiLSTM structure of a runoff prediction method provided by an embodiment of the present invention;

[0051] Figure 4 A schematic diagram of original daily runoff information of three stations in a runoff prediction method provided by an embodiment of the present invention;

[0052] Figure 5 Schematic diagram of modal aliasing of the last IMF at each station in a runoff prediction method provided by an embodiment of the present invention; (a) is a comparison between IMF6 and IMF7; (b) is a comparison between IMF7 and IMF8;

[0053] Figure 6 FE schematic diagram of IMFs at three sites of a runoff prediction method provided by an embodiment of the present invention;

[0054] Figure 7 Schematic diagram of the division of IMF under different r conditions for a runoff prediction method provided by an embodiment of the present invention; (a) is a comparison of IMF1, IMF2, IMF4, and IMF5 obtained after decomposition of the daily runoff series at station B; (b) is a comparison of IMF1 and IMF2 obtained after decomposition of the daily runoff series at station A; (c) is a comparison of IMF1 and IMF2 obtained after decomposition of the daily runoff series at station C;

[0055] Figure 8 A schematic diagram of the extraction results of trend items, period items, and random items at three stations in a runoff prediction method provided by an embodiment of the present invention;

[0056] Figure 9 A schematic diagram of daily runoff prediction at three stations according to a runoff prediction method provided in an embodiment of the present invention;

[0057] Figure 10 A schematic diagram of runoff trend items and original runoff extracted by three different methods at three stations in a runoff prediction method provided by an embodiment of the present invention;

[0058] Figure 11 A schematic diagram of a real part wavelet coefficient contour map and a wavelet variance curve of a three-station periodic term of a runoff prediction method provided by an embodiment of the present invention;

[0059] Figure 12 A schematic diagram of the fuzzy entropy of an IMF composed of random terms and the fuzzy entropy of random terms in a runoff prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0060] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0061] See also Figure 1 The embodiment of the present invention provides a runoff prediction method, specifically designing a VFMB prediction model; first, using variational mode decomposition (VMD) and fuzzy entropy (FE) to evaluate the signal complexity of the subsequences obtained after decomposing the original sequence, based on the difference in fuzzy entropy of different subsequences, the signal complexity and change characteristics of the recombined subsequences are comprehensively analyzed, and a runoff subsequence classification standard is proposed. A VFM runoff classification method is established to obtain three runoff items in the daily runoff sequence, namely, trend item, periodic item, and random item; finally, a neural network model is used to predict the different runoff items separately, and then the sum is summed to obtain the final prediction result.

[0062] The specific process of the VFMB prediction model is as follows Figure 2 As shown, including:

[0063] Step 1: Use VMD to adaptively decompose the original runoff data and observe the fast Fourier transform spectrum of the last IMF of the decomposed runoff sequence to determine whether modal aliasing occurs. This method is used to determine the optimal number of decomposition layers.

[0064] Step 2: After the decomposition in step 1, fuzzy entropy (FE) is used to evaluate the signal complexity of all runoff IMFs. All IMFs are divided according to the different fuzzy entropies to obtain reconstructed runoff components with different signal complexities.

[0065] Step 3: The stage of classification of different runoff components. In this stage, all reconstructed runoff components are given practical meanings.

[0066] ①The recombinant component composed of IMFs with the lowest signal complexity was selected as the runoff trend item. The trend comparison analysis of the original runoff sequence and the recombinant component was performed to compare the similarity of the change characteristics of the runoff trend item and the original sequence.

[0067] ②The recombined components composed of IMFs with medium signal complexity are selected as the runoff periodic terms, and the wavelet analysis is performed on the periodic terms.

[0068] ③The recombinant component composed of IMFs with higher signal complexity is selected as the runoff random term.

[0069] Step 4: Use 80% of the data as the training set, 10% as the validation set, and 10% as the test set; first normalize the divided runoff trend term, cycle term, and trend term; then use BiLSTM to predict the three divided runoff terms separately, and sum the prediction results to obtain the final prediction result. The normalization formula is:

[0070]

[0071] Where: x norm represents the normalized value; x i represents the observed value; x max Indicates the maximum value; x min Indicates the minimum value.

[0072] Specifically:

[0073] Variational Mode Decomposition (VMD) is applied as follows:

[0074] Variational mode decomposition can decompose the signal into a series of intrinsic mode components (IMFs). It does not require a preset window function during decomposition, and the parameter selection is simple. At the same time, it overcomes the defects of empirical mode decomposition (EMD) that are prone to mode mixing and endpoint effects. The runoff series is affected by factors such as seasonal rainfall and presents non-stationary characteristics. It is more appropriate to use VMD to decompose the runoff series. VMD treats each IMF as an AM-FM signal u k (t), to evaluate the bandwidth, each constrained modal optimization problem is formulated as:

[0075]

[0076] Where: {u k} represents the set of K patterns decomposed by VMD; {ω k} represents the corresponding center frequencies of all IMFs; Represents the partial derivative of the function with respect to t; δ(t) represents the impulse function; * represents convolution.

[0077] Used to convert constrained optimization problems into unconstrained optimization problems, the augmented Lagrangian is defined as follows:

[0078]

[0079] Where: α represents the penalty parameter; λ represents the Lagrange multiplier; and then the alternating direction multiplier method (ADMM) algorithm is used to update each component u k and the center frequency ω of each component k , and finally the saddle point of the unconstrained model is obtained, which is the optimal solution of the above equation. The updated equation is:

[0080]

[0081]

[0082]

[0083] Where: n is the number of iterations; τ is the noise tolerance; ε is the identification accuracy; u k (t) represents the real part u of the inverse FT k (ω); Respectively f(t),λ n The Fourier transform of (t).

[0084] Fuzzy entropy (FE) is applied as follows:

[0085] Fuzzy entropy (FE) can be effectively applied to time series polluted by noise and has little dependence on the length of the time series. For the time series {u(i):1≤i≤N}, its formal vector is:

[0086]

[0087] in: represents m consecutive u values, starting from the i-th point, and can be obtained by deleting the baseline:

[0088]

[0089] Given a vector The adjacent vectors are calculated by the following formula and Similarity for:

[0090]

[0091] in: Indicates the maximum absolute difference of corresponding scalar components For each vector Average of its adjacent vectors All similarities can be obtained:

[0092]

[0093] Build and The fuzzy entropy of a time series function can be defined as:

[0094]

[0095] Runoff IMFs are categorized as follows:

[0096] The IMFs obtained by decomposing the runoff sequence have different central frequencies and signal complexities, so the FE of all IMFs is calculated first, and then the average value of FE is calculated. according to The IMFs are divided into three categories: Category 1 (trend items): Category II (cyclical items): IMFs’ FE Between, the third category (random items): Where (1-r) and (1+r) represent the complexity range of IMFs; the method is composed of adaptive decomposition, signal complexity discrimination, and runoff subsequence classification criteria.

[0097] Bidirectional long short-term memory BiLSTM is:

[0098] BiLSTM runs two independent LSTMs at each time step, one from the beginning to the end of the sequence, t∈[1,T], and the other from the end to the beginning of the sequence, t∈[T,1]. The outputs of these two LSTMs are combined to consider both past and future information. The structure of BiLSTM is shown in the figure below. Figure 3 As shown, the hidden layer state Ht of BiLSTM at time t contains the positive and reverse for:

[0099]

[0100]

[0101]

[0102] The performance indicators are:

[0103] The present invention uses the Nash-Sutcliffe efficiency coefficient (NSE), the determination coefficient (R 2 ), mean absolute error (MAE) and root mean square error (RMSE) are used to evaluate the prediction results of the model. The specific formula is as follows:

[0104]

[0105]

[0106]

[0107]

[0108] Where: N represents the number of test samples of the prediction model; Q0(i) represents the actual value of the monthly runoff series; Q f (i) indicates the corresponding predicted value; Indicates the average of actual values; Represents the mean of the predicted values.

[0109] Specific experiments:

[0110] 1. Decomposition of runoff time series.

[0111] Collect the original daily runoff information of three hydrological stations, such as Figure 4 As shown in Figure 1, in order to decompose the runoff sequence, the VMD method is used to process the data and divide the complex runoff sequence into a series of IMFs with different central frequencies. The appropriate VMD decomposition level K is very important. In order to avoid over-decomposition or incomplete decomposition of the runoff sequence, K is determined by observing whether the last IMF has modal aliasing. For station A, when K = 7, as shown in Figure 1, Figure 5 As shown in (a), no modal aliasing occurs. When K = 8, as shown in Figure 5 As shown in (b), the modal aliasing phenomenon occurs in the 8th-order IMF. To meet the orthogonality constraint, the decomposition level of the stream sequence of station A is set to K = 7. The same method is used to determine K for stations B and C, and both are set to K = 6.

[0112] 2. Signal complexity of runoff IMFs.

[0113] Different runoff IMFs have different variation characteristics. FE is often used to evaluate the complexity of short time series signals contaminated by noise. To evaluate the complexity of different IMFs, it is necessary to calculate the FE of all IMFs. When calculating FE, three parameters m, r, and N must be selected and determined. For the length of the comparison window m, a larger m can reconstruct the dynamic process in more detail. The FE of the IMFs of the three stations is as follows: Figure 6 As shown, it can be seen that at the same station, some IMFs have similar signal complexity and similar changing patterns.

[0114] 3. Classification method of runoff IMFs based on finite element average value.

[0115] In order to reconstruct the runoff subsequences with different signal complexities reasonably, the runoff subsequence division standard was formulated according to the mean value of FE. The IMFs are divided into three categories according to the difference in r; when r is 0.1, IMF1, IMF2, IMF4, and IMF5 obtained after the decomposition of the daily runoff series at station B are divided into one category. Figure 7As shown in (a), the IMF1 of station B is significantly different from the other three IMFs. The variation characteristics of IMF1 are close to the general trend of the original daily runoff, and the other four IMFs have a certain periodic variation law. When r is 0.3, the IMF1 and IMF2 obtained after the decomposition of the daily runoff series of station A are divided together, as shown in Figure 7 As shown in (b), the characteristics of IMF1 are close to the change trend of runoff at station A, while IMF2 has certain periodic characteristics. The two have different characteristics. When r is 0.1, 0.2, and 0.3, IMF1, IMF3, and IMF6 at station C are combined as trend items. When r is 0.4, IMF1 and IMF6 are divided together. Figure 7 As can be seen from (c), the two have different variation patterns. When r is 0.2, the similar daily runoff IMFs of the three stations are reasonably divided together, and there is no situation where the IMFs with different variation patterns are incorrectly divided together. Therefore, for the IMFs of the three test stations, the average FE of these IMFs is calculated. according to The differences in these similar IMFs can be divided into three categories: Cyclical item: IMFs FE Between, random items:

[0116] 4. Extraction of runoff trend terms, periodic terms and random terms.

[0117] In the runoff sequence, the trend term reflects the overall change direction of the runoff, which changes relatively slowly. The periodic term may contain multiple runoff change cycles and is more complex than the trend term. The runoff random term contains most of the noise interference and irregular components and has the highest signal complexity. After calculating the FE of the runoff IMFs at each station, it can be found that some IMFs at the same station have similar complexity. The extraction results of the trend term, periodic term, and random term of the three stations are shown in Figure 2. Figure 8 shown.

[0118] 5. Comparison of prediction results between VFMB prediction model and comparison model.

[0119] In the VFMB prediction model proposed in the present invention, the fuzzy entropy of the IMFs after each runoff decomposition is first calculated, and then the signal complexity evaluation and subsequence reorganization of the IMFs are performed to extract three runoff components with practical significance, namely the trend term, the periodic term, and the random term. Finally, the BiLSTM model is used to predict the trend term, the periodic term, and the random term, and the prediction results of the three terms are added to obtain the final prediction result. In order to prove the prediction performance of the model proposed in the present invention, four models, CNN, GRU, LSTM, and BiLSTM, are selected as benchmark models for comparison. The main hyperparameters of the models are shown in Table 1. Four evaluation indicators are used to evaluate the prediction results of each model. The prediction results of the training set, validation set, and test set of all models are shown in Table 2. The VFMB prediction model designed by the present invention not only divides different runoff components at the three test sites but also realizes accurate runoff prediction, overcoming the characteristics of drastic seasonal changes and large runoff in the daily runoff series in the Huaihe River Basin. The prediction results of the five models, CNN, LSTM, GRU, BiLSTM, and VFMB, are shown in Table 2. Figure 9 shown.

[0120] Table 1 Hyperparameters of different models

[0121]

[0122] Table 2 Evaluation indicators of three sites

[0123]

[0124] 6. Test prediction results.

[0125] The present invention selects 7 groups of flood prediction results at each test site to compare the flood forecasting capabilities of different models. The specific flood peak forecast results and errors are shown in Table 3.

[0126] Table 3 Flood forecast results of three hydrological stations

[0127]

[0128]

[0129] It can be seen that the RMSE and MAE of the VFMB prediction model designed in the present invention are significantly lower than those of the other four comparison models at the three stations; the lowest RMSE at station A is 207.82, and the lowest MAE is 107.94; the lowest RMSE at station B is 279.11, and the lowest MAE is 231.82; at station C, the lowest RMSE is 50.72, and the lowest MAE is 41.63. It can be seen that the VFMB prediction model has the smallest prediction deviation for the flood peak, and the flood forecast results are closer to the actual flood.

[0130] like Figure 10As shown, after analyzing the extracted trend items using the VFMB prediction model designed by the present invention, the trend components of the runoff at the three sites are consistent with the changing trend of the original runoff, proving the accuracy of the VFMB prediction model designed by the present invention.

[0131] like Figure 11 As shown in the figure, the red area indicates higher runoff, the blue area indicates lower runoff, and the other colors indicate intermediate runoff. It can be seen that in 2005, the runoff cycle item of station A had two cycles of 11 days and 24 days. During the period of 235 days to 260 days, the runoff cycle item experienced a change of high flow-low flow-high flow-low flow. Winter and spring are low flow periods, which is consistent with the change law of actual runoff. In 2006, the runoff cycle item of station B had two cycles of 34 days and 40 days. The runoff cycle The term shows obvious and stable periodic changes. On the scale of 34 days, its oscillation frequency is high and very complex. Within 100 days to 350 days, the flow shows the characteristics of repeated low and high flows. In 2005, the runoff period term of Station C has three cycles of 16 days, 26 days and 33 days, which is similar to the variation pattern of Station A. Within 220 days to 250 days, on the scale of 16 days, the period term of Station C shows obvious low flow-high flow-low flow-high flow variation, which is in line with the variation pattern of actual runoff.

[0132] like Figure 12 As shown, it can be seen that stations A, B, and C all contain several IMFs with higher FE. These runoff components fluctuate complexly and contain more noise. The runoff IMFs of each station are divided into three categories, among which the runoff random term of each station is reorganized by several IMFs with higher complexity. It can be seen that after the reorganization, the FE of the random term of the runoff of the three stations is smaller than before the reorganization. At stations A and C, the FE of the random term is significantly lower than before the reorganization.

[0133] The present invention can extract trend terms, periodic terms and random terms in runoff while providing accurate runoff forecasts, overcoming the problem of insufficient forecasting ability of a single model for daily runoff; at the same time, similar runoff components can be divided together according to the differences in signal complexity of runoff subsequences, and subsequence reorganization can be performed, and parameter selection is simple, without the need for multiple debugging; moreover, the extracted trend terms can reflect the short-term and long-term change trends of runoff, the extracted periodic terms have obvious periodic characteristics, and the change period is consistent with the change trend of the original runoff during the flood season, and the extracted trend terms are more consistent with the trend of the original runoff than the trend terms obtained by the STL method; and the designed model, while improving the prediction accuracy, reconstructs the subsequences in the "decomposition-integration" prediction model research into runoff components with practical significance, better revealing the changing laws and inherent characteristics of complex natural daily runoff.

[0134] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A runoff prediction method, characterized in that: The following steps are involved: Obtain historical runoff data from hydrological stations and form runoff series data; The runoff sequence data is decomposed by using variational mode decomposition (VMD) to obtain the intrinsic mode component (IMFs) subsequences with different center frequencies. The signal complexity of the intrinsic mode component (IMFs) subsequences is evaluated by using fuzzy entropy (FE) to obtain reconstructed runoff components with different signal complexities. The VFM runoff component extraction method is used to set the recombined runoff components with low complexity in the recombined runoff components as runoff trend items, the recombined runoff components with medium complexity in the recombined runoff components are set as runoff period items, and the recombined runoff components with high complexity in the recombined runoff components are set as runoff random items; wherein the low, medium and high complexity of the recombined runoff components are determined by pre-set thresholds; The BiLSTM model is used to predict the runoff trend item, runoff period item and runoff random item to obtain the trend item prediction value, period item prediction value and random item prediction value. The trend item prediction value, period item prediction value and random item prediction value are added together, and the runoff change trend of the hydrological station is predicted based on the addition result.

2. A runoff prediction method according to claim 1, characterized in that: When the runoff sequence data is decomposed by using variational mode decomposition VMD, the fast Fourier transform spectrum of the last IMFs subsequence of the decomposed runoff sequence is observed to determine whether the decomposed intrinsic mode component IMFs subsequence has mode aliasing, and to determine the number of decomposition layers when decomposing the runoff sequence data.

3. A runoff prediction method according to claim 1, characterized in that: The method of using fuzzy entropy FE to evaluate the complexity of the intrinsic mode component IMFs subsequence to obtain the complexity of the intrinsic mode component IMFs subsequences with different center frequencies includes: Variational mode decomposition (VMD) defines each IMF as a frequency-amplitude modulated signal, and represents the intrinsic mode component IMFs subsequences of different center frequencies as {u(i):1≤i≤N}. For a given N-dimensional time series [u(1),u(2),…,u(N)], the phase space dimension m (m≤N-2) and the similarity tolerance r are defined, and the phase space is reconstructed; and it is quantized into a formal vector as follows: Introduce fuzzy membership function: The r-dimensional similarity tolerance, for i=1,2…,N-m+1, is: The maximum absolute distance between window vectors X(i) and X(j) is: For each i, find the average value and get: For each vector Average of its adjacent vectors All similarities of , we get: And according to each vector after averaging The adjacent vectors of All similarities of and Then the fuzzy entropy FE of the intrinsic mode component IMFs subsequence with different center frequencies is:

4. A runoff prediction method according to claim 1, characterized in that: The acquisition of the runoff trend term, the runoff cycle term and the runoff random term includes: After obtaining the intrinsic mode component IMFs subsequences with different center frequencies, the fuzzy entropy FE is used to obtain the complexity of the intrinsic mode component IMFs subsequences with different center frequencies, and the average complexity of all the intrinsic mode component IMFs subsequences is obtained at the same time. According to the average The IMFs subsequence of the center frequency is divided into runoff trend term, runoff period term and runoff random term according to the difference of , including: The runoff trend term is: The runoff period term is: The runoff random term is: (1-r) and (1+r) are the ranges of the complexity of the IMFs subsequences, r=0.15×SD is the boundary width, and SD is the standard deviation of the time series.

5. A runoff prediction method according to claim 1, characterized in that: When the BiLSTM model is used to predict the runoff trend term, the runoff cycle term and the runoff random term, the runoff trend term, the runoff cycle term and the runoff random term are normalized, and the normalization method is: Where: x norm represents the normalized value; x i represents the observed value; x max Indicates the maximum value; x min Indicates the minimum value.

6. A runoff prediction device, characterized in that: include: The data processing module is used to obtain the historical runoff data of the hydrological station and form the runoff sequence data; The decomposition module is used to decompose the runoff sequence data using variational mode decomposition (VMD) to obtain intrinsic mode component IMFs subsequences with different center frequencies; the fuzzy entropy FE is used to evaluate the signal complexity of the intrinsic mode component IMFs subsequence to obtain reconstructed runoff components with different signal complexities; The VFM runoff component extraction method is used to set the recombined runoff components with low complexity in the recombined runoff components as runoff trend items, the recombined runoff components with medium complexity in the recombined runoff components are set as runoff period items, and the recombined runoff components with high complexity in the recombined runoff components are set as runoff random items; wherein the low, medium and high complexity of the recombined runoff components are determined by pre-set thresholds; The prediction module is used to predict the runoff trend item, runoff period item and runoff random item using the BiLSTM model, obtain the trend item prediction value, period item prediction value and random item prediction value, add the trend item prediction value, period item prediction value and random item prediction value, and predict the runoff change trend of the hydrological station according to the addition result.

7. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is used to implement the steps of a runoff prediction method as described in any one of claims 1 to 5 when executing the computer program stored in the memory.

8. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the steps of a runoff prediction method as described in any one of claims 1 to 5.

Images