Method and device for forecasting runoff in cold highland area, equipment and medium

Abstract

The embodiment of the invention discloses a runoff forecasting method and device for a cold highland area, equipment and a medium. The method comprises the following steps: constructing a historical daily precipitation sequence set and a historical daily runoff sequence set corresponding to each season; performing wavelet decomposition on the sequence to obtain a plurality of historical daily rainfall component sequences and historical daily runoff component sequences; performing model training based on the historical daily rainfall component sequence and the historical daily runoff component sequence to generate a daily runoff component prediction model; inputting the predicted daily rainfall component sequence into the daily runoff component prediction model to obtain a plurality of predicted daily runoff component sequences; and superposing the forecast daily runoff component sequences to obtain forecast daily runoff. According to the method, the daily runoff volume of each season is respectively forecasted, the seasonal characteristics of runoff are fully considered, the daily runoff volume is forecasted through the daily rainfall component sequence and the daily runoff component sequence containing the multi-scale information, and the precision of the forecasting result is effectively improved.

Description

Technical Field

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

[0002] The impact of climate change on the water cycle system is becoming increasingly severe, leading to an increase in the frequency and intensity of extreme hydrological events worldwide. Hydrological simulation can effectively predict and prevent extreme hydrological events.

[0003] Snowmelt runoff and rainfall runoff are important sources of river runoff in alpine mountainous areas, and their formation is influenced by multiple factors. The highly nonlinear nature of these factors makes runoff prediction in these areas extremely challenging. In recent years, with the advancement of computer technology and artificial intelligence, data-driven methods have been widely used in runoff prediction. However, existing data-driven methods generally suffer from low runoff prediction accuracy. Summary of the Invention

[0004] Based on this, it is necessary to propose runoff forecasting methods, devices, equipment and media in high-altitude and cold mountainous areas to address the above problems in order to improve the accuracy of runoff prediction results.

[0005] To achieve the above objectives, the present application provides, in a first aspect, a method for predicting runoff in high-altitude mountainous areas, the method comprising:

[0006] Collect historical daily precipitation data and historical daily runoff data for each season in the alpine mountainous area, and construct a historical daily precipitation sequence set and a historical daily runoff sequence set corresponding to a target season, wherein the target season is any one of all seasons, and the historical daily precipitation sequences in the historical daily precipitation sequence set correspond one-to-one to the historical daily runoff sequences in the historical daily runoff sequence set, and the historical daily precipitation sequences and the historical daily runoff sequences with a one-to-one correspondence are used as data pairs;

[0007] Performing wavelet decomposition on the historical daily precipitation sequence and the historical daily runoff sequence contained in the data pair to obtain a plurality of historical daily precipitation component sequences and a plurality of historical daily runoff component sequences for each data pair;

[0008] Model training is performed based on the historical daily precipitation component series and historical daily runoff component series of each data pair to generate a daily runoff component prediction model;

[0009] Predicting daily precipitation at a preset future time to obtain a predicted daily precipitation sequence, and performing wavelet decomposition on the predicted daily precipitation sequence to obtain a plurality of predicted daily precipitation component sequences, wherein the future time belongs to a preset time period of the target season;

[0010] Inputting the predicted daily precipitation component sequence into the daily runoff component prediction model to obtain a plurality of predicted daily runoff component sequences for the future time;

[0011] All of the forecast daily runoff component sequences for the future time are superimposed to obtain the forecast daily runoff for the future time.

[0012] Furthermore, the model training is performed based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair to generate a daily runoff component prediction model, specifically including:

[0013] Based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair, a target multi-layer perceptron model is constructed, wherein the target multi-layer perceptron model is used to output an initial predicted daily runoff component sequence according to the daily precipitation component sequence;

[0014] A target weighted Markov model is constructed according to the residual of the target multi-layer perceptron model, wherein the target weighted Markov model is used to correct the initial predicted daily runoff component sequence output by the target multi-layer perceptron model to obtain a corrected predicted daily runoff component sequence;

[0015] The target multi-layer perceptron model and the target weighted Markov model are coupled to generate a daily runoff component prediction model.

[0016] Furthermore, the target multi-layer perceptron model is constructed based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair, specifically including:

[0017] Model training is performed based on the n historical daily precipitation component sequences and the i-th historical daily runoff component sequence in all data pairs to obtain the i-th multilayer perceptron model, wherein the i-th multilayer perceptron model is used to predict the i-th daily runoff component sequence based on the n daily precipitation component sequences, 1≤i≤n, and the target multilayer perceptron model is composed of the 1st to nth multilayer perceptron models.

[0018] Furthermore, the initial predicted daily runoff component sequence output by the i-th multilayer perceptron model is obtained by the following formula:

[0019]

[0020] Where Q di is the initial predicted daily runoff component sequence output by the i-th multilayer perceptron model, is the weight matrix of the i-th multilayer perceptron model in the j-th layer, σ(·) is the preset activation function, P 1×n is the input vector, which consists of a sequence of n daily precipitation components. is the bias vector of the i-th multilayer perceptron model at the j-th layer, and L is the number of hidden layers.

[0021] Furthermore, constructing a target weighted Markov model based on the residual of the target multilayer perceptron model specifically includes:

[0022] Constructing a residual sequence of the i-th multilayer perceptron model in the target multilayer perceptron model, and calculating the weights of each order of the residual sequence;

[0023] The residual sequence is defined as d states, and the transition probability of the residual of the i-th multilayer perceptron model in each state is calculated according to the residual sequence and the weight of each order of the residual sequence, and the state corresponding to the maximum transition probability is used as the target state;

[0024] Calculating the eigenvalues of the residual sequence according to the probability that the residual of the i-th multilayer perceptron model is in each state;

[0025] An i-th weighted Markov model is constructed according to the eigenvalue of the residual sequence, the target state, and the upper and lower limits of the target state, wherein the i-th weighted Markov model is used to correct the initial predicted daily runoff component sequence output by the i-th multilayer perceptron model, and the target weighted Markov model is composed of the 1st to nth weighted Markov models.

[0026] Furthermore, the calculation of the weights of each order of the residual sequence specifically includes:

[0027] Calculating autocorrelation coefficients of various orders based on parameters in the residual sequence;

[0028] The weights of each order of the residual sequence are calculated based on the autocorrelation coefficients of each order.

[0029] Furthermore, the weighted Markov model is expressed by the following formula:

[0030]

[0031] Where, ε i is the correction value of the i-th multilayer perceptron model, h is the target state, H is the eigenvalue, T h is the upper limit of the target state, B h is the lower limit of the target state.

[0032] To achieve the above-mentioned purpose, the second aspect of the present application provides a runoff forecasting device for high-altitude and cold mountainous areas, the device comprising: a data acquisition unit, a data processing unit, a model training unit and a runoff forecasting unit;

[0033] The data processing unit is used to collect historical daily precipitation data and historical daily runoff data of each season in the alpine mountainous area, and construct a historical daily precipitation sequence set and a historical daily runoff sequence set corresponding to a target season, wherein the target season is any one of all seasons, and the historical daily precipitation sequences in the historical daily precipitation sequence set correspond to the historical daily runoff sequences in the historical daily runoff sequence set in a one-to-one correspondence, and the historical daily precipitation sequences and the historical daily runoff sequences with a one-to-one correspondence are used as data pairs;

[0034] The data processing unit is used to perform wavelet decomposition on the historical daily precipitation sequence and the historical daily runoff sequence contained in the data pair to obtain a plurality of historical daily precipitation component sequences and a plurality of historical daily runoff component sequences for each data pair;

[0035] The model training unit is used to perform model training based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair to generate a daily runoff component prediction model;

[0036] The runoff forecast unit is configured to predict daily precipitation at a preset future time to obtain a predicted daily precipitation sequence, and perform wavelet decomposition on the predicted daily precipitation sequence to obtain a plurality of predicted daily precipitation component sequences, wherein the future time belongs to a preset time period of the target season;

[0037] Inputting the predicted daily precipitation component sequence into the daily runoff component prediction model to obtain a plurality of predicted daily runoff component sequences for the future time;

[0038] All of the forecast daily runoff component sequences for the future time are superimposed to obtain the forecast daily runoff for the future time.

[0039] To achieve the above-mentioned objectives, the third aspect of the present application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method described in the first aspect.

[0040] To achieve the above-mentioned objectives, the fourth aspect of the present application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method described in the first aspect.

[0041] The embodiments of the present invention have the following beneficial effects:

[0042] An embodiment of the present invention proposes a runoff forecasting method for a high-altitude and cold mountainous area, the method comprising: collecting historical daily precipitation data and historical daily runoff data of each season, and constructing a historical daily precipitation sequence set and a historical daily runoff sequence set corresponding to a target season, wherein the target season is any one of all seasons, the historical daily precipitation sequences in the historical daily precipitation sequence set correspond one-to-one to the historical daily runoff sequences in the historical daily runoff sequence set, the target historical daily precipitation sequence and the target historical daily runoff sequence corresponding to the target historical daily precipitation sequence are used as a target data pair, the target historical daily precipitation sequence is any historical daily precipitation sequence in the historical daily precipitation sequence set; for all historical daily precipitation sequences and historical daily runoff sequences Wavelet decomposition is performed to obtain several historical daily precipitation component sequences and several historical daily runoff component sequences for each data pair; model training is performed based on the historical daily precipitation component sequences and historical daily runoff component sequences of each data pair to generate a daily runoff component prediction model; daily precipitation at a preset future time is predicted to obtain a predicted daily precipitation sequence, and the predicted daily precipitation sequence is subjected to wavelet decomposition to obtain several predicted daily precipitation component sequences, wherein the future time belongs to the target season; the predicted daily precipitation component sequence is input into the daily runoff prediction model to obtain several predicted daily runoff component sequences for the future time; all predicted daily runoff component sequences for the future time are superimposed to obtain the predicted daily runoff for the future time. The present invention forecasts daily runoff separately for each season, fully taking into account the seasonal characteristics of runoff to improve the accuracy of the forecast results; secondly, by performing wavelet decomposition on the daily precipitation sequence and the daily runoff sequence, multi-scale information of hydrological variables is extracted to forecast daily runoff based on the daily precipitation component sequence and the daily runoff component sequence containing multi-scale information, effectively improving the accuracy of the forecast results. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] in:

[0045] Figure 1 Schematic diagram of the flow chart of the runoff forecasting method in the alpine mountainous area according to an embodiment of the present invention;

[0046] Figure 2 This is a structural block diagram of the runoff forecasting device for high-altitude mountainous areas according to an embodiment of the present invention.

[0047] Figure 3 2 is a diagram showing the internal structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0049] Due to the large differences in temperature, humidity, rainfall and other factors in different seasons, the hydrological phenomena in the same area in different seasons will also be significantly different. In order to better simulate and predict future hydrological phenomena, the embodiment of the present invention proposes a runoff forecasting method for high-altitude mountainous areas. By considering the influence of seasonal factors, the error caused by seasonal factors in the runoff forecast results is avoided, thereby improving the accuracy of the runoff forecast results in high-altitude mountainous areas. Figure 1 , Figure 1 This is a flow chart of a method for predicting runoff in a high-altitude mountainous area according to an embodiment of the present invention. The method includes:

[0050] Step 110: Collect historical daily precipitation data and historical daily runoff data for each season in the alpine mountainous area, and construct a historical daily precipitation sequence set and a historical daily runoff sequence set corresponding to the target season, wherein the target season is any one of all seasons, and the historical daily precipitation sequences in the historical daily precipitation sequence set correspond one-to-one to the historical daily runoff sequences in the historical daily runoff sequence set, and the historical daily precipitation sequences and historical daily runoff sequences with a one-to-one correspondence are used as data pairs.

[0051] Alpine mountainous areas, characterized by high altitudes, cold climates, and complex terrain, are the source of many rivers. Predicting runoff in these regions not only facilitates rational water resource planning and ensures water supply to downstream areas, but also provides early warning of natural disasters such as floods and mudslides caused by melting ice and snow, thereby minimizing losses.

[0052] Because runoff in alpine mountainous areas varies significantly from season to season, this paper analyzes runoff for each season to improve the accuracy of predictions. Specifically, daily precipitation and runoff data from past years in alpine regions are collected. Historical daily runoff and precipitation series are constructed for spring, summer, autumn, and winter, respectively. This yields sets of historical daily precipitation and runoff series corresponding to each season.

[0053] It is understandable that daily precipitation and daily runoff at the same time affect each other and have a one-to-one correspondence. Therefore, the daily precipitation series and daily runoff series with a one-to-one correspondence can be used as a data pair. For example, if a historical daily precipitation series contains the daily precipitation for each day in March of 2019, then the historical daily runoff series corresponding to the historical daily precipitation series contains the daily runoff for each day in March of 2019.

[0054] Step 120: Perform wavelet decomposition on the historical daily precipitation sequence and the historical daily runoff sequence contained in the data pair to obtain several historical daily precipitation component sequences and several historical daily runoff component sequences for each data pair.

[0055] Specifically, the Daubechies wavelet can be used to decompose the historical daily precipitation series and the historical daily runoff series in each data pair to obtain several historical daily precipitation component series and several historical daily runoff component series for each data pair.

[0056] Taking the decomposition of the historical daily precipitation series as an example, the historical daily precipitation series is decomposed by formula (1):

[0057]

[0058] Where, P(t) is the decomposition of the historical daily precipitation series, P di (t) is the i-th historical daily precipitation component sequence, and n is the total number of historical daily precipitation component sequences.

[0059] In one embodiment of the present invention, the scaling function and wavelet function of the Daubechies wavelet satisfy formulas (2) and (3):

[0060]

[0061] Where h k is the low-pass filter coefficient, g k is the high-pass filter coefficient, both depend on the order of Daubechies wavelet, and k is the order of Daubechies wavelet.

[0062] In one embodiment of the present invention, the Daubechies 6 wavelet can be used to decompose the historical daily precipitation series for four seasons to obtain several daily precipitation component sequences. For example, the historical daily precipitation series can be decomposed into eight historical daily precipitation component sequences, where component sequences d1-d7 are high-frequency components that represent the periodicity of the original series, and component sequence d8 is a low-frequency component that represents the trend of the original series.

[0063] It is understandable that the decomposition process of the historical daily runoff series is basically the same as the decomposition process of the historical daily precipitation series, so we will not elaborate on it here.

[0064] By performing wavelet decomposition on daily precipitation series and daily runoff series to extract multi-scale information of daily precipitation and daily runoff, daily runoff forecasting can be performed based on the daily precipitation component series and daily runoff component series containing multi-scale information, which can effectively improve the accuracy of the forecast results.

[0065] Step 130: Perform model training based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair to generate a daily runoff component prediction model.

[0066] In an embodiment of the present invention, a prediction model is trained based on the relationship between historical daily precipitation component sequences and historical daily runoff component sequences. Specifically, a training sample set is constructed based on the historical daily precipitation component sequences and historical daily runoff component sequences for each data pair. The training sample set is then divided into a training set and a test set according to a preset ratio (for example, the ratio can be set to 4:1). The model is trained and tested based on the training and test sets to obtain a trained daily runoff component prediction model. The daily runoff component prediction model takes the daily precipitation component sequence as input and outputs the daily runoff component sequence as output.

[0067] Step 140: predict the daily precipitation at a preset future time to obtain a predicted daily precipitation sequence, and perform wavelet decomposition on the predicted daily precipitation sequence to obtain several predicted daily precipitation component sequences, wherein the future time belongs to a preset time period of the target season.

[0068] After model training is complete, the generated daily runoff component prediction model can be used to predict future daily runoff components. Specifically, first determine the time period for which daily runoff forecasting is desired, such as a specific future time, and determine the season to which the future time belongs. This is then used to predict the daily runoff components for that season using the daily runoff component prediction model. Secondly, the daily precipitation for the future time is predicted to obtain a predicted daily precipitation sequence. Finally, the predicted daily precipitation is subjected to wavelet decomposition to obtain several future daily precipitation component sequences.

[0069] Step 150: Input the predicted daily precipitation component sequence into the daily runoff component prediction model to obtain several predicted daily runoff component sequences for future times.

[0070] Specifically, several predicted daily precipitation component sequences for the future time are input into the daily runoff component prediction model corresponding to the season of the future time, and several predicted daily precipitation component sequences output by the daily runoff component prediction model are obtained.

[0071] Step 160: Superimpose all the forecast daily runoff component sequences for the future time to obtain the forecast daily runoff for the future time.

[0072] After predicting all the forecast daily runoff component sequences for the future time, all the forecast daily runoff component sequences are superimposed to obtain the forecast daily runoff sequence for the future time, so as to obtain the forecast daily runoff for any day in the future time.

[0073] In the embodiment of the present invention, the forecast daily runoff series for the future time is calculated by formula (4):

[0074]

[0075] Where, Q is the forecast daily runoff series; Q di is the runoff component sequence for the i-th forecast day, and n is the total number of runoff component sequences for the forecast day.

[0076] In an embodiment of the present invention, a corresponding daily runoff component prediction model is constructed for each season to predict the daily runoff in the time periods corresponding to different seasons, thereby eliminating the impact of seasonal changes on the prediction results; in addition, an embodiment of the present invention also performs wavelet decomposition on the daily precipitation sequence and the daily runoff sequence to effectively extract the multi-scale information of the hydrological variables, so as to forecast the daily runoff based on the daily precipitation component sequence and the daily runoff component sequence containing the multi-scale information, thereby improving the accuracy of the forecast results.

[0077] For high-altitude mountainous areas, runoff is affected by a variety of factors. Since these influencing factors have highly nonlinear characteristics, existing runoff prediction methods are difficult to capture. In addition, systematic errors may occur when predicting daily runoff, resulting in inaccurate prediction results. Based on this, an embodiment of the present invention proposes a daily runoff component prediction model that couples a multi-layer perceptron model and a weighted Markov model. The nonlinear characteristics are obtained through the daily runoff component prediction model, and the predicted daily runoff component sequence is corrected to make the corrected prediction results more accurate. Specifically, step 130, based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair, a model training is performed to generate a daily runoff component prediction model, which specifically includes:

[0078] Step 310: Based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair, a target multi-layer perceptron model is constructed, wherein the target multi-layer perceptron model is used to output an initial predicted daily runoff component sequence based on the daily precipitation component sequence.

[0079] The multilayer perceptron model can fit complex nonlinear relationships. Therefore, the multilayer perceptron model is used to conduct deep learning on the relationship between the daily precipitation component sequence and the daily runoff component sequence, and the multilayer perceptron model is used to predict the daily runoff component sequence to improve the accuracy of the prediction results.

[0080] In one embodiment of the present invention, the target multi-layer perceptron model construction process is as follows: model training is performed based on the n historical daily precipitation component sequences and the i-th historical daily runoff component sequence in all data pairs to obtain the i-th multi-layer perceptron model, wherein the i-th multi-layer perceptron model is used to predict the i-th daily runoff component sequence based on the n daily precipitation component sequences, 1≤i≤n, and the target multi-layer perceptron model is composed of the 1st to nth multi-layer perceptron models.

[0081] Specifically, if a data pair contains n historical daily precipitation component sequences and n historical daily runoff component sequences, then this data pair can generate n training samples, each of which contains the n historical daily precipitation component sequences and one daily runoff component sequence from the data pair. It is understood that each component sequence is uniquely numbered, so the i-th multilayer perceptron model is trained based on the training samples corresponding to the daily runoff component sequences with the same number.

[0082] For example, data pair A and data pair B both contain n historical daily precipitation component sequences and n historical daily runoff component sequences. Data pair A can generate at least n training samples, with the i-th training sample being (the first historical daily precipitation component sequence, the second historical daily precipitation component sequence, ..., the n-th historical daily precipitation component sequence, the i-th historical daily runoff component sequence); data pair B can also generate at least n training samples, with the i-th training sample being (the first historical daily precipitation component sequence, the second historical daily precipitation component sequence, ..., the n-th historical daily precipitation component sequence, the i-th historical daily runoff component sequence). Therefore, the i-th multilayer perceptron model can be trained using the i-th training sample of data pair A and the i-th training sample of data pair B, so that the i-th multilayer perceptron model predicts the i-th daily runoff component sequence based on the n daily precipitation component sequences.

[0083] The specific model training process is as follows:

[0084] First, all training samples are divided into a training set and a test set according to a preset ratio. The i-th multilayer perceptron model is trained based on the samples in the training set, and the mean square error of the i-th multilayer perceptron model is calculated. The mean square error is calculated using formula (5):

[0085]

[0086] Where, and are the predicted value and true value of the i-th daily runoff component series of the k-th sample, respectively, and M is the number of samples in the training set.

[0087] Second, the loss function of the i-th multilayer perceptron model is determined based on the mean squared error (MSE). The gradient of the loss function with respect to each weight and bias of the i-th multilayer perceptron model is calculated, and the model weights and biases are updated using gradient descent or other optimization algorithms. During training, the i-th multilayer perceptron model is validated using the test set to evaluate its accuracy and adjust hyperparameters such as the number of hidden layers and neurons based on the results.

[0088] In the embodiment of the present invention, n multi-layer perceptron models can be obtained using the above method, and finally a target multi-layer perceptron model is constructed based on the n multi-layer perceptron models.

[0089] In one embodiment of the present invention, the initial predicted daily runoff component sequence output by the i-th multi-layer perceptron model is obtained by formula (6):

[0090]

[0091] Where Q di is the initial predicted daily runoff component sequence output by the i-th multi-layer perceptron model, is the weight matrix of the i-th multilayer perceptron model in the j-th layer, σ(·) is the preset activation function, P 1×n is the input vector, which consists of n daily precipitation component sequences, P 1×n =[P d1 ,P d2 ,P d3 …P dn ], is the bias vector of the i-th multilayer perceptron model at the j-th layer, and L is the number of hidden layers.

[0092] Step 320: construct a target weighted Markov model based on the residual of the target multi-layer perceptron model. The target weighted Markov model is used to correct the initial predicted daily runoff component sequence output by the target multi-layer perceptron model to obtain a corrected predicted daily runoff component sequence.

[0093] After the target multi-layer perceptron model is constructed, the residual of the target multi-layer perceptron model is calculated, and a target weighted Markov model is constructed based on the residual of the target multi-layer perceptron model, so as to correct the predicted value of the target multi-layer perceptron model using the target weighted Markov model.

[0094] The specific construction process of the target weighted Markov model is as follows:

[0095] Step 321. Construct the residual sequence of the i-th multilayer perceptron model in the target multilayer perceptron model and calculate the weights of each order of the residual sequence.

[0096] In the embodiment of the present invention, the residual sequence of each multilayer perceptron model is constructed separately, such as the residual sequence of the i-th multilayer perceptron model in, Then calculate the weights of each order of the residual sequence of the i-th multilayer perceptron model.

[0097] Specifically, the specific process of calculating the weights of each order of the residual sequence is as follows:

[0098] Step 3211. Calculate the autocorrelation coefficients of each order based on the parameters in the residual sequence.

[0099] In the embodiment of the present invention, the autocorrelation coefficients of each order of the residual sequence of the i-th multilayer perceptron model can be calculated by formula (7):

[0100]

[0101] Where r c is the autocorrelation coefficient of order c, order c is the number of state transition steps selected to predict system variables, m is the maximum order, generally 4 to 6, u1 and u2 are the means of the two residual sequences before and after divided by the order, and n is the total number of residuals in the residual sequence.

[0102] Step 3212: Calculate the weights of each order of the residual sequence based on the autocorrelation coefficients of each order.

[0103] In the embodiment of the present invention, the weights of each order of the residual sequence of the i-th multilayer perceptron model are calculated by formula (8):

[0104]

[0105] Where w c is the weight of order c, r c is the autocorrelation coefficient of order c, and m is the maximum order.

[0106] Step 322: Define the residual sequence as d states, calculate the transition probability of the residual of the i-th multi-layer perceptron model in each state according to the residual sequence and the weights of each order of the residual sequence, and take the state corresponding to the maximum transition probability as the target state.

[0107] The weighted Markov chain can be defined by the state transition probability matrix. When the residual sequence is divided into d states, the one-step transition probability matrix P1 is a d-order square matrix. The value p in the f-th row and g-th column of the one-step transition probability matrix P1 is f,g Expressed as:

[0108]

[0109] Where p f,g is the probability of transitioning from state f to state g in one step, S f is the number of states f in the residual sequence; S f,g is the number of one-step transitions from state f to state g.

[0110] Based on this, the state transition probability matrix P of order c c It can be expressed as:

[0111]

[0112] Taking the first selected basic moment as the initial state, combining the state transition probability matrix of each order and the state of the lag period, the predicted state transition probability matrix P is constructed. The predicted state transition probability matrix P can be expressed as:

[0113]

[0114] Where, is the probability value of state f at the cth order.

[0115] The transition probability of the residual of the i-th multilayer perceptron model in each state can be calculated by formula (13):

[0116]

[0117] Where p f is the transition probability that the residual of the i-th multilayer perceptron model is in state f.

[0118] The transition probabilities of the residuals of the i-th multilayer perceptron model falling into each state are compared, and the state with the largest transition probability is the predicted state corresponding to the predicted target.

[0119] Step 323. Calculate the eigenvalues of the residual sequence based on the probability that the residual of the i-th multilayer perceptron model is in each state.

[0120] First, according to the probability of the residual of the i-th multilayer perceptron model being in each state, the fuzzy numbers of each order are determined. The fuzzy numbers are obtained by formula (13):

[0121]

[0122] Where, d f is a fuzzy number, δ is the maximum probability effect index. The larger its value is, the more it can highlight the maximum probability effect. It is generally set to 2 to 4.

[0123] Secondly, the eigenvalues of the residual sequence are calculated based on the fuzzy numbers of each order. The eigenvalues are calculated using formula (14):

[0124]

[0125] Where H is the eigenvalue.

[0126] Step 324: Construct the i-th weighted Markov model based on the eigenvalues of the residual sequence, the target state, and the upper and lower limits of the target state, wherein the i-th weighted Markov model is used to correct the initial predicted daily runoff component sequence output by the i-th multilayer perceptron model, and the target weighted Markov model is composed of the 1st to nth weighted Markov models.

[0127] In the embodiment of the present invention, the weighted Markov model is expressed by formula (15):

[0128]

[0129] Where, ε i is the correction value of the i-th multilayer perceptron model, h is the target state, H is the eigenvalue, T h is the upper limit of the target state, B h is the lower limit of the target state.

[0130] It can be understood that the weighted Markov model corresponding to each multilayer perceptron model can be obtained by the above method. After obtaining n weighted Markov models corresponding to n multilayer perceptron models, the target weighted Markov model is constructed based on the n weighted Markov models.

[0131] Step 330: Couple the target multi-layer perceptron model and the target weighted Markov model to generate a daily runoff component prediction model.

[0132] In the embodiment of the present invention, the daily runoff component prediction model is obtained by coupling the target multi-layer perceptron model and the target weighted Markov model. The daily runoff component prediction model outputs Q di =Q i +ε i , where Q di is the runoff component sequence of the i-th forecast day, Q i is the initial predicted daily runoff component sequence, ε i is the correction value of the i-th multilayer perceptron model.

[0133] In one embodiment of the present invention, a runoff forecasting device for high-altitude and cold mountainous areas is also provided. Figure 2 , Figure 22 is a structural block diagram of a runoff forecasting device for high-altitude mountainous areas according to an embodiment of the present invention. The device includes: a data acquisition unit 201 , a data processing unit 202 , a model training unit 203 and a runoff forecasting unit 204 .

[0134] The data processing unit 201 is used to collect historical daily precipitation data and historical daily runoff data in each season in the alpine mountainous area, and construct a historical daily precipitation sequence set and a historical daily runoff sequence set corresponding to the target season, wherein the target season is any one of all seasons, and the historical daily precipitation sequences in the historical daily precipitation sequence set correspond one-to-one to the historical daily runoff sequences in the historical daily runoff sequence set, and the historical daily precipitation sequences and historical daily runoff sequences with a one-to-one correspondence are used as data pairs.

[0135] The data processing unit 202 is configured to perform wavelet decomposition on the historical daily precipitation sequence and the historical daily runoff sequence contained in the data pair to obtain a plurality of historical daily precipitation component sequences and a plurality of historical daily runoff component sequences for each data pair.

[0136] The model training unit 203 is used to perform model training based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair to generate a daily runoff component prediction model.

[0137] The runoff forecasting unit 204 is used to predict the daily precipitation at a preset future time to obtain a predicted daily precipitation sequence, and perform wavelet decomposition on the predicted daily precipitation sequence to obtain a plurality of predicted daily precipitation component sequences, wherein the future time belongs to a preset time period of the target season.

[0138] The predicted daily precipitation component sequence is input into the daily runoff component prediction model to obtain several predicted daily runoff component sequences in the future.

[0139] All the forecast daily runoff component sequences for the future time are superimposed to obtain the forecast daily runoff for the future time.

[0140] The runoff forecasting device for alpine mountainous areas proposed in an embodiment of the present invention forecasts daily runoff separately for each season, fully taking into account the seasonal characteristics of runoff to improve the accuracy of the forecast results; secondly, by performing wavelet decomposition on the daily precipitation sequence and the daily runoff sequence, the multi-scale information of the hydrological variables is effectively extracted, and the daily runoff is forecasted based on the daily precipitation component sequence and the daily runoff component sequence containing multi-scale information, thereby effectively improving the accuracy of the forecast results.

[0141] Figure 3 FIG1 shows the internal structure of a computer device in one embodiment of the present invention. The computer device can be a terminal or a system. Figure 3As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement the various steps in the above method embodiment. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can implement the various steps in the above method embodiment. It will be understood by those skilled in the art that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0142] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes each step in the above method embodiment.

[0143] In one embodiment, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the processor executes the steps in the above method embodiment.

[0144] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0145] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0146] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person of ordinary skill in the art may make various modifications 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 present invention shall be determined by the appended claims.

Claims

1. A method for predicting runoff in alpine mountainous areas, characterized in that: The method comprises: Collect historical daily precipitation data and historical daily runoff data for each season in the alpine mountainous area, and construct a historical daily precipitation sequence set and a historical daily runoff sequence set corresponding to a target season, wherein the target season is any one of all seasons, and the historical daily precipitation sequences in the historical daily precipitation sequence set correspond one-to-one to the historical daily runoff sequences in the historical daily runoff sequence set, and the historical daily precipitation sequences and the historical daily runoff sequences with a one-to-one correspondence are used as data pairs; Performing wavelet decomposition on the historical daily precipitation sequence and the historical daily runoff sequence contained in the data pair to obtain a plurality of historical daily precipitation component sequences and a plurality of historical daily runoff component sequences for each data pair; Model training is performed based on the historical daily precipitation component series and historical daily runoff component series of each data pair to generate a daily runoff component prediction model; Predicting daily precipitation at a preset future time to obtain a predicted daily precipitation sequence, and performing wavelet decomposition on the predicted daily precipitation sequence to obtain a plurality of predicted daily precipitation component sequences, wherein the future time belongs to a preset time period of the target season; Inputting the predicted daily precipitation component sequence into the daily runoff component prediction model to obtain a plurality of predicted daily runoff component sequences for the future time; All of the forecast daily runoff component sequences for the future time are superimposed to obtain the forecast daily runoff for the future time.

2. The method according to claim 1, wherein The model training is performed based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair to generate a daily runoff component prediction model, specifically including: Based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair, a target multi-layer perceptron model is constructed, wherein the target multi-layer perceptron model is used to output an initial predicted daily runoff component sequence according to the daily precipitation component sequence; A target weighted Markov model is constructed according to the residual of the target multi-layer perceptron model, wherein the target weighted Markov model is used to correct the initial predicted daily runoff component sequence output by the target multi-layer perceptron model to obtain a corrected predicted daily runoff component sequence; The target multi-layer perceptron model and the target weighted Markov model are coupled to generate a daily runoff component prediction model.

3. The method according to claim 2, wherein The target multi-layer perceptron model is constructed based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair, specifically including: Model training is performed based on the n historical daily precipitation component sequences and the i-th historical daily runoff component sequence in all data pairs to obtain the i-th multilayer perceptron model, wherein the i-th multilayer perceptron model is used to predict the i-th daily runoff component sequence based on the n daily precipitation component sequences, 1≤i≤n, and the target multilayer perceptron model is composed of the 1st to nth multilayer perceptron models.

4. The method according to claim 3, wherein The initial predicted daily runoff component sequence output by the i-th multilayer perceptron model is obtained by the following formula: Where Q di is the initial predicted daily runoff component sequence output by the i-th multilayer perceptron model, is the weight matrix of the i-th multilayer perceptron model in the j-th layer, σ(·) is the preset activation function, P 1×n is the input vector, which consists of a sequence of n daily precipitation components. is the bias vector of the i-th multilayer perceptron model at the j-th layer, and L is the number of hidden layers.

5. The method according to claim 3, wherein The target weighted Markov model is constructed according to the residual of the target multi-layer perceptron model, specifically comprising: Constructing a residual sequence of the i-th multilayer perceptron model in the target multilayer perceptron model, and calculating the weights of each order of the residual sequence; The residual sequence is defined as d states, and the transition probability of the residual of the i-th multilayer perceptron model in each state is calculated according to the residual sequence and the weight of each order of the residual sequence, and the state corresponding to the maximum transition probability is used as the target state; Calculating the eigenvalues of the residual sequence according to the probability that the residual of the i-th multilayer perceptron model is in each state; An i-th weighted Markov model is constructed according to the eigenvalue of the residual sequence, the target state, and the upper and lower limits of the target state, wherein the i-th weighted Markov model is used to correct the initial predicted daily runoff component sequence output by the i-th multilayer perceptron model, and the target weighted Markov model is composed of the 1st to nth weighted Markov models.

6. The method according to claim 5, wherein The calculating of the weights of each order of the residual sequence specifically includes: Calculating autocorrelation coefficients of various orders based on parameters in the residual sequence; The weights of each order of the residual sequence are calculated based on the autocorrelation coefficients of each order.

7. The method according to claim 5, wherein The weighted Markov model is expressed by the following formula: Where, ε i is the correction value of the i-th multilayer perceptron model, h is the target state, H is the eigenvalue, T h is the upper limit of the target state, B h is the lower limit of the target state.

8. A runoff forecasting device for high-altitude and cold mountainous areas, characterized in that: The device comprises: a data acquisition unit, a data processing unit, a model training unit and a runoff forecasting unit; The data processing unit is used to collect historical daily precipitation data and historical daily runoff data of each season in the alpine mountainous area, and construct a historical daily precipitation sequence set and a historical daily runoff sequence set corresponding to a target season, wherein the target season is any one of all seasons, and the historical daily precipitation sequences in the historical daily precipitation sequence set correspond to the historical daily runoff sequences in the historical daily runoff sequence set in a one-to-one correspondence, and the historical daily precipitation sequences and the historical daily runoff sequences with a one-to-one correspondence are used as data pairs; The data processing unit is used to perform wavelet decomposition on the historical daily precipitation sequence and the historical daily runoff sequence contained in the data pair to obtain a plurality of historical daily precipitation component sequences and a plurality of historical daily runoff component sequences for each data pair; The model training unit is used to perform model training based on the historical daily precipitation component sequence and the historical daily runoff component sequence of each data pair to generate a daily runoff component prediction model; The runoff forecast unit is configured to predict daily precipitation at a preset future time to obtain a predicted daily precipitation sequence, and perform wavelet decomposition on the predicted daily precipitation sequence to obtain a plurality of predicted daily precipitation component sequences, wherein the future time belongs to a preset time period of the target season; Inputting the predicted daily precipitation component sequence into the daily runoff component prediction model to obtain a plurality of predicted daily runoff component sequences for the future time; All of the forecast daily runoff component sequences for the future time are superimposed to obtain the forecast daily runoff for the future time.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 7.

10. A computer device comprising a memory and a processor, characterized in that: The memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 7.

Images