Multi-scale incoming water sequence generation method and system considering runoff change characteristics

Abstract

The invention discloses a multi-scale incoming water sequence generation method and system considering runoff change characteristics, and the method comprises the steps: building annual and monthly incoming water joint distribution based on a Copula function, carrying out the monthly average flow estimation under different incoming water frequencies, and obtaining a monthly scale runoff process; secondly, respectively carrying out feature extraction on more than ten-year flood and small-scale flood to obtain session flood flow hydrograph of different durations; thirdly, session floods with different durations and flood volumes are combined, the occurrence time of the floods is adjusted, superposition is carried out to obtain a high-water-period inflow water sequence closest to the historical flood form, in-month flow distribution is carried out, and the inflow water scale of a key month of the flood season is shortened to the daily scale; and finally, combining the monthly-scale dry-season flow with the daily-scale wet-season flow to obtain a monthly-daily mixed-scale annual water sequence considering runoff change characteristics.

Description

Technical Field

[0001] This invention relates to the field of heat wave subseason forecasting technology, specifically to a multi-scale inflow sequence generation method and system that considers runoff variation characteristics. Background Technology

[0002] Predicting annual water series for rivers is quite challenging. In practice, river basin management agencies often use information from meteorological agencies to predict annual water levels (abundant, average, or low), obtaining estimates of the total annual water volume or average annual flow, which are then allocated to each month according to a certain proportion. This method typically only estimates the total annual water volume and average monthly flow, resulting in monthly forecast data. The generated annual water series cannot include the different flood characteristics of each month, nor does it contain daily-scale water volume information.

[0003] Furthermore, when allocating the total annual runoff to each month according to a certain proportion, a typical annual process is generally sampled and scaled down. The annual runoff distribution in wet, normal, and dry years is scaled down proportionally to obtain the annual runoff process line at the design frequency. Although the above process is simple to operate, the simple scaling of the water inflow in a representative year cannot take into account the intra-annual and inter-annual differences in water inflow at different frequencies.

[0004] This type of annual inflow data, which does not consider the characteristics of water inflow variations, can only be used for rough water resource scheduling and power generation planning. Furthermore, due to the significant difference between the average monthly flow during the high-water season and the actual flow, and the excessively large time scale (monthly or decadal), plans based on monthly inflows often experience substantial deviations during implementation. In particular, when using monthly inflow data to formulate annual power generation plans, there is often a large discrepancy between the planned and actual water heads, causing the power generation plan to deviate from reality. This significantly reduces the effectiveness of the annual power generation plan and creates considerable trouble for water resource management and reservoir scheduling. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a multi-scale inflow sequence generation method and system that considers runoff variation characteristics to solve the problems of excessively large time scales and significant forecast deviations in the existing technology, thereby providing more refined and reliable basic data for the preparation of annual power generation plans.

[0006] Technical solution: The present invention provides a method for generating multi-scale inflow sequences considering runoff variation characteristics, comprising the following steps:

[0007] S1. Establish the frequency distribution of annual water volume and obtain the marginal distribution of annual water volume; estimate the marginal distribution of monthly water volume and map the marginal distribution of each month to a uniform distribution; construct the joint distribution between annual water volume and monthly water volume.

[0008] S2. Evaluate the goodness of fit of different joint distribution models and select the optimal joint distribution model; based on the selected joint distribution model, generate the annual monthly runoff according to the specified inflow frequency.

[0009] S3. For the high-water season, process large and small floods separately to generate daily-scale runoff processes; identify and extract the characteristics of small-scale floods and classify them into flood waveform families of different durations; classify the baseflow and obtain the original baseflow of the river channel; subtract the original baseflow from the flood hydrograph to obtain the superposition amount of each flood event;

[0010] S4. Using the month as a window, by adjusting the occurrence time and interval of floods, multiple floods are superimposed to generate the monthly inflow process line; the monthly window is slidable and the incompletely calculated flood process is inherited to complete the superposition of the entire inflow sequence during the high water season, thus obtaining the assumed inflow process.

[0011] S5. Assess the morphological similarity between the assumed inflow process and the historical runoff process of the same period, and select the optimal matching scheme; allocate the monthly flow according to the optimal matching scheme, and shorten the inflow forecast step of the key months of the flood season to the daily scale.

[0012] S6. Combine the monthly dry season flow with the daily wet season flow to obtain the annual water series at the monthly-daily mixed scale.

[0013] Furthermore, in step S1, the distribution of water inflow for each month is estimated using the kernel density estimation method, and its cumulative distribution function is obtained through numerical integration.

[0014] Furthermore, in step S3, the differential recognition algorithm is used to identify the peaks and troughs of small-scale floods, and the small-scale floods are divided into flood waveform families with different durations by cluster analysis.

[0015] Furthermore, the cluster analysis specifically involves: initially dividing the flood events into multiple clusters based on their duration; updating the cluster centers through iterative calculations until the cluster centers stabilize; outputting multiple cluster centers and adjusting their durations to integer multiples, and allocating flood volumes to obtain daily-scale flow process lines.

[0016] Furthermore, in step S3, a digital filtering method is used to divide the dry season flow data into baseflows to obtain the original baseflow of the river channel.

[0017] Furthermore, the digital filtering method treats the runoff process as a composite signal, with the baseflow being the low-frequency component and precipitation runoff being the high-frequency component, and separates the baseflow through filtering.

[0018] 7. A method for generating multi-scale inflow sequence considering runoff variation characteristics according to claim 1, characterized in that, in step S4, different monthly runoff processes are generated by adjusting the number, frequency combination, or peak location of flood events.

[0019] 8. A method for generating multi-scale inflow sequence considering runoff variation characteristics according to claim 1, characterized in that, in step S4, the dynamic time warping method is used to calculate the morphological similarity between the assumed inflow process and the historical runoff process of the same period, and the minimum path cumulative distance is used as the criterion.

[0020] 9. A multi-scale inflow sequence generation system considering runoff variation characteristics, characterized in that it comprises:

[0021] The construction module is used to establish the frequency distribution of annual total water volume, obtain the marginal distribution of annual total water volume, estimate the marginal distribution of monthly water inflow, and map the marginal distribution of each month to a uniform distribution; and construct the joint distribution between annual total water volume and monthly water inflow.

[0022] Evaluation module: Used to evaluate the goodness of fit of different joint distribution models and select the optimal joint distribution model; based on the selected joint distribution model, it generates the annual monthly runoff according to the specified inflow frequency;

[0023] The overlay module is used to process large and small floods separately during the high-water season to generate daily-scale runoff processes; identify and extract the characteristics of small-scale floods and classify them into flood waveform families of different durations; classify the baseflow and obtain the original baseflow of the river channel; and subtract the original baseflow from the flood hydrograph to obtain the overlay amount of each flood event.

[0024] Water Inflow Process Module: This module uses a monthly window to generate monthly water inflow process lines by superimposing multiple flood events by adjusting the occurrence time and interval of the floods; it slides the monthly window and inherits the flood processes that are not fully calculated to complete the superposition of the entire water inflow sequence during the high-water season and obtain the assumed water inflow process.

[0025] Optimal module: used to evaluate the morphological similarity between the assumed inflow process and the historical runoff process of the same period, select the optimal matching scheme; allocate the monthly flow according to the optimal matching scheme, and shorten the inflow forecast step of the key months of the flood season to the daily scale;

[0026] Combination module: Used to combine monthly dry season flow with daily wet season flow to obtain a monthly-daily mixed annual water series.

[0027] An electronic device according to the present invention includes a memory and a processor. The memory stores a computer program, and the processor executes the program to implement the steps of the method.

[0028] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: By establishing a joint distribution of monthly and annual inflow based on a Copula function, and combining monthly and daily inflow data, this invention generates a mixed monthly-daily scale annual inflow sequence, significantly improving the precision and accuracy of inflow forecasting. Furthermore, this invention employs differential recognition algorithms and cluster analysis methods to extract and classify features of small-scale floods, generating flow process lines for flood events of different durations, and accurately delineates baseflow through digital filtering, improving the accuracy of flood processing. Through dynamic time warping, this invention evaluates the morphological similarity between assumed inflow processes and historical runoff processes, selecting the optimal matching scheme, further improving forecast reliability. The modular design of the system enables automated processing of multi-scale inflow sequence generation, improving efficiency and practicality. This invention provides more refined and reliable basic data for water resource management and reservoir scheduling. Attached Figure Description

[0029] Figure 1 This is a flowchart of the present invention;

[0030] Figure 2 This is a schematic diagram of the superimposed flood generating monthly runoff process of the present invention;

[0031] Figure 3 This is a graph showing the theoretical distribution and historical data of annual water frequency in this invention;

[0032] Figure 4 This is a graph of the probability density function of the average monthly flow rate of the present invention;

[0033] Figure 5 This is a graph showing the monthly Copula function sampling results of this invention;

[0034] Figure 6 These are monthly runoff process diagrams for three frequencies according to the present invention;

[0035] Figure 7 This invention provides a flow chart of a flood event occurring once every ten years or more.

[0036] Figure 8 This is a schematic diagram of the flow composition of the present invention;

[0037] Figure 9 This is a simulation result diagram of the 25% frequency water flow sequence of the present invention;

[0038] Figure 10 This is a simulation result diagram of the 50% frequency water flow sequence of the present invention;

[0039] Figure 11 This is a simulation result diagram of the 75% frequency water flow sequence of the present invention. Detailed Implementation

[0040] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0041] like Figure 1 As shown, this embodiment of the invention provides a method for generating multi-scale inflow sequences that considers runoff variation characteristics, including the following steps:

[0042] S1. Assuming that the annual water volume frequency of the basin follows a P-III type distribution, the frequency is calculated and the parameters are calibrated using historical data to obtain the marginal distribution of the random variable X (annual water volume).

[0043] S2, using the Gaussian kernel function to estimate the monthly water inflow Y1, Y2, Y3, ... Y 12 The distribution of water flow is obtained by integrating the composite Simpson's rule to obtain the cumulative distribution function. A probability integral transformation is then used to map the marginal distributions of each month to a uniform distribution on [0, 1]. Since the monthly water inflows are Y1, Y2, Y3, ... Y... 12 The distribution type of the random variable X is unknown, and its marginal distribution lacks available expressions and parameters. Gaussian kernel density estimation (GKDE) can be used to estimate its distribution. This method estimates the distribution of a random variable using a sample through cumulative integration. The probability density functions for different values ​​x of the random variable X are as follows: where F(x) is the estimated probability density function at point x, n is the number of data points, h is the smoothing parameter, and K is the Gaussian kernel function.

[0044] (1)

[0045] (2)

[0046] S3 uses the Frank-Copula, Clayton-Copula, and Gumbel-Copula functions to construct the joint distribution of monthly and annual water inflows, and uses the maximum likelihood method to estimate the parameters.

[0047] S4. Calculate the Spearman rank correlation coefficient, Euclidean distance, and Akaike Information Criterion (AIC) value to measure the goodness of fit of different Copula functions. Compare the goodness of fit values ​​of different Copula functions to select the final monthly and yearly combined water inflow distribution.

[0048] S5. Conditional simulation is performed using the established joint distribution of annual and monthly inflow. The annual inflow frequency is selected, and the annual and monthly inflow generation method based on the Copula function is used to estimate the annual monthly runoff at the specified inflow frequency, thus obtaining the total inflow for each month.

[0049] Monthly runoff is divided into dry season and wet season. Since the average flow is similar in dry season months, the monthly average flow can be directly used as the dry season flow. The inflow varies significantly in wet season months and needs to be handled separately for different flood sizes.

[0050] S6. For major floods with a return period of more than once in ten years during the high-water season, the same-scale runoff process within the month is obtained by scaling down the calculation using the same-scale method. That is, the scaling-up ratio is determined by comparing the design flood volume with the typical flood volume, and then the typical flood process is scaled up to obtain flood processes of different frequencies. Large floods at frequencies of 10%, 5%, 3.3%, 2%, and 1% can be performed to obtain the corresponding flood processes.

[0051] S7. For small floods with a return period of less than once in ten years during the high-water season, since it is impossible to obtain a flood process that matches reality by reducing the typical design flood process, it is necessary to identify and extract features of small floods. Therefore, a second-order difference identification algorithm is used to identify the peaks and troughs of small floods in a long series of inflow sequences, calculate the duration of the flood, and divide the small floods into families of flood waveforms with different durations using cluster analysis. The identification of peaks and troughs of small floods in a long series of inflow sequences using the second-order difference identification algorithm includes the following steps:

[0052] S71, for the flow sequence Q(i), first calculate the first difference:

[0053] (3)

[0054] S72, perform another difference operation on the first-order difference result to obtain the second-order difference result:

[0055] (4)

[0056] S73, then determine whether Q(i) is a peak or trough of the flood wave based on the sign of the second-order difference value:

[0057] (5)

[0058] S74, after identifying all small-scale floods in a long-term runoff process, uses K-means clustering to extract flood features. This includes the following steps:

[0059] S741, first divide the flood events into K clusters according to their duration, and determine the cluster center by minimizing the sum of squared intervals within the cluster;

[0060] S742, re-divide the clusters according to the new cluster centers and calculate the new cluster centers;

[0061] S743 iterates multiple times until the cluster centers no longer change, outputting K cluster centers. The sum of squared intra-cluster distances is calculated as follows. Where, T i and W i Let represent the duration and volume of the i-th flood, respectively. and The duration and flood volume of the cluster center.

[0062] (6)

[0063] S744 extends or shortens the duration of the flood at the cluster center to an integer multiple of 24 hours in the final output, and distributes the flood volume to the daily-scale flow process lines of small-scale floods on each day. In this invention, K=5 is used, which can obtain small-scale flood flow processes with durations of 3 days, 4 days, 5 days, 6 days, and 7 days.

[0064] S8. A digital filtering method is used to segment the baseflow of the dry season flow data from the past ten years, and the original baseflow of the river channel is obtained through weighted averaging. The monthly runoff process obtained by the staggered superposition of multiple floods leads to repeated accumulation of baseflow from multiple flood events. Assuming that the original baseflow is generated by groundwater recharge during the dry season or in the absence of precipitation, the baseflow is altered by the direct increase in runoff during the flood season. To avoid double counting or omissions, the original baseflow component in each flood event must be removed before superposition to obtain the direct runoff generated by precipitation and the baseflow change caused by precipitation, which can be superimposed as the amount of each flood event. The purpose of the baseflow segmentation in this invention is to obtain the impact of the corresponding precipitation on the river channel flow by subtracting the original baseflow from the flood event data. Therefore, dry season runoff data is used for baseflow segmentation to serve as the original baseflow in the absence of precipitation during the wet season.

[0065] Furthermore, digital filtering is a widely used method in signal processing. It uses mathematical techniques to process a composite signal composed of signals of different frequencies, removing low-frequency or high-frequency components to obtain the desired information. In this invention, the runoff process is the composite signal. The baseline flow can be considered as the low-frequency component of the signal, which changes gradually and has a long period. Precipitation runoff is considered as the high-frequency component of the signal, which varies greatly, occurs occasionally, and has a short period.

[0066] The most basic digital filtering method is the Lyne-Hollick filter, which eliminates phase shift through two recursive steps, forward and backward. The calculation formula is as follows: Where α is the filtering parameter, with a value range of [0.9, 0.99]; Q... b (i) represents the fundamental current in time period i, Q sum (i) represents the total runoff during time period i.

[0067] (7)

[0068] S9 obtains the flood process curves under different inflow frequencies, and subtracts the original baseflow to obtain the "overlapping amount of each flood event".

[0069] S10 uses an exhaustive method to superimpose flood data, with a one-month window length: several typical flood hydrographs are selected, and different flood occurrence times and intervals are tried. Multiple floods are superimposed to obtain the monthly inflow hydrograph. The flood duration can exceed the window's defined range, but the excess is included in the next window to ensure the total flood volume within the window approximates the total monthly inflow. A schematic diagram of the assumed monthly inflow process for superimposed flood events is attached. Figure 2 As shown in the figure, the flood component of each flood event is the "superimposable amount of each flood event" minus the baseflow. Different monthly runoff processes can be generated by adjusting the number and frequency combination of flood events, or by adjusting the peak position of the same flood event combination.

[0070] S11, the window length is shifted backward each time, and the flood process that was not fully calculated in the previous window is inherited. Step 10 is repeated until the superposition of the entire flood season inflow sequence is completed, and the "hypothetical inflow process" is obtained.

[0071] S12, calculate the morphological similarity between the "hypothetical water inflow process" formed by various superposition methods and the historical runoff processes of the same period, and obtain the scheme with the best matching effect. The morphological similarity between the hypothetical water inflow process and the historical process of the same period can be measured by using the Dynamic Time Warping (DTW) method to calculate the minimum alignment path cumulative distance. The smaller the cumulative distance, the more similar the morphologies of the two sequences are. DTW is a classic algorithm for measuring the morphological similarity of two time series. Compared with directly calculating the Euclidean distance of the sequences, it can effectively avoid misjudgments caused by time series misalignment and supports the morphological similarity judgment of two sequences of different lengths. For two water inflow sequences X={ }、Y={ } Calculate the Euclidean distance between each point in the two sequences, generating an n×m Euclidean distance matrix. Generate a path from one corner (1, 1) to another corner (n, m) of the matrix. The cumulative path distance is the sum of the elements traversed. The shortest path is found using dynamic programming, which is the minimum cumulative path distance. The calculation formula is as follows: Let K be the Euclidean distance between the points along the path and K be the number of points along the path.

[0072] (8)

[0073] S13, using the best matching scheme, allocates the monthly flow obtained in S5, shortening the water inflow forecast step for key months of the flood season to a daily scale.

[0074] S14 combines the monthly dry season flow with the daily wet season flow to obtain a monthly-daily mixed annual runoff sequence that takes into account runoff variation characteristics.

[0075] Specific experiment:

[0076] S1. Assuming that the annual water volume frequency of the basin follows a P-III type distribution, the frequency is calculated and the parameters are calibrated using historical data to obtain the marginal distribution of the random variable X (annual water volume).

[0077] In this embodiment, a key section of a reservoir in a southern river basin is used as an example for case analysis. The average annual runoff is considered as a random variable X, assuming that variable X follows a P-III distribution. The fit between historical data and the theoretical distribution curve is observed to verify its distribution pattern. Based on the average annual runoff data of the past fifty years (1975-2024), the fitting method is used for parameter estimation, and the fitting results are shown in Table 1. After multiple fitting methods, the parameters with the best fit are: average flow rate of 75.8 m³ / s, coefficient of variation of 0.66, and skewness coefficient of 1.32. The theoretical cumulative distribution curve of the average annual flow rate and historical data are attached. Figure 3 As shown. From Figure 3 As can be seen, the theoretical cumulative distribution curve of the annual average flow fits well with the historical data. The historical data are all located near the theoretical curve and are concentrated in the latter half of the curve, indicating that the annual average runoff of the basin conforms to the P-III distribution.

[0078] Table 1. Results of the final alignment. ;

[0079] S2, using the Gaussian kernel function to estimate the monthly water inflow Y1, Y2, Y3, ... Y 12 The distribution of the cumulative distribution function is obtained by integral of the composite Simpson's rule, and the marginal distribution of each month is mapped to a uniform distribution on [0, 1] by probability integral transformation.

[0080] Treating the monthly water inflow of each hydrological year as random variables Y1, Y2, Y3, ... Y 12 Because of variable Y n The distribution type is unclear and needs to be selected and confirmed based on historical data. The monthly average runoff for different months over the past fifty years (1975-2024) is used as random variables Y1, Y2, Y3, ... Y 12 For the sample, the marginal distribution of average runoff for each month was obtained using the Gaussian kernel density estimation method, and the probability density function of average runoff for each month was plotted. The probability density curves of average runoff for each month during the high-water season (May to October) and the low-water season (November to April) are attached. Figure 4 As shown.

[0081] Clearly, the probability density curves of water inflow in each month are all approximately normal distribution curves. The differences are: the skewness coefficient is larger during the wet season, the monthly average flow distribution range is larger, and the flow value is more dispersed; the skewness coefficient is smaller during the dry season, the monthly average flow distribution range is smaller, and the flow value is more concentrated.

[0082] After obtaining the probability density function of the average monthly flow, it is necessary to further obtain the cumulative distribution function of the average flow. However, the probability density function obtained by Gaussian kernel density estimation is actually a set of points rather than an analytical function expression. Therefore, numerical integration is required to obtain its cumulative distribution function. This invention uses the composite Simpson's rule to obtain the distribution function, that is, to calculate the definite integral through polynomial fitting in the absence of an analytical probability density function. The formula is as follows:

[0083]

[0084] In the formula, x and f(x) are the point sets of the probability density function, a and b are the minimum and maximum values ​​of x, respectively, n is the number of partitions of the interval [a,b], and h=(ba) / n.

[0085] The cumulative distribution function obtained by integrating the composite Simpson's rule is still a set of points, and this set of points is considered as the marginal distribution function of the random variable. To establish the relationship between the random variable X and Y1, Y2, Y3, ... Y using the Copula function... 12 The relationship between them needs to be established by using probability integral transformation to transform the random variables Y1, Y2, Y3, ... Y... 12 The marginal distribution is mapped to a uniform distribution on [0, 1].

[0086] S3. The joint annual and monthly inflow distribution was constructed using the Frank-Copula, Clayton-Copula, and Gumbel-Copula functions, respectively, and the parameters were estimated using the maximum likelihood method. The Spearman rank correlation coefficient, Euclidean distance, and Akaike Information Criterion (AIC) value were calculated to measure the goodness of fit of different Copula functions. The goodness of fit values ​​of different Copula functions were compared to select the final joint annual and monthly inflow distribution.

[0087] The results of fitting the runoff distribution between the average monthly flow and the total annual inflow are shown below. 2. Among the parameters mentioned above, the Spearman rank correlation coefficient measures the correlation between two random variables, with a value range of [-1, 1]. The closer to -1, the stronger the negative correlation between the two variables; the closer to 1, the stronger the positive correlation. Euclidean distance measures the geometric linear distance between two vectors in Euclidean space. Calculating the Euclidean distance between sampled data fitted by the Copula function can verify whether the generated samples have the same distribution pattern as the original data, i.e., whether the fitted Copula function captures the dependency between the two random variables. Generally speaking, a large Euclidean distance indicates that the samples cover too much of the marginal area of ​​the sampling range, meaning that the Copula function overestimates the extreme correlation between the two variables; a small Euclidean distance indicates that the generated samples are clustered in the central area of ​​the sampling range, meaning that the Copula function has not captured the correlation between the two random variables. The Akaike Information Criterion (AIC value) reflects the goodness of fit and complexity of the model. This invention adopts a simple evaluation criterion: the smaller the AIC value, the better the model fit.

[0088] Table 2 Fitting results for three Copula distributions

[0089] As shown in Table 2, except for December, the Spearman rank correlation coefficients for all other months are positive, with higher coefficients from April to October and lower coefficients for the remaining months. This indicates a strong positive correlation between annual and monthly runoff in the basin from April to October, while the correlation is weaker in other months. This phenomenon may be due to the following reasons: during the dry season, precipitation runoff is low, and the flow is mainly composed of baseflow, which is relatively stable and accounts for a small proportion of the total annual inflow, thus lacking a strong correlation with the total annual inflow; during the wet season, the flow is mainly composed of precipitation runoff, accounting for a large proportion of the total annual inflow, and a strong positive correlation exists between the two.

[0090] In the calculation results, except for the Gumbel-Copula function sample with a slightly larger Euclidean distance in August, the Euclidean distances of the other fitted functions are all around 0.5 and are very stable. This indicates that the Copula function and its parameter calibration method and results established in this invention are reasonable and perform stably when dealing with data from different months.

[0091] Based on the AIC values ​​for each month under the three different models in the table, it can be concluded that the correlation between annual total water volume and monthly inflow is more suitable to be established using the Gumbel-Copula function in July and August, while the correlation for other months is more suitable using the Frank-Copula function. Since the Gumbel-Copula function is suitable for correlation studies of random variables with upper tail correlation, and July and August are precisely the months with the highest inflow and the highest probability of extreme situations, this aligns with the hydrological patterns of the watershed.

[0092] Scatter plots of Copula function sampling samples for each month are attached. Figure 5 As shown, because the samples have been mapped to the [0, 1] interval, unlike the cumulative distribution curve before the transformation, the closer the frequency in the figure is to 1, the greater the corresponding water inflow. It can be seen from the figure that the sampling results of the annual total water volume and monthly average flow during the dry season are relatively dispersed, showing no strong correlation, indicating that the increase in the annual total water volume is not significantly related to the monthly water inflow during the dry season. As the wet season approaches, the sampling results gradually concentrate near the 45° line, and in July and August, the closer the water inflow frequency is to 1, the more concentrated the samples are near the 45° line, exhibiting a very strong upper tail correlation. This indicates that the large annual total water volume in this basin is mainly due to the increase in the monthly average flow during the wet season, especially the extreme floods occurring in July and August, which have the closest relationship with the annual water frequency.

[0093] S4 utilizes the established joint annual and monthly inflow distribution for conditional simulation. A selected annual inflow frequency is chosen, and a Copula-based method for generating annual and monthly inflows is employed to estimate the monthly runoff at the specified frequency, yielding the total monthly inflow. Monthly runoff acquisition is divided into dry and wet seasons. Since the average flow is similar in dry season months, the monthly average flow can be directly used as the dry season flow. Inflows vary significantly in wet season months, requiring separate processing for different flood sizes.

[0094] From Table 2 and appendix Figure 5 The conclusion is that the monthly average flow from May to October is strongly correlated with the annual runoff. The established Copula function can be used to estimate the expected flow for each month at a specified flow frequency. The flow from November to April is relatively close and does not fluctuate much. The expected flow from May to October can be subtracted from the annual runoff and then allocated monthly with reference to the multi-year monthly average flow.

[0095] Three commonly used frequencies (25%, 50%, and 75%) were selected for estimation, and the monthly water inflow data are shown in Table 3. Figure 6As can be seen from the figure, the average flow rate is relatively low from November to April of the following year, and the flow rate is similar under different water inflow frequencies; the water inflow varies greatly from May to October, and the flow rate difference between July and August is the largest under different water inflow frequencies. The flow rate in a wet year is about twice that in a dry year, which is the most important factor affecting the annual runoff.

[0096] Table 3. Estimated monthly average flow rates for the three frequencies.

[0097] The aforementioned method not only established a way to estimate the average monthly flow under different inflow frequencies, but also concluded that the average flow is similar from November to April, while the inflow during the wet season from May to October differs significantly. Therefore, the monthly average flow can be directly used as the flow during the dry season. The focus of inflow sequence generation below will be on parts 5 to 10. Since the main reason for the increased inflow during the wet season is precipitation runoff, the superposition of multiple precipitation runoffs constitutes the river inflow during the flood season. By superimposing multiple floods caused by precipitation in staggered order of occurrence, the inflow process curve can be generated from the perspective of causes.

[0098] S5. For major floods with a return period of more than once in ten years during the high-water season, the same-scale runoff process within the month is obtained by scaling down the calculation using the same-scale method. That is, the scaling-up ratio is determined by comparing the design flood volume with the typical flood volume, and then the typical flood process is scaled up to obtain flood processes of different frequencies. Large floods at frequencies of 10%, 5%, 3.3%, 2%, and 1% can be calculated to obtain the corresponding flood processes.

[0099] The design peak discharge and design flood volume of the basin are shown in Table 4, and the duration and composition of a typical 10-year flood are shown in Table 5. For rivers with a good peak-volume relationship, the same-ratio method is simple and yields stable results for flood scaling calculations. The same-ratio scaling method is specifically divided into two types: "peak ratio" scaling and "volume ratio" scaling. Since this invention allocates floods under the premise of a determined flood volume, the "volume ratio" scaling method is selected, and the scaling factor is:

[0100]

[0101] W p To ensure loudness at a design frequency of p, W d This represents the flood volume during a typical flood event.

[0102] First, the amplification ratio is determined by comparing the design flood volume with the typical flood volume. Then, by amplifying the typical flood process, flood processes of different frequencies can be obtained. To facilitate subsequent research, large-scale floods at frequencies of 10%, 5%, 3.3%, 2%, and 1% were conducted, and the resulting flood processes are shown in the attached figure. Figure 7 As shown.

[0103] Table 4. Design Peak Flow and Design Flood Volume of the Basin

[0104] Table 5. Duration and Flood Volume Composition of a Typical 10-Year Flood

[0105] S6. For small floods with a return period of less than once in ten years during the high-water season, since it is impossible to obtain a flood process that matches the actual situation by reducing the typical design flood process, it is necessary to identify and extract features of small floods. Therefore, the second-order difference identification algorithm is used to identify the peaks and troughs of small floods in long series of inflow sequences, calculate the duration of the flood, and divide the small floods into flood waveform families with different durations through cluster analysis.

[0106] Small-scale floods are typically caused by short periods of rainfall and recede rapidly after the rainfall stops. Their duration is much shorter than that of floods with a return period of more than 10 years, making it impossible to obtain a flood process that reflects reality by narrowing down a typical design flood process. Therefore, this section starts with historical runoff processes and uses a second-order difference identification algorithm to identify the peaks and troughs of small-scale floods in long-series inflow sequences, and calculates the duration of the floods.

[0107] After identifying all small-scale floods in a long series of runoff processes, K-means clustering is used to extract flood features. The duration of the floods at the cluster centers is extended outward or shortened inward to an integer multiple of 24 hours, and the flood volume is allocated to the daily-scale flow process lines of the small-scale floods for each day. In this invention, K=5 is used, which yields small-scale flood flow processes with durations of 3 days, 4 days, 5 days, 6 days, and 7 days. The flood flow for each duration is shown in Table 6.

[0108] Table 6 Duration and Volume Distribution of Small-Scale Floods

[0109] S7 uses digital filtering to divide the baseflow of the dry season flow data of the past ten years, and obtains the original baseflow of the river channel by weighted averaging.

[0110] The monthly runoff process obtained by the staggered superposition of multiple floods will result in the repeated accumulation of baseflow from multiple floods. Assuming that the original baseflow is generated by groundwater recharge during the dry season or in the absence of precipitation, the baseflow will be altered as a direct increase in runoff during the flood season. A schematic diagram is attached. Figure 8As shown: a represents the baseflow during the dry season or under no-precipitation conditions, b represents the baseflow during a flood event, and c represents the hydrograph of the flood event. The area between a and b represents the change in baseflow caused by precipitation, and the area between b and c represents the increase in runoff directly caused by precipitation. To avoid double counting or omissions, the original baseflow component in the flood event must be removed before superposition to obtain the direct runoff generated by precipitation and the change in baseflow caused by precipitation, i.e., the area between a and c, which serves as the superimposed quantity for the flood event.

[0111] Therefore, the baseflow segmentation of this invention differs in purpose from common baseflow segmentation. The purpose of baseflow segmentation in this invention is to obtain the impact of precipitation on river flow by subtracting the original baseflow from the flood event. Therefore, baseflow segmentation is performed using runoff data from the dry season, which is used as the original baseflow under the condition of no precipitation during the wet season.

[0112] Digital filtering is a widely used method in signal processing. It uses mathematical techniques to process a composite signal composed of signals of different frequencies, removing low-frequency or high-frequency components to obtain the desired information. In this invention, the runoff process is the composite signal. The baseline flow can be considered the low-frequency component of the signal, characterized by gradual changes and a long period, while precipitation runoff is considered the high-frequency component, exhibiting larger amplitudes, occurring occasionally, and having a short period. The most basic digital filtering method is the Lyne-Hollick filter, which eliminates phase shift through two recursive steps: forward and reverse.

[0113]

[0114] In the formula, α is the filtering parameter, with a value range of [0.9, 0.99]; Q b (i) represents the fundamental current in time period i, Q sum (i) represents the total runoff during time period i.

Claims

1. A method for generating multi-scale inflow sequences considering runoff variation characteristics, characterized in that, Includes the following steps: S1. Establish the frequency distribution of annual water volume and obtain the marginal distribution of annual water volume; estimate the marginal distribution of monthly water volume and map the marginal distribution of each month to a uniform distribution; construct the joint distribution between annual water volume and monthly water volume. S2. Evaluate the goodness of fit of different joint distribution models and select the optimal joint distribution model; based on the selected joint distribution model, generate the annual monthly runoff according to the specified inflow frequency. S3. For the high-water season, process large and small floods separately to generate daily-scale runoff processes; identify and extract the characteristics of small-scale floods and classify them into flood waveform families of different durations; classify the baseflow and obtain the original baseflow of the river channel; subtract the original baseflow from the flood hydrograph to obtain the superposition amount of each flood event; S4. Using the month as a window, by adjusting the occurrence time and interval of floods, multiple floods are superimposed to generate the monthly inflow process line; the monthly window is slidable and the incompletely calculated flood process is inherited to complete the superposition of the entire inflow sequence during the high water season, thus obtaining the assumed inflow process. S5. Assess the morphological similarity between the assumed inflow process and the historical runoff process of the same period, and select the optimal matching scheme; allocate the monthly flow according to the optimal matching scheme, and shorten the inflow forecast step of the key months of the flood season to the daily scale. S6. Combine the monthly dry season flow with the daily wet season flow to obtain the annual water series at the monthly-daily mixed scale.

2. The method for generating multi-scale inflow sequences considering runoff variation characteristics according to claim 1, characterized in that, In step S1, the distribution of water inflow in each month is estimated using the kernel density estimation method, and its cumulative distribution function is obtained through numerical integration.

3. The method for generating multi-scale inflow sequences considering runoff variation characteristics according to claim 1, characterized in that, In step S3, the differential recognition algorithm is used to identify the peaks and troughs of small-scale floods, and the cluster analysis method is used to divide the small-scale floods into flood waveform families with different durations.

4. The method for generating multi-scale inflow sequences considering runoff variation characteristics according to claim 3, characterized in that, The cluster analysis is as follows: the flood events are initially divided into multiple clusters according to their duration; the cluster centers are updated through iterative calculation until the cluster centers are stable; multiple cluster centers are output, and their durations are adjusted to integer multiples, and the flood volume is allocated to obtain the daily-scale flow process line.

5. The method for generating multi-scale inflow sequences considering runoff variation characteristics according to claim 1, characterized in that, In step S3, the digital filtering method is used to divide the dry season flow data into baseflows to obtain the original baseflow of the river channel.

6. The method for generating multi-scale inflow sequences considering runoff variation characteristics according to claim 5, characterized in that, Digital filtering treats the runoff process as a composite signal, with the baseflow being the low-frequency component and precipitation runoff being the high-frequency component, and separates the baseflow through filtering.

7. The method for generating multi-scale inflow sequences considering runoff variation characteristics according to claim 1, characterized in that, In step S4, different monthly runoff processes are generated by adjusting the number, frequency combination, or peak location of flood events.

8. The method for generating multi-scale inflow sequences considering runoff variation characteristics according to claim 1, characterized in that, In step S4, the dynamic time warping method is used to calculate the morphological similarity between the assumed inflow process and the historical runoff process of the same period, with the minimum path cumulative distance as the criterion.

9. A multi-scale inflow sequence generation system considering runoff variation characteristics, characterized in that, include: The construction module is used to establish the frequency distribution of annual total water volume, obtain the marginal distribution of annual total water volume, estimate the marginal distribution of monthly water inflow, and map the marginal distribution of each month to a uniform distribution; and construct the joint distribution between annual total water volume and monthly water inflow. Evaluation module: Used to evaluate the goodness of fit of different joint distribution models and select the optimal joint distribution model; based on the selected joint distribution model, it generates the annual monthly runoff according to the specified inflow frequency; Overlay module: Used to process major and minor floods separately during the high-water season to generate daily-scale runoff processes; Identify and extract the characteristics of small-scale floods, classify them into flood waveform families with different durations; classify the baseflow and obtain the original baseflow of the river channel; subtract the original baseflow from the flood hydrograph to obtain the superposition amount of each flood event; Water Inflow Process Module: This module uses a monthly window to generate monthly water inflow process lines by superimposing multiple flood events by adjusting the occurrence time and interval of the floods; it slides the monthly window and inherits the flood processes that are not fully calculated to complete the superposition of the entire water inflow sequence during the high-water season and obtain the assumed water inflow process. Optimal module: used to evaluate the morphological similarity between the assumed inflow process and the historical runoff process of the same period, select the optimal matching scheme; allocate the monthly flow according to the optimal matching scheme, and shorten the inflow forecast step of the key months of the flood season to the daily scale; Combination module: Used to combine monthly dry season flow with daily wet season flow to obtain a monthly-daily mixed annual water series.

10. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the program to implement the steps of the method according to claims 1-8.

Images