Method for predicting ice flood in Ningxia section of Yellow River

Abstract

A Yellow River Ningxia section ice flood prediction method comprises the following steps: S0, collecting daily weather, highest temperature, lowest temperature and daily average runoff information in a monitored river reach in a specific historical period, and constructing a data set by taking the weather, the highest temperature and the lowest temperature as input and the daily average runoff as output; s1, constructing an ice flood prediction model by adopting the data set and taking a support vector regression model and a radial basis function network model as basic models; s2, on the basis of the ice flood prediction model, a corresponding change rule between the daily average runoff volume and the weather, the highest air temperature and the lowest air temperature is obtained; s3, collecting daily weather, highest temperature and lowest temperature information in a future prediction period in the monitored river reach; s4, inputting the weather and temperature information in the S3 into an ice flood prediction model, and obtaining daily average runoff information of the monitored river reach in a future prediction time period; and S5, predicting the ice flood level of the monitored river reach in the future prediction time period.

Description

Technical Field

[0001] The present invention relates to the technical field of flood monitoring, and in particular to a method for predicting ice floods in the Ningxia section of the Yellow River. Background Art

[0002] The Ningxia section of the Yellow River is geographically unique, flowing from south to north. Because temperatures decrease with increasing latitude, the lower reaches of the Ningxia section freeze over before the upper reaches in winter. When upstream water continuously flows downstream, while the lower reaches are already frozen but not yet open, the limited space beneath the ice can no longer accommodate the incoming water, causing the river to stagnate and surge, leading to a sharp rise in water levels. In spring, the situation is reversed. The lower reaches of the Ningxia section open later than the upper reaches. The resulting ice peaks from the upper reaches surge downstream, carrying icy debris with them. As they flow through the still-opened lower reaches, they are easily blocked by blocked channels. Particularly in narrow channels and bends, unmelted ice can easily accumulate and block the river, forming ice dams and ice jams, further raising water levels. Once water levels rise sufficiently, an ice flood occurs, submerging riverbanks and overflowing into surrounding areas, posing a serious threat to the safety of people living along the banks.

[0003] In recent years, ice floods have frequently occurred in the Ningxia section of the Yellow River, each time causing huge losses to the lives and production of residents along the river. Therefore, ice flood warning work in the Ningxia section of the Yellow River is particularly important to help relevant departments know the ice flood information in advance, take response measures as soon as possible, protect people’s lives and property, and reduce people’s losses.

[0004] Among the existing ice flood warning technologies, the most advanced is to obtain video image information of ice in the river channel through remote sensing technology, and then make judgments and warnings on ice floods; because ice flood warnings based on video images of ice can only be made when ice has formed and can be recognized by images, the flood season is approaching at this time. Although ice flood warnings can be issued, the time left for people to prepare is still short. Especially when dealing with larger-scale ice floods, people still need earlier warnings to have more time to prepare for the flood situation. Summary of the Invention

[0005] In view of this, it is necessary to provide an ice flood prediction method for the Ningxia section of the Yellow River that can provide earlier warning of ice conditions, so as to ensure that people have more sufficient preparation time and improve the monitoring level of ice prevention work.

[0006] A method for predicting ice floods in the Ningxia section of the Yellow River comprises the following steps:

[0007] S0. Collect daily weather, maximum temperature, minimum temperature, and average daily runoff information within the monitored river section during a specific historical period, and construct a dataset using weather, maximum temperature, minimum temperature as input and average daily runoff as output;

[0008] S1. Using the above dataset, an ice flood prediction model is constructed based on the support vector regression model and radial basis function network model;

[0009] S2. Based on the ice flood prediction model, obtain the corresponding change pattern between daily average runoff and weather, maximum temperature and minimum temperature;

[0010] S3. Collect daily weather information and maximum and minimum temperature information within the future forecast period within the monitored river section;

[0011] S4, inputting the weather and maximum and minimum temperature information in S3 into the ice flood prediction model to obtain the average daily runoff information of the monitored river section during the future prediction period;

[0012] S5. Based on the average daily runoff information of the monitored river section during the future forecast period obtained in S4, and based on the corresponding relationship between the average daily runoff and the ice flood level, a forecast is made on the ice flood level of the monitored river section during the future forecast period.

[0013] Preferably, in step S1, the ice flood prediction model is constructed based on the support vector regression model and the radial basis function network model, which specifically includes the following steps:

[0014] S10, divide the data set in S0 into training set and test set;

[0015] S11, preset the key external parameter penalty parameter and γ value, and use the training set to train the support vector regression model f SVR (x) train to obtain the internal parameters that match the preset parameters and the corresponding trained support vector regression model f′ SVR (x); preset the key external parameter γ′ value, and use the training set to train the radial basis function network model f RBFN (x) Perform training to obtain internal parameters that match the preset parameters and the corresponding trained radial basis function network model f′ RBFN (x);

[0016] S12, the trained support vector regression model f′ SVR (x) and the trained radial basis function network model f′ RBFN (x) is weighted and combined to obtain the basic model f(x) for ice flood prediction, as shown in Formula 1, where q1 and q2 are f′ respectively. SVR (x) and f′ RBFN The weighting coefficient of (x), the sum of q1 and q2 is 1;

[0017] f(x)=q1f′ SVR (x)+q2f′ RBFN (x) Formula (1)

[0018] S13, preset weighting coefficients q1 and q2, and use the test set to test the ice flood prediction basic model f(x) to obtain the test accuracy;

[0019] S14, using the root mean square error (RMSE) and the mean absolute error (MAE) as evaluation criteria, setting the accuracy boundary, and comparing the test accuracy with the accuracy boundary. If the test accuracy is lower than the accuracy boundary, returning to S11 and re-executing; if the test accuracy is higher than the accuracy boundary, continuing to S15;

[0020] S15, setting the corresponding key external parameter penalty parameter, γ value, and γ′ value to the optimal penalty parameter, optimal γ value, and optimal γ′ value, setting the weighting coefficients q1 and q2 to the optimal weighting coefficients q′1 and q′2, and assigning them to the ice flood prediction basic model f(x);

[0021] S16. Use the data set to train the ice flood prediction basic model f(x) obtained in S15 to obtain the ice flood prediction model f′(x), as shown in Formula 2.

[0022] f′(x)=q′1f′ SVR (x)+q′2f′ RBFN (x) Formula (2)

[0023] Preferably, in step S11, the preset key external parameter penalty parameter and γ value are used to train the support vector regression model f SVR (x) train to obtain the internal parameters that match the preset parameters and the corresponding trained support vector regression model f′ SVR (x), specifically:

[0024] Input the input data and output data of the training set into f SVR (x), as shown in Formula 3, where x and f SVR (x) corresponds to input data and output data respectively, and the internal parameter ω SVR is the weight vector, b SVR is the bias term of SVR, <ω SVR ,x> represents the inner product, and the operation obtains the inner parameter ω SVR and b SVR , and further obtain f′ SVR (x).

[0025] f SVR (x)=<ω SVR ,x>+b SVRFormula (3)

[0026] Preferably, in step S11, the preset key external parameter γ′ value is used to train the radial basis function network model f RBFN (x) Perform training to obtain internal parameters that match the preset parameters and the corresponding trained radial basis function network model f′ RBFN (x), specifically:

[0027] Input the input data and output data of the training set into f RBFN (x), as shown in formula (4), where x and f RBFN (x) corresponds to input data and output data respectively, m is the number of neurons in the hidden layer, c j is the center of the j-th Gaussian kernel, σ j is the width parameter, ω jk is the weight from the hidden layer to the output layer, b RBFN is the bias term of RBFN; the internal parameters m and c are obtained by operation j , σ j 、ω jk and b RBFN , and further obtain f′ RBFN (x).

[0028]

[0029] Preferably, in step S14, the evaluation criteria of the root mean square error RMSE and the mean absolute error MAE are shown in formula (5) and formula (6), respectively, where i is the data sequence number of the test set, p is i The output data obtained when the test set input data is input into the ice flood prediction basic model. is the original output data of the test set.

[0030]

[0031] Preferably, in step S4, the correspondence between the daily average runoff and the ice flood level is:

[0032] If the average daily runoff is less than 500m 3 / s, the ice flood level is level IV, which is a mild ice flood;

[0033] If the average daily runoff is between 500m 3 / s to 1500m 3 / s, the ice flood level is level III, which is a moderate ice flood;

[0034] If the average daily runoff is between 1500m 3 / s to 3000m 3 / s, the ice flood level is Level II, which is a severe ice flood;

[0035] If the average daily runoff is higher than 3000m 3 / s, the ice flood grade is Level I, which is an extremely severe ice flood.

[0036] Preferably, when collecting the daily weather data in the monitored river section during a specific historical period to construct a data set, and when collecting the daily weather data in the monitored river section during a future forecast period as input data, a correspondence is set between the weather and the digital code; the correspondence is: sunny corresponds to 0, cloudy corresponds to 1, overcast corresponds to 2, light rain corresponds to 3, moderate rain corresponds to 4, heavy rain corresponds to 5, thunderstorm corresponds to 6, light snow corresponds to 7, moderate snow corresponds to 8, and heavy snow corresponds to 9.

[0037] Preferably, the collected weather, maximum temperature, minimum temperature and average daily runoff are cleaned before constructing the data set.

[0038] The above-mentioned ice flood prediction method for the Ningxia section of the Yellow River establishes an ice flood prediction model based on a support vector regression model and a radial basis function network model, and uses weather, maximum temperature, minimum temperature and daily average runoff data in a specific historical period of the monitored river section to train the ice flood prediction model, thereby obtaining a variation pattern between the daily average runoff and the weather, maximum temperature and minimum temperature based on the specific historical period. Based on this pattern, the daily average runoff information of the monitored river section in the future prediction period is obtained on the basis of obtaining the weather, maximum temperature and minimum temperature in the future prediction period. Further, based on the correspondence between the daily average runoff information and the ice flood level, the ice flood level in the monitoring period in the future prediction period is predicted. Compared with the prior art technology of early warning of ice floods based on video images of ice and icicles, the ice flood prediction model of the present invention, which is based on historical data, can predict ice conditions earlier when ice and icicles have not yet formed and the flood season has not yet approached, and can also analyze ice condition trends in the long term, providing more sufficient time for relevant departments to prepare for ice prevention work and improving the monitoring level of ice prevention work. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The figure is a schematic diagram of the overall process of the ice flood prediction method for the Ningxia section of the Yellow River in the present invention.

[0040] Figure 2 This is a schematic diagram of the construction process of the ice flood prediction model in the present invention.

[0041] Figure 3 This is the support vector regression model fitting effect diagram when the penalty parameter is 10000 and the γ value is 50 in the present invention.

[0042] Figure 4 This is a diagram showing the fitting effect of the radial basis function network model when the γ′ value is 5 in the present invention.

[0043] Figure 5 This is a fitting effect diagram of the ice flood prediction model when q1 and q2 are both 0.5 in the present invention. DETAILED DESCRIPTION

[0044] The technical solutions and technical effects of the embodiments of the present invention are further elaborated below in conjunction with the accompanying drawings of the present invention.

[0045] In a specific embodiment of the present invention, the Yellow River section from Xiaheyan to Shizuishan in Ningxia is used as the monitoring section, and a data set is constructed based on the weather, maximum temperature, minimum temperature and average daily runoff information for a specific historical period of three years from 2021 to 2024;

[0046] Please see Figure 1 A method for predicting ice floods in the Ningxia section of the Yellow River comprises the following steps:

[0047] S0. Collect daily weather, maximum temperature, minimum temperature, and average daily runoff information for the Yellow River from Xiaheyan, Ningxia to Shizuishan from 2021 to 2024, and construct a data set with weather, maximum temperature, and minimum temperature as input and average daily runoff as output; set a correspondence between weather and digital codes, and the correspondence is: sunny corresponds to 0, cloudy corresponds to 1, overcast corresponds to 2, light rain corresponds to 3, moderate rain corresponds to 4, heavy rain corresponds to 5, thunderstorm corresponds to 6, light snow corresponds to 7, moderate snow corresponds to 8, and heavy snow corresponds to 9; Table 1 shows some of the data;

[0048] S1. Using the above dataset, a flood prediction model is constructed based on the support vector regression model and radial basis function network model. Please refer to Figure 2 As shown:

[0049] S10, divide the data set in S0 into a training set and a test set, where 2 / 3 of the data set is converted into a training set and the remaining 1 / 3 is converted into a test set;

[0050] S11, preset the key external parameter penalty parameter and γ value, and use the training set to train the support vector regression model f SVR (x) train to obtain the internal parameters that match the preset parameters and the corresponding trained support vector regression model f′ SVR (x); preset the key external parameter γ′ value, and use the training set to train the radial basis function network model f RBFN (x) Perform training to obtain internal parameters that match the preset parameters and the corresponding trained radial basis function network model f′ RBFN (x);

[0051] S12, the trained support vector regression model f′ SVR(x) and the trained radial basis function network model f′ RBFN (x) is weighted and combined to obtain the basic model f(x) for ice flood prediction, as shown in Formula 1, where q1 and q2 are f′ respectively. SVR (x) and f′ RBFN The weighting coefficient of (x), the sum of q1 and q2 is 1;

[0052] f(x)=q1f′ SVR (x)+q2f′ RBFN (x) Formula (1)

[0053] S13, preset weighting coefficients q1 and q2, and use the test set to test the ice flood prediction basic model f(x) to obtain the test accuracy;

[0054] S14, using the root mean square error (RMSE) and the mean absolute error (MAE) as evaluation criteria, setting the accuracy boundary, and comparing the test accuracy with the accuracy boundary. If the test accuracy is lower than the accuracy boundary, returning to S11 and re-executing; if the test accuracy is higher than the accuracy boundary, continuing to S15;

[0055] In the above stage, after repeated adjustments and evaluations, it was determined that the key external parameter penalty parameter of the support vector regression model was 10000, the γ value was 50, the key external parameter γ′ of the radial basis function network model was 5, and the weighting coefficients q1 and q2 were 0.5 and 0.5 respectively. When the test accuracy was higher than the accuracy boundary;

[0056] S15, setting the corresponding key external parameter penalty parameter, γ value, and γ′ value to the optimal penalty parameter, optimal γ value, and optimal γ′ value, setting the weighting coefficients q1 and q2 to the optimal weighting coefficients q′1 and q′2, and assigning them to the ice flood prediction basic model f(x);

[0057] That is, in the ice flood prediction model, the optimal penalty parameter, optimal γ value and optimal γ′ value are 10000, 50 and 5 respectively; the optimal weighting coefficients q′1 and q′2 are 0.5 and 0.5;

[0058] S16. Use the data set to train the ice flood prediction basic model f(x) obtained in S15 to obtain the ice flood prediction model f′(x), as shown in Formula 2.

[0059] f′(x)=0.5f′ SVR (x)+0.5f′ RBFN (x) Formula (2)

[0060] S2. Based on the ice flood prediction model, obtain the corresponding change pattern between daily average runoff and weather, maximum temperature and minimum temperature;

[0061] S3. Collect daily weather information and maximum and minimum temperature information within the future forecast period within the monitored river section;

[0062] S4, inputting the weather and maximum and minimum temperature information in S3 into the ice flood prediction model to obtain the average daily runoff information of the monitored river section during the future prediction period;

[0063] S5. Based on the average daily runoff information of the monitored river section during the future forecast period obtained in S4, and based on the corresponding relationship between the average daily runoff and the ice flood level, a forecast is made on the ice flood level of the monitored river section during the future forecast period.

[0064] In this embodiment, in steps S11 to S14, the support vector regression model needs to be adjusted; during the adjustment process, the penalty parameter is finally determined to be 10000. This parameter will improve the prediction accuracy of the model. At the same time, it is also found that as the γ value gradually increases, the test results of the support vector regression model are closer to the accuracy boundary, that is, the model fits the data of the training set better. However, an excessively high γ value will cause the model training cost to rise sharply. Therefore, it is necessary to find a balance between optimizing the fitting effect and controlling the training cost. Experiments have found that when the γ value is set to 50, further increasing the γ value will only increase the model fitting effect to a very limited extent. Therefore, after comprehensive consideration, the γ value is set to 50 and the penalty parameter is set to 10000. Under this parameter setting, the model is tested with the test set data. The fitting effect between the original data of the test set and the predicted data is as shown in the figure. Figure 3 As shown;

[0065] In this embodiment, in steps S11 to S14, the key external parameter γ′ value of the radial basis function network model needs to be repeatedly adjusted and trained; in the process of repeated adjustments, it is found that as the γ′ value gradually increases, the model's fit shows a trend of getting better and better. However, as the γ′ value increases, the model's training cost also increases significantly. Therefore, in the process of model training, it is necessary to reduce the γ′ value as much as possible while ensuring that the model has a good fit. Experiments have found that after the γ′ value exceeds 5, even if it continues to increase, the change in the model's fit is no longer significant, so the γ′ value is finally set to 5; under this parameter setting, the model is tested with the test set data, and the fitting effect between the original data of the test set and the predicted data is as follows. Figure 4 As shown;

[0066] In this embodiment, in steps S11 to S14, the weighting coefficients q1 and q2 need to be repeatedly adjusted and trained to improve the fit of the ice flood prediction model. Experiments have found that when q1 and q2 are both 0.5, the fit of the ice flood prediction model is the best, as shown in Figure 5.

[0067] Table 1

[0068]

[0069]

[0070] Furthermore, in step S11, the preset key external parameter penalty parameter and γ value are used to train the support vector regression model f SVR (x) train to obtain the internal parameters that match the preset parameters and the corresponding trained support vector regression model f′ SVR (x), specifically:

[0071] Input the input data and output data of the training set into f SVR (x), as shown in Formula 3, where x and f SVR (x) corresponds to input data and output data respectively, and the internal parameter ω SVR is the weight vector, b SVR is the bias term of SVR, <ω SVR ,x> represents the inner product, and the operation obtains the inner parameter ω SVR and b SVR , and further obtain f′ SVR (x).

[0072] f SVR (x)=<ω SVR ,x>+b SVR Formula (3)

[0073] Furthermore, in step S11, the preset key external parameter γ′ value is used to train the radial basis function network model f RBFN (x) Perform training to obtain internal parameters that match the preset parameters and the corresponding trained radial basis function network model f′ RBFN (x), specifically:

[0074] Input the input data and output data of the training set into f RBFN (x), as shown in formula (4), where x and f RBFN (x) corresponds to input data and output data respectively, m is the number of neurons in the hidden layer, c j is the center of the j-th Gaussian kernel, σ j is the width parameter, ω jk is the weight from the hidden layer to the output layer, b RBFN is the bias term of RBFN; the internal parameters m and c are obtained by operation j , σ j 、ω jk and b RBFN , and further obtain f′ RBFN (x).

[0075]

[0076] Furthermore, in step S14, the evaluation criteria of the root mean square error RMSE and the mean absolute error MAE are shown in formula (5) and formula (6), respectively, where i is the data sequence number of the test set, p is i is the output data obtained when the input data of the test set is input into the basic model of ice flood prediction, p i * is the original output data of the test set.

[0077]

[0078] Furthermore, in step S4, the corresponding relationship between the daily average runoff and the ice flood level is:

[0079] If the average daily runoff is less than 500m 3 / s, the ice flood level is level IV, which is a mild ice flood;

[0080] If the average daily runoff is between 500m 3 / s to 1500m 3 / s, the ice flood level is level III, which is a moderate ice flood;

[0081] If the average daily runoff is between 1500m 3 / s to 3000m 3 / s, the ice flood level is Level II, which is a severe ice flood;

[0082] If the average daily runoff is higher than 3000m 3 / s, the ice flood grade is Level I, which is an extremely severe ice flood.

[0083] Furthermore, when collecting the daily weather data in the monitored river section during a specific historical period to construct a data set, and when collecting the daily weather data in the monitored river section during a future forecast period as input data, a correspondence is set between the weather and the digital code; the correspondence is: sunny corresponds to 0, cloudy corresponds to 1, overcast corresponds to 2, light rain corresponds to 3, moderate rain corresponds to 4, heavy rain corresponds to 5, thunderstorm corresponds to 6, light snow corresponds to 7, moderate snow corresponds to 8, and heavy snow corresponds to 9.

[0084] Furthermore, the collected weather, maximum temperature, minimum temperature and daily runoff were cleaned before constructing the dataset.

[0085] In a specific embodiment of the present invention, an ice flood prediction model is used to predict the relationship between weather, maximum temperature, minimum temperature and average daily runoff, and the performance of the prediction results is evaluated using the root mean square error (RMSE) and the mean absolute error (MAE).

[0086] At the same time, the data set in the specific embodiment of the present invention is used to train the support vector regression model and the radial basis function network model respectively, and the trained support vector regression model and the trained radial basis function network model are used to predict the relationship between weather, maximum temperature, minimum temperature and daily average runoff, and the prediction result performance is evaluated using RMSE and MAE. The evaluation results are shown in Table 2:

[0087] Table 2

[0088]

[0089] The root mean square error (RMSE) indicates how well the model controls the deviation between the predicted value and the true value. The smaller the RMSE, the better the model controls the deviation between the predicted value and the true value, and the more accurately the prediction results can approach the true value. The mean absolute error (MAE) reflects the average absolute deviation between the predicted value and the true value. A smaller MAE means that the model has a lower average level of overall prediction error and more stable and reliable prediction results.

[0090] As can be seen from Table 2, the root mean square error (RMSE) of the ice flood prediction model is 93.91, and the root mean square error (RMSE) of the support vector regression model is 93.19, which is significantly lower than the 108.52 of the radial basis function network model. This shows that the ice flood prediction model and the support vector regression model perform better in controlling the degree of deviation between the predicted value and the true value, and are more accurate in approaching the true value.

[0091] Although the root mean square error (RMSE) of the ice flood prediction model is similar to that of the support vector regression model, the mean absolute error (MAE) of the ice flood prediction model is only 39.39, which is lower than the 41.50 of the support vector regression model. This indicates that the ice flood prediction model has a lower average level of overall prediction error and more stable and reliable prediction results. Therefore, the ice flood prediction model of the present invention has a higher reference value in practical applications.

[0092] The above-mentioned ice flood prediction method for the Ningxia section of the Yellow River establishes an ice flood prediction model based on a support vector regression model and a radial basis function network model, and uses weather, maximum temperature, minimum temperature and daily average runoff data in a specific historical period of the monitored river section to train the ice flood prediction model, thereby obtaining a variation pattern between the daily average runoff and the weather, maximum temperature and minimum temperature based on the specific historical period. Based on this pattern, the daily average runoff information of the monitored river section in the future prediction period is obtained on the basis of obtaining the weather, maximum temperature and minimum temperature in the future prediction period. Further, based on the correspondence between the daily average runoff information and the ice flood level, the ice flood level in the monitoring period in the future prediction period is predicted. Compared with the prior art technology of early warning of ice floods based on video images of ice and icicles, the ice flood prediction model of the present invention, which is based on historical data, can predict ice conditions earlier when ice and icicles have not yet formed and the flood season has not yet approached, and can also analyze ice condition trends in the long term, providing more sufficient time for relevant departments to prepare for ice prevention work and improving the monitoring level of ice prevention work.

[0093] The above disclosure is only a preferred embodiment of the present invention, and it is certainly not intended to limit the scope of the present invention. A person skilled in the art can understand that all or part of the processes of the above embodiment and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for predicting ice floods in the Ningxia section of the Yellow River, characterized in that: The following steps are involved: S0. Collect daily weather, maximum temperature, minimum temperature, and average daily runoff information within the monitored river section during a specific historical period, and construct a dataset using weather, maximum temperature, minimum temperature as input and average daily runoff as output; S1. Using the above dataset, an ice flood prediction model is constructed based on the support vector regression model and radial basis function network model; S2. Based on the ice flood prediction model, obtain the corresponding change pattern between daily average runoff and weather, maximum temperature and minimum temperature; S3. Collect daily weather information and maximum and minimum temperature information within the future forecast period within the monitored river section; S4, inputting the weather and temperature information in S3 into the ice flood prediction model to obtain the average daily runoff information of the monitored river section during the future prediction period; S5. Based on the average daily runoff information of the monitored river section during the future forecast period obtained in S4, and based on the corresponding relationship between the average daily runoff and the ice flood level, a forecast is made on the ice flood level of the monitored river section during the future forecast period.

2. The method for predicting ice floods in the Ningxia section of the Yellow River according to claim 1, characterized in that: In step S1, a flood prediction model is constructed based on a support vector regression model and a radial basis function network model, specifically comprising the following steps: S10, divide the data set in S0 into training set and test set; S11, preset the key external parameter penalty parameter and γ value, and use the training set to train the support vector regression model f SVR (x) train to obtain the internal parameters that match the preset parameters and the corresponding trained support vector regression model f′ SVR (x); preset the key external parameter γ′ value, and use the training set to train the radial basis function network model f RBFN (x) Perform training to obtain internal parameters that match the preset parameters and the corresponding trained radial basis function network model f′ RBFN (x); S12, the trained support vector regression model f′ SVR (x) and the trained radial basis function network model f′ RBFN (x) is weighted and combined to obtain the basic model f(x) for ice flood prediction, as shown in Formula 1, where q1 and q2 are f′ respectively. SVR (x) and f′ RBFN The weighting coefficient of (x), the sum of q1 and q2 is 1; f(x)=q1f′ SVR (x)+q2f′ RBFN (x) Formula (1) S13, preset weighting coefficients q1 and q2, and use the test set to test the ice flood prediction basic model f(x) to obtain the test accuracy; S14, using the root mean square error (RMSE) and the mean absolute error (MAE) as evaluation criteria, setting the accuracy boundary, and comparing the test accuracy with the accuracy boundary. If the test accuracy is lower than the accuracy boundary, returning to S11 and re-executing; if the test accuracy is higher than the accuracy boundary, continuing to S15; S15, the corresponding key external parameter penalty parameter, γ value, γ ′ The values are set to the optimal penalty parameter, the optimal γ value and the optimal γ ′ value, weighting coefficients q1 and q2 are set to the optimal weighting coefficient q ′ 1 and q ′ 2. And give the ice flood prediction basic model f(x); S16, use the data set to train the ice flood prediction basic model f(x) obtained in S15 to obtain the ice flood prediction model f ′ (x), as shown in Formula 2. f′(x)=q′1f′ SVR (x)+q′2f′ RBFN (x) Formula (2).

3. The method for predicting ice floods in the Ningxia section of the Yellow River according to claim 2, wherein: In step S11, the preset key external parameter penalty parameter and γ value are used to train the support vector regression model f SVR (x) train to obtain the internal parameters that match the preset parameters and the corresponding trained support vector regression model f′ SVR (x), specifically: Input the input data and output data of the training set into f SVR (x), as shown in Formula 3, where x and f SVR (x) corresponds to input data and output data respectively, and the internal parameter ω SVR is the weight vector, b SVR is the bias term of SVR, <ω SVR ,x> represents the inner product, and the operation obtains the inner parameter ω SVR and b SVR , and further obtain f′ SVR (x). f SVR (x)=<ω SVR ,x>+b SVR Formula (3).

4. The method for predicting ice floods in the Ningxia section of the Yellow River according to claim 1, wherein: In step S11, the preset key external parameter γ ′ Value, using the training set to the radial basis function network model f RBFN (x) Perform training to obtain internal parameters that match the preset parameters and the corresponding trained radial basis function network model f′ RBFN (x), specifically: Input the input data and output data of the training set into f RBFN (x), as shown in formula (4), where x and f RBFN (x) corresponds to input data and output data respectively, m is the number of neurons in the hidden layer, c j is the center of the jth Gaussian kernel, σ j is the width parameter, ω jk is the weight from the hidden layer to the output layer, b RBFN is the bias term of RBFN; the internal parameters m and c are obtained by operation j , σ j 、ω jk and b RBFN , and further obtain f′ RBFN (x).

5. The method for predicting ice floods in the Ningxia section of the Yellow River according to claim 2, wherein: In step S14, the evaluation criteria of the root mean square error RMSE and the mean absolute error MAE are shown in formula (5) and formula (6), respectively, where i is the data sequence number of the test set, p is i The output data obtained when the test set input data is input into the ice flood prediction basic model. is the original output data of the test set.

6. The method for predicting ice floods in the Ningxia section of the Yellow River according to claim 1, wherein: In step S4, the corresponding relationship between the daily average runoff and the ice flood level is: If the average daily runoff is less than 500m 3 / s, the ice flood level is level IV, which is a mild ice flood; If the average daily runoff is between 500m 3 / s to 1500m 3 / s, the ice flood level is level III, which is a moderate ice flood; if the average daily runoff is between 1500m 3 / s to 3000m 3 / s, the ice flood level is Level II, which is a severe ice flood; If the average daily runoff is higher than 3000m 3 / s, the ice flood grade is Level I, which is an extremely severe ice flood.

7. The method for predicting ice floods in the Ningxia section of the Yellow River according to claim 1, wherein: When collecting the daily weather data in the monitored river section during a specific historical period to construct a data set, and when collecting the daily weather data in the monitored river section during a future forecast period as input data, a correspondence is set between the weather and the digital code; the correspondence is: sunny corresponds to 0, cloudy corresponds to 1, overcast corresponds to 2, light rain corresponds to 3, moderate rain corresponds to 4, heavy rain corresponds to 5, thunderstorm corresponds to 6, light snow corresponds to 7, moderate snow corresponds to 8, and heavy snow corresponds to 9.

8. The method for predicting ice floods in the Ningxia section of the Yellow River according to claim 1, wherein: The collected weather, maximum temperature, minimum temperature and daily average runoff were cleaned and then the dataset was constructed.

Images