CN113379110B - Medium-and-long-term runoff forecast result trend testing method - Google Patents

Abstract

The invention relates to a self-adaptive drainage basin medium and long term runoff forecasting model architecture method, which aims to divide a forecasting drainage basin into different sub drainage basin subareas for forecasting according to the characteristics of the forecasting drainage basin; establishing forecasting factor sets of different sub-watersheds; adopting an ensemble forecasting method for each sub-basin; adjusting parameters of different models in the forecasting method through a self-adaptive method; adopting a river channel calculation method to obtain a final river basin forecast value for the forecast result of the sub-river basin; the deterministic coefficient of the forecasting result is periodically checked to judge whether the forecasting factor needs to be updated or not and the composition of the forecasting method, and the runoff forecasting result obtained by using the method can provide reliable basis for urban flood control or large-scale reservoir inflow forecasting.

Description

A trend test method for medium and long-term runoff forecast results

technical field

The application of the present invention is the filing date of November 21, 2017, the application number is: 201711163861.1, and the divisional application of the invention patent application entitled "An adaptive model framework method for medium and long-term runoff forecasting in a river basin". The invention relates to a runoff forecasting framework method, in particular to a method for checking the trend of medium and long-term runoff forecasting results.

Background technique

Runoff forecasting belongs to the category of hydrological forecasting and is an important part of applied hydrology. It is an applied science and technology that predicts future runoff changes on the basis of mastering objective hydrological laws. prerequisite for implementation. Runoff forecasting can be divided into short-term runoff forecasting and medium- and long-term runoff forecasting according to the forecast period. Generally, it is bounded by the time of watershed confluence. If the forecasted forecast period is less than or equal to the watershed confluence time, it is called short-term forecast, and the forecast forecast period is greater than the watershed confluence time. known as medium and long-term forecasts. Among them, a forecast period of one day is a medium-term forecast, a forecast period of more than one day is a long-term forecast, and a forecast period of more than one year is a long-term forecast.

Medium- and long-term runoff forecasting is to make qualitative or quantitative predictions of the runoff state in a certain period of time in the future based on known information. First, analyze the rainfall, sea surface temperature, atmospheric circulation system and weather system that affect the forecast object, select and determine the predictors with strong physical mechanism and significant correlation, and use statistical calculation methods to calculate whether the correlation between them is significant. On this basis, the corresponding forecasting model is established, and the optimal forecasting model is evaluated and optimized. Finally, through the simulation and trial report of the model, the accuracy of the model is analyzed, and the uncertainty analysis of the deterministic forecast results is carried out, and the forecast results are given, which can be applied to production practice.

Accurate mid- and long-term runoff forecast is an important guarantee for improving the utilization rate of water resources, realizing the optimal operation of hydropower stations in the basin, and improving the economic benefits of hydropower stations. Especially in the context of the reform of the power market, it is particularly important to improve the accuracy and forecast period of medium and long-term forecasting, and to formulate a scientific and reasonable cascade joint optimization power generation plan in the basin, which is particularly important for the efficient joint dispatch of basin cascade reservoirs.

In recent years, with the continuous construction of large-scale water conservancy projects in our country, it has played a huge role in alleviating the shortage of water resources and improving the ecological environment. The role of these water conservancy projects has been brought into full play, and the economic benefits, social benefits and ecological environmental benefits of these water conservancy projects have been brought into full play, and the realization of the project construction goals has been ensured. How to predict runoff more accurately is the primary issue faced in the operation and management of water resources in various projects, and it is also one of the key issues that determine the success or failure of various projects. However, due to the influence of human activities, climate change and other factors, the current forecasting system urgently needs to be improved in terms of coverage and forecasting factors. Especially under the changing conditions of the underlying surface, the runoff mechanism between various parts of the basin has also changed accordingly, and the simulation effect of the method ignoring local differences has declined. The update forecast model cannot be adjusted according to the forecast results.

SUMMARY OF THE INVENTION

The present invention designs an adaptive model framework method for mid- and long-term runoff forecasting in a watershed. The technical problem it solves is that the current forecasting system urgently needs to be improved in terms of coverage, forecasting factors, etc., and there is no medium-term forecasting and long-term forecasting in the forecasting process. Different forecasting models are not established according to the characteristics of different sub-basins of the watershed. At the same time, the forecasting model itself lacks the ability of self-evaluation and modification, so the model blocks cannot be updated in time.

In order to solve the above-mentioned technical problem, the present invention adopts the following scheme:

1. An adaptive model framework method for medium and long-term runoff forecasting in a watershed, comprising the following steps:

Step 1. Collect the basic data of the forecast watershed;

Step 2. Based on the basic data, use the linear regression method to establish a linear regression equation between the annual flow series x(t) and its time series t, and then test the trend of the time series;

Step 3. Divide the watershed into several sub-basins;

Step 4. Identification of predictors;

Step 5. Establish a prediction model library;

Step 6. Through deterministic coefficient evaluation, in each sub-basin, for the three methods of physical formation method, hydrological statistics method and artificial intelligence, each method selects a model with a higher deterministic coefficient to form an ensemble forecast for the sub-basin. composition;

Step 7, according to the result calculated in step 6, determine the weight value of different methods of physical formation method, hydrological statistics method and artificial intelligence, and carry out ensemble forecast;

Step 8, using the minimum root mean square error algorithm to adjust the weight value in step 7, so that the root mean square error between the predicted value of each sub-basin and the measured value is minimized, and the predicted value of each sub-basin is output;

Step 9. According to the forecast value of each sub-basin, carry out channel calculation to obtain the runoff process of the outlet section of the entire watershed, and complete the forecast process;

Step 10. Calculate the coefficient of certainty for the prediction result of the entire watershed;

Step 11. On a certain day of each month, carry out a trend test on the certainty coefficient of the daily forecast result of the previous year;

Step 12: Determine whether it is necessary to update the forecast factor set and reselect the forecast model.

Further, the basic data in the step 1 includes:

Basic data A, the daily, ten-day, monthly and annual flow data of the main control hydrological station in the watershed, which are used in the step 2 and the step 11;

Basic data B, the ten-day, monthly, annual maximum flow, minimum flow characteristic value and occurrence time of each main control hydrological station, the first runoff and the last runoff process data, which are used in the step 3;

Basic data C, the daily, ten-day, monthly and annual rainfall of the main rainfall stations in the basin, which are used in the step 4;

Basic data D. Collect 74 circulation indicators, as well as the numerical forecast results of the European Center for Medium-Range Weather Forecasts ECMWF or the US National Center for Environmental Prediction NCEP and the meteorological impact factors of the reanalysis data, which are used in the step 4. Further, in the step 2, a linear regression equation is constructed according to the basic data A of the watershed main control hydrological station provided in the step 1, and the linear regression equation gives whether the time series has an increasing or decreasing trend, and is:

x(t)=a×t+b;

where x(t) is the time series, t is the corresponding time series, a is the slope of the linear equation, which represents the average trend change rate of the time series, and b is the intercept; the values of a and b can be estimated by the least squares method.

Further, in the step 3, the criteria for dividing the watershed into several sub-basins are: the flow change trend of the control station determined in step 2, the underlying surface conditions of the upstream sub-basins of each control station and the sub-basins with the same runoff method are merged. It is a sub-basin, and different sub-basins are distinguished from each other. The ten-day, monthly and annual maximum flow, minimum flow characteristic value and occurrence time of each main control hydrological station in the basic data B in the step 1 are used as the underlying surface conditions for judgment. and the conditions of the mode of abortion.

Further, in the step 4, predicting factors are identified for different sub-basins respectively, and the predicting factors include: previous precipitation and runoff, 74 circulation indicators, meteorological factor data, sea surface temperature, solar activity factor and human activity factor; wherein , the solar activity factor selects the relative number of sunspots and the associated geomagnetic index, the solar 10cm wave radio current as the predictor; human activities are reflected by the urban impervious hardened ground area and the scheduling rules of the hydropower station; the meteorological factor data comes from the above steps The numerical prediction results and reanalysis data of the European Center for Medium-Range Weather Forecasts ECMWF or the US National Center for Environmental Prediction NCEP in the basic data D in 1;

The correlation analysis method is used to analyze the degree of correlation between different forecast factors and different sub-basin flows. The calculation formula is as follows:

相关系数；n为资料样本数；X_i为X的第i个样本值；Y_i为Y的第i个样本值；

为X的样本均值；

为Y的样本均值；X代表某一子流域出口断面的流 量，Y代表某一种预报因子，分别计算不同的预报因子和子流域出口断面流量之间的相关关 系；

Correlation coefficient; n is the number of data samples; X _i is the ith sample value of X; Y _i is the ith sample value of Y;

is the sample mean of X;

is the sample mean of Y; X represents the flow at the outlet section of a sub-basin, and Y represents a certain forecast factor, and the correlation between different forecast factors and the flow at the outlet section of the sub-basin is calculated respectively;

The value range of the correlation coefficient R _XY is [-1,1]; if R _XY is greater than 0, it indicates that there is a positive correlation between the forecast object Y and the forecast factor X; if R _XY is less than 0, it indicates that there is a positive correlation between the forecast object Y and the forecast factor X negative correlation; _RXY equal to 0, indicating that there is no correlation between the forecast object Y and the predictor _X ; the greater the absolute value of RXY, the higher the correlation between the forecast object Y and the predictor X; for different subtypes For the watershed, the top 10% of the predictors with the highest degree of correlation were selected as the set of predictors for different sub-basins.

Further, the establishment of the prediction model library in step 5 includes three methods: physical genetic method, hydrological statistical method and artificial intelligence method, physical genetic method includes multiple linear regression model and multiple threshold regression model, hydrological statistical method includes time series decomposition model and Rank similarity prediction model, artificial intelligence model including artificial neural network model and support vector machine model.

Further, the deterministic coefficient formula used in the step 6 is:

为实测序列的均值，m为资料序列 的长度。where DC is the coefficient of certainty, y ₀ (i) is the measured value, y _c (i) is the predicted value,

is the mean value of the measured series, and m is the length of the data series.

Further, the calculation formula in the step 7 is as follows:

水文统计法的权重为

人工智能法 模拟的权重为

则集成预报值为：In step 6, the deterministic coefficients of the physical cause method, hydrological statistics method and artificial intelligence are A, B, and C respectively, then the weight of the simulation result of the physical cause method is

The weight of hydrostatistics is

The weight of artificial intelligence method simulation is

Then the integrated forecast value is:

R=w ₁ y ₁ +w ₂ y ₂ +w ₃ y ₃

In the formula, w ₁ , w ₂ , and w ₃ are the weight values, y ₁ , y ₂ , and y ₃ are the forecast values of each method, and R _i is the ensemble forecast value of each sub-watershed.

Further, on the 1st of each month in the step 11, the trend test is carried out on the certainty coefficient of the daily forecast result of the previous year (12 months), and the test object of the certainty coefficient is the daily forecast result of the basic data A in the step 1 and the main control. For the daily flow data of the hydrological station, the Kandel rank correlation test method is used, and the calculation formula is as follows:

N为确定性系数序列的总长度，x_i，x_j为系列中的数值，sgn为符号函数，返回值如果数字大于0，则sgn返回1，数字等于0，则返回0，数字小于0，则返回-1，数字参数的符号决定了sgn函数的返回值；i，j为系列中数值的编号，从1到n；n为系列的长度；τ为常数。In the formula, U is the coefficient of certainty;

N is the total length of the deterministic coefficient sequence, x _i , x _j are the numerical values in the series, sgn is the sign function, the return value If the number is greater than 0, sgn returns 1, if the number is equal to 0, it returns 0, if the number is less than 0, Returns -1, the sign of the numeric parameter determines the return value of the sgn function; i, j are the numbers of the values in the series, from 1 to n; n is the length of the series; τ is a constant.

Further, in the step 12, for the calculation result of the step 11, if |U|>U _α/2 and U is greater than 0, it means that the change trend of the deterministic coefficient sequence is significant, the deterministic coefficient sequence is on the rise, and the predicted result Better, no need to update the forecast factor set and re-select the forecast model;

When |U|>U _α/2 and U is less than 0, the sequence shows a downward trend, indicating that the forecast results have a downward trend. At this time, go back to step 4 to re-identify the forecast factor set, and repeat steps 6-10;

Among them, α is the significance level, and U _α/2 is obtained through the normal distribution table through the given significance level.

The adaptive watershed medium and long-term runoff forecasting model framework method has the following beneficial effects:

The present invention divides the forecasted watershed into different sub-watershed divisions for forecasting according to the characteristics of the forecasted watershed; establishes forecast factor sets of different sub-watersheds; adopts the method of ensemble forecasting for each sub-watershed; parameters; for the forecast results of the sub-basins, the channel calculation method is used to obtain the final forecast value of the basin; the certainty coefficient of the forecast results is periodically checked to determine whether the forecast factors and forecast methods need to be updated, and the runoff forecast results obtained by using this method can be Provide a reliable basis for urban flood control or inflow prediction of large reservoirs.

Description of drawings:

Fig. 1 is a flow chart of the method of constructing an adaptive medium and long-term runoff forecast model in a watershed according to the present invention.

Detailed ways

Below in conjunction with embodiment, the present invention is further described:

Step 1. Collect basic data for forecasting the basin, mainly including: (1) daily, ten-day, monthly and annual flow data of the main control hydrological stations in the basin; (2) ten-day, monthly and annual maximum flow, minimum Flow characteristic value and occurrence time, first and last runoff process data; (3) daily, ten-day, monthly and annual rainfall of main rainfall stations in the basin; (4) collection of 74 circulation indicators, and European medium-term weather Meteorological influencing factors such as the numerical forecast results (reanalysis data) of the forecast center ECMWF or the US National Environmental Forecast Center NCEP.

Step 2. For the main control hydrological station, use the linear regression method to establish a linear regression equation between the annual flow series x(t) and its time series t, and then test the trend of the time series. The decreasing trend, the linear regression equation is:

x(t)=a×t+b

where x(t) is the time series, t is the corresponding time series, a is the slope of the linear equation, which represents the average trend change rate of the time series, and b is the intercept. The values of a and b can be estimated by the least squares method.

Step 3. Divide the watershed into a number of sub-basins. The basis for the division mainly includes: the flow change trend of the control station judged in step 2, the underlying surface conditions and flow-producing methods of the upstream sub-basins of each control station, and the sub-basins that are all the same are merged into A sub-basin, sub-basins with different conditions are distinguished from each other;

Step 4. Identification of forecasting factors. Identifying forecasting factors for different sub-basins. The forecasting factors mainly include: previous precipitation and runoff, 74 circulation indicators, sea surface temperature, solar activity factors, human activity factors, etc., using the correlation analysis method To analyze the degree of correlation between different forecast factors and different sub-basin flows, the calculation formula is:

；为X的样本均值

；为Y的样本均值。In the formula, R _XY is the correlation coefficient between X and Y; n is the number of data samples; X _i is the ith sample value of X; Y _i is the ith sample value of Y

; is the sample mean of X

; is the sample mean of Y.

The value range of the correlation coefficient R _XY is [-1,1]. _{If RXY} is greater than 0, it indicates that there is a positive correlation between forecast object Y and predictor X; if _RXY is less than 0, it indicates that there is a negative correlation between forecast object Y and predictor X; if _RXY is equal to 0, it indicates that forecast object Y and predictor X are negatively correlated. There is no correlation between X. The greater the absolute value of R _XY , the higher the degree of correlation between the forecast object Y and the forecast factor X. For different sub-watersheds, select the top 10% of the predictors with the highest degree of correlation as the set of predictors for different sub-watersheds;

Step 5. Establish a forecasting model library. The forecasting model library mainly includes three methods: physical cause method, hydrological statistics method and artificial intelligence method. Physical cause method includes multiple linear regression model and multiple threshold regression model, and hydrological statistics method includes time series decomposition. Model and rank similarity prediction model, artificial intelligence model including artificial neural network model and support vector machine model;

Step 6. Through deterministic coefficient evaluation, in each sub-basin, for the three methods of physical formation method, hydrological statistics method and artificial intelligence, each method selects a model with a higher deterministic coefficient to form an ensemble forecast for the sub-basin. The composition of the coefficient of certainty is as follows:

为实测序列的均 值，m为资料序列的长度。where DC is the coefficient of certainty, y ₀ (i) is the measured value, y _c (i) is the predicted value,

is the mean value of the measured series, and m is the length of the data series.

水文统计法的权重为人

工智能法模拟的权重为

则集成预报值为：Step 7. According to the result calculated in step 6, determine the weights of different methods of physical formation method, hydrological statistical method and artificial intelligence, and carry out ensemble forecasting. It is assumed that the certainty coefficients of physical formation method, hydrological statistical method and artificial intelligence in step 6 are respectively: A, B, C, the weight of the simulation result of the physical cause method is

The weight of hydrostatistics is human

The weight of artificial intelligence method simulation is

Then the integrated forecast value is:

R=w1y1+w2y2+w3y3

In the formula, w ₁ , w ₂ , and w ₃ are the weight values, y ₁ , y ₂ , and y ₃ are the forecast values of each method, and R _i is the ensemble forecast value of each sub-watershed.

Step 8. Use the minimum root mean square error algorithm to adjust the values of w ₁ , w ₂ , and w ₃ in step 7, so that the root mean square error between the predicted value and the measured value of each sub-basin is minimized, and the predicted value is output. The minimum root mean square error algorithm can be realized by mat lab program.

Step 9. According to the forecast results of each sub-basin, carry out the channel routing to obtain the runoff process of the outlet section of the basin, and complete the forecasting process. The channel routing can use the Muskingen method or the neural network method;

Step 10. Calculate the coefficient of certainty for the result of the basin forecast, the formula is as in Step 6;

Step 11. On the 1st of each month, carry out the trend test on the certainty coefficient of the daily forecast results of the previous year (12 months), adopt the Kandel rank correlation test method, and the calculation formula is:

N为确定性系数序列的总长度，x_i，x_j为系列中的数值，sgn为符号函数，返回值如果数字大于0，则sgn返回1，数字等于0，则返回0，数字小于0，则返回-1，数字参数的符号决定了sgn函数的返回值。In the formula, U is the coefficient of certainty

N is the total length of the deterministic coefficient sequence, x _i , x _j are the numerical values in the series, sgn is the sign function, the return value If the number is greater than 0, sgn returns 1, if the number is equal to 0, it returns 0, if the number is less than 0, Returns -1, and the sign of the numeric parameter determines the return value of the sgn function.

Step 12: For step 11, determine whether it is necessary to update the forecast factor set and re-select the forecast model. If |U|>U _α/2 and U is greater than 0, it means that the change trend of the deterministic coefficient sequence is significant, and the deterministic coefficient sequence is The upward trend, the prediction results are better, and there is no need to update the forecast factor set and re-select the forecast model. When |U|>U _α/2 and U is less than 0, the sequence shows a downward trend, indicating that the forecast results have a downward trend. At this time, go back to step 4 to re-identify the predictor set, and repeat steps 6-10.

Claims (4)

1. A method for checking the trend of medium and long-term runoff forecast results, comprising the following steps: Step 1. Collect the basic data of the forecast watershed; The basic data in the step 1 includes: Basic data A, the daily, ten-day, monthly and annual flow data of the main control hydrological station in the basin, which are used in steps 2 and 11; Basic data B, the ten-day, monthly, annual maximum flow, minimum flow characteristic value and occurrence time of each main control hydrological station, the first runoff and the last runoff process data, which are used in step 3; Basic data C, the daily, ten-day, monthly and annual rainfall of the main rainfall stations in the basin, which are used in step 4; Basic data D. Collect 74 circulation indicators, as well as the numerical forecast results of the European Center for Medium-Range Weather Forecasts ECMWF or the National Center for Environmental Prediction NCEP of the United States and the meteorological impact factors of the reanalysis data, which are used in step 4; Step 2. Based on the basic data, use the linear regression method to establish a linear regression equation between the annual flow series x(t) and its time series t, and then test the trend of the time series; Step 3. Divide the watershed into several sub-basins; In the step 3, the watershed is divided into several sub-basins, and the division is mainly based on the sub-basins that are the same as the flow change trend of the control station determined in step 2, the underlying surface conditions and the flow-producing mode of the upstream sub-basins of each control station. Merged into a sub-basin, and sub-basins with different conditions are distinguished from each other; Step 4. Identification of predictors; In the step 4, forecast factors are identified for different sub-basins respectively, and the forecast factors include: precipitation and runoff in the previous period, 74 circulation indicators, meteorological factor data, sea surface temperature, solar activity factor and human activity factor; The activity factor selects the relative number of sunspots, the associated geomagnetic index, and the solar 10cm radio current as the predictor; human activities are reflected by the urban impervious hardened ground area and the scheduling rules of hydropower stations; the meteorological factor data comes from the step 1. Numerical forecast results and reanalysis data of the European Center for Medium-Range Weather Forecasts ECMWF or the National Center for Environmental Prediction NCEP in the basic data D; The correlation analysis method is used to analyze the degree of correlation between different forecast factors and different sub-basin flows. The calculation formula is as follows:

In the formula, the correlation coefficient R _XY is the correlation coefficient between X and Y; n is the number of data samples; X _i is the ith sample value of X; Y _i is the ith sample value of Y;

is the sample mean of X;

is the sample mean of Y; X represents the flow at the outlet section of a sub-basin, and Y represents a certain forecast factor, and the correlation between different forecast factors and the flow at the outlet section of the sub-basin is calculated respectively; The value range of the correlation coefficient R _XY is [-1,1]; if R _XY is greater than 0, it indicates that there is a positive correlation between the forecast object Y and the forecast factor X; if R _XY is less than 0, it indicates that there is a positive correlation between the forecast object Y and the forecast factor X negative correlation; _RXY equal to 0, indicating that there is no correlation between the forecast object Y and the predictor _X ; the greater the absolute value of RXY, the higher the correlation between the forecast object Y and the predictor X; for different subtypes Watershed, select the top 10% of the predictors with the highest degree of correlation as the set of predictors for different sub-basins; Step 5. Establish a prediction model library; Step 6. Through deterministic coefficient evaluation, in each sub-basin, for the three methods of physical formation method, hydrological statistics method and artificial intelligence, each method selects a model with a higher deterministic coefficient to form an ensemble forecast for the sub-basin. composition; The certainty coefficient formula used in the step 6 is:

where DC is the coefficient of certainty, y ₀ (i) is the measured value, y _c (i) is the predicted value,

is the mean value of the measured sequence, m is the length of the data sequence; Step 7, according to the result calculated in step 6, determine the weight value of different methods of physical formation method, hydrological statistics method and artificial intelligence, and carry out ensemble forecasting; The calculation formula in the step 7 is as follows: In the step 6, the deterministic coefficients of the physical formation method, the hydrological statistical method and the artificial intelligence are respectively A, B, and C, then the weight of the simulation result of the physical formation method is:

The weight of hydrostatistics is

The weight of artificial intelligence method simulation is

Then the integrated forecast value is: R=w ₁ y ₁ +w ₂ y ₂ +w ₃ y ₃ In the formula, w ₁ , w ₂ , w ₃ are the weight values, y ₁ , y ₂ , y ₃ are the forecast values of each method, and R _i is the ensemble forecast value of each sub-watershed; Step 8, using the minimum root mean square error algorithm to adjust the weight value in step 7, so that the root mean square error between the predicted value of each sub-basin and the measured value is minimized, and the predicted value of each sub-basin is output; Step 9. According to the forecast value of each sub-basin, carry out the channel calculation to obtain the runoff process of the outlet section of the entire basin, and complete the forecast process. The neural network method can be used for the channel calculation; Step 10. Calculate the coefficient of certainty for the prediction result of the entire watershed; The certainty coefficient formula used in the step 10 is:

where DC is the coefficient of certainty, y ₀ (i) is the measured value, y _c (i) is the predicted value,

is the mean value of the measured sequence, m is the length of the data sequence; Step 11. On a certain day of each month, carry out a trend test on the certainty coefficient of the daily forecast result of the previous year; In the step 11, on the 1st of each month, a trend test is performed on the certainty coefficient of the daily forecast results of the previous year (12 months), and the test objects of the certainty coefficient are the daily forecast results of the basic data A in step 1 and the main control hydrological stations. The daily flow data of , using the Kandel rank correlation test method, the calculation formula is:

In the formula, U is the coefficient of certainty;

N is the total length of the deterministic coefficient sequence, x _i , x _j are the numerical values in the series, sgn is the sign function, the return value If the number is greater than 0, sgn returns 1, if the number is equal to 0, it returns 0, if the number is less than 0, Returns -1, the sign of the numeric parameter determines the return value of the sgn function; i, j are the numbers of the values in the series, from 1 to n; n is the length of the series; τ is a constant; Step 12. Determine whether it is necessary to update the forecast factor set and reselect the forecast model; In the step 12, for the calculation result of the step 11, when |U|>U _α/2 and U is less than 0, the sequence shows a downward trend, indicating that the forecast result has a downward trend. At this time, return to step 4 to re-identify the forecast factor set, And repeat steps 6-10; Among them, α is the significance level, and U _α/2 is obtained through the normal distribution table through the given significance level. 2. A method for checking the trend of medium and long-term runoff forecasting results according to claim 1, characterized in that: in the step 12, for the calculation result of the step 11, if |U|>U _α/2 and U is greater than 0 , indicating that the change trend of the sequence of certainty coefficients is significant, the sequence of certainty coefficients is on the rise, the prediction results are good, and there is no need to update the forecast factor set and re-select the forecast model; Among them, α is the significance level, and U _α/2 is obtained through the normal distribution table through the given significance level. 3. A kind of medium and long-term runoff forecasting result trend checking method according to claim 1, is characterized in that: in described step 2, according to the basic data A of the watershed main control hydrological station provided in step 1 to construct a linear regression equation, the linear regression The equation gives whether the time series has an increasing or decreasing trend and is: x(t)=a×t+b; where x(t) is the time series, t is the corresponding time series, a is the slope of the linear equation, which represents the average trend change rate of the time series, and b is the intercept; the values of a and b can be estimated by the least squares method. 4. a kind of medium and long-term runoff forecast result trend checking method according to claim 2, is characterized in that: the ten-day, month, year maximum flow, minimum flow of each main control hydrological station in the basic data B described in the step 1 The eigenvalues and occurrence time are used as the conditions for judging the conditions of the underlying surface and the mode of runoff.

Images