Time-space intelligent area rainfall classification weighting mixing quantitative forecasting method based on Beidou water vapor

Abstract

The invention discloses a time-space intelligent area rainfall classification weighting mixing quantitative forecasting method based on Beidou water vapor. The method comprises the following steps: 1, acquiring atmospheric precipitable water and meteorological historical data inverted by Beidou; 2, constructing and training a BPNN rain-free model and a BPNN rain model; 3, training based on the univariate meteorological parameters in combination with CNN and LSTM network models; 4, on the basis of CNN and LSTM network model input, constructing a BPNN rain-free hybrid model and a BPNN rain hybrid model, and training; and 5, subsequent area rainfall classification weighting mixing quantitative forecasting is carried out. The method is simple in steps and reasonable in design, weighting is carried out on the basis of predicted values of the BPNN rain-free model and the BPNN rain model and predicted values of the BPNN rain-free mixed model and the BPNN rain mixed model, subsequent quantitative rainfall forecasting of the to-be-studied area is achieved, and short-term rainfall forecasting precision is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of rainfall quantitative forecasting, and specifically relates to a spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor. Background Art

[0002] Atmospheric water vapor is the main driving force for the occurrence and development of rainfall weather. Short-term sudden rainfall weather is mainly caused by the rapid temporal and spatial changes of water vapor, which seriously threatens people's lives, property and social stability. With the frequent occurrence and increasing intensity of extreme sudden rainfall, my country's demand for accurate forecasts of disastrous rainfall weather caused by water vapor has become more urgent. The water vapor detection technology of the Beidou navigation satellite system is not affected by the weather. It has the advantages of high precision, low cost, and continuous operation. It can provide high-precision, high-resolution multi-dimensional water vapor information. Therefore, the forecast of atmospheric precipitable water volume PWV based on Beidou inversion came into being.

[0003] Existing rainfall forecasting methods using PWV mainly include two types: traditional statistical methods based on the least squares fitting principle and neural network methods. PWV short-term rainfall forecast based on traditional statistical methods is mainly based on long-term historical data. Long-term rainfall, PWV and meteorological parameters are fitted through the least squares principle to achieve short-term rainfall forecast and warning. The nonlinear relationship between rainfall and PWV and meteorological data is obtained based on the neural network method, which can also achieve accurate prediction and warning of short-term rainfall. However, neither of them can realize the classification of current rain models and no rain models, which makes the rainfall in the study area, that is, the surface rainfall prediction accuracy, low.

[0004] Therefore, a reasonably designed spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor is needed. The atmospheric precipitable water inverted by Beidou and historical meteorological data are used to train the BPNN no-rain model and the BPNN rain model. The meteorological parameters predicted by the CNN and LSTM network models are combined with the real values ​​to obtain the BPNN no-rain hybrid model and the BPNN rain hybrid model. Finally, the predicted values ​​of the BPNN no-rain model and the BPNN rain model, as well as the predicted values ​​of the BPNN no-rain hybrid model and the BPNN rain hybrid model are weighted to achieve the quantitative forecast of rainfall in the subsequent study area and improve the accuracy of short-term rainfall forecast. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor in view of the deficiencies in the above-mentioned prior art. The method has simple steps and reasonable design. The atmospheric precipitable water inverted by Beidou and historical meteorological data are first used to train a BPNN rainless model and a BPNN rainy model, and the meteorological parameters predicted by CNN and LSTM network models are combined with real values ​​to obtain a BPNN rainless hybrid model and a BPNN rainy hybrid model. Finally, weighting is performed based on the predicted values ​​of the BPNN rainless model and the BPNN rainy model and the predicted values ​​of the BPNN rainless hybrid model and the BPNN rainy hybrid model, so as to realize the quantitative forecast of rainfall in the subsequent studied area and improve the accuracy of short-term rainfall forecast.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor, characterized in that the method comprises the following steps:

[0007] Step 1: Obtain atmospheric precipitable water and historical meteorological data inverted by Beidou:

[0008] Obtain the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour cumulative rainfall corresponding to each annual accumulation time in the area to be studied to form a training set; where s is a positive integer;

[0009] Step 2: Construction and training of BPNN rainless model and BPNN rainy model:

[0010] Step 201, constructing N BPNN rainless models and N BPNN rainy models;

[0011] Step 202, determine whether the s-hour rainfall corresponding to an annual accumulation time in the training set is greater than 0.3 mm. If it is greater than or equal to 0.3 mm, then the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the previous i-1 annual accumulation times consecutive to the annual accumulation time are used as input, and the s-hour rainfall in the subsequent annual accumulation time is used as the true output, and are respectively input into N BPNN rain models for training to obtain N trained BPNN rain models; wherein s is a positive integer;

[0012] If it is less than 0.3 mm, then the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the previous i-1 annual accumulation time are used as input, and the s-hour rainfall in the subsequent annual accumulation time is used as the true output, and are respectively input into N BPNN rain-free models for training to obtain N trained BPNN rain-free models; where i is a positive integer;

[0013] Step 3: Training based on single variable meteorological parameters combined with CNN and LSTM network models:

[0014] Step 301, construct a first CNN and LSTM network model, a second CNN and LSTM network model, a third CNN and LSTM network model, a fourth CNN and LSTM network model and a fifth CNN and LSTM network model;

[0015] Step 302: input the s-hour average atmospheric precipitable water, s-hour average temperature, s-hour average air pressure, s-hour average relative humidity and s-hour cumulative rainfall corresponding to the annual accumulation into the first CNN and LSTM network model, the second CNN and LSTM network model, the third CNN and LSTM network model, the fourth CNN and LSTM network model and the fifth CNN and LSTM network model for training, to obtain the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model and the trained fifth CNN and LSTM network model;

[0016] Step 4: Based on the CNN and LSTM network model input, build the BPNN rainless mixed model and the BPNN rainy mixed model and train them:

[0017] Step 401, constructing N BPNN rainless mixed models and N BPNN rainy mixed models;

[0018] Step 402: based on the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model, and the trained fifth CNN and LSTM network model, obtain the s-hour average atmospheric precipitable water prediction value, s-hour average temperature prediction value, s-hour average air pressure prediction value, s-hour average relative humidity prediction value, and s-hour rainfall prediction value corresponding to the subsequent annual accumulation time;

[0019] Step 403: According to the method of step 202, the predicted value of atmospheric precipitable water for s hours, the predicted value of temperature for s hours, the predicted value of air pressure for s hours, the predicted value of relative humidity for s hours, and the predicted value of rainfall for s hours corresponding to a subsequent annual accumulation are added as inputs, and the N BPNN mixed models without rain and the N BPNN mixed models with rain are trained respectively to obtain N trained BPNN mixed models without rain and N trained BPNN mixed models with rain.

[0020] Step 5: Subsequent surface rainfall classification weighted hybrid quantitative forecast:

[0021] Step 501: In subsequent data collection moments in the area to be studied, based on N trained BPNN rain models and N trained BPNN no-rain models, obtain the first predicted values ​​of rainfall in s hours for N subsequent annual accumulation times;

[0022] Step 502: In subsequent collection moments in the area to be studied, based on the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model, and the trained fifth CNN and LSTM network model, as well as N BPNN mixed models with rain and N BPNN mixed models without rain, obtain second predicted values ​​of rainfall in s hours corresponding to N subsequent annual accumulation times;

[0023] Step 503, according to S=S1×0.5+S2×0.5, obtain the weighted quantitative value of rainfall corresponding to the next annual accumulation in the area to be studied; wherein S1 is the average of the second predicted values ​​of rainfall in s hours corresponding to N next annual accumulations, and S2 is the average of the second predicted values ​​of rainfall in s hours corresponding to N next annual accumulations.

[0024] The above-mentioned spatial-temporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor, step one, the specific process is as follows:

[0025] Step 101, using a computer to obtain the atmospheric precipitable water PWV at each measurement time of the Beidou inversion in the area to be studied according to the formula PWV=Π·ZWD; wherein ZWD represents the zenith wet delay, and Π represents the atmospheric water vapor conversion coefficient;

[0026] The atmospheric precipitable water PWV inverted by Beidou at each measurement time in the study area within one hour is averaged to obtain the 1-hour atmospheric precipitable water mean;

[0027] Step 102: Use a temperature and humidity sensor to monitor the area to be studied, obtain the temperature, air pressure, and relative humidity of the area to be studied at each sampling time within one hour, and perform mean processing on the temperature, air pressure, and relative humidity of the area to be studied at each sampling time within one hour to obtain a 1-hour temperature mean, a 1-hour air pressure mean, and a 1-hour relative humidity mean;

[0028] Step 103: Use a rainfall sensor to monitor the area to be studied, obtain the rainfall at each sampling time in the area to be studied within one hour, and accumulate and sum the rainfall at each sampling time in the area to be studied within one hour to obtain a 1-hour rainfall cumulative value;

[0029] Step 104, repeating steps 101 to 103 multiple times, obtaining the average value of atmospheric precipitable water for each hour, the average value of temperature for each hour, the average value of air pressure for each hour, the average value of relative humidity for each hour, and the cumulative value of rainfall for each hour;

[0030] Step 105: According to Get the annual cumulative time T corresponding to the jth time point j ; Where t(j) represents the cumulative day corresponding to the date at the jth time point; j is a natural number, and the value of j ranges from 0 to 23;

[0031] according to Get the annual cumulative time T corresponding to the j-(s-1)th time point j-(s-1) ;

[0032] The annual accumulation time T j-(s-1) To T j The corresponding 1-hour temperature averages are then processed to obtain the annual cumulative temperature T j The corresponding s hourly mean temperature; the annual accumulation time T j-(s-1) To T j The corresponding 1-hour pressure averages are then averaged to obtain the annual cumulative time T j The corresponding s-hour average pressure; the annual accumulation time T j-(s-1) To T j The corresponding 1-hour relative humidity averages are then averaged to obtain the annual cumulative time T j The corresponding s hour relative humidity average; the annual accumulation time T j-(s-1) To T j The corresponding 1-hour rainfall accumulation values ​​are then accumulated and summed to obtain the annual accumulation time T j The corresponding s-hour rainfall accumulation value; the annual accumulation time T j-(s-1) To T j The corresponding hourly atmospheric precipitation water averages are then processed to obtain the annual cumulative time T j The corresponding hourly average of atmospheric precipitable water.

[0033] The above-mentioned spatial-temporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor, the acquisition of the zenith wet delay ZWD and the atmospheric water vapor conversion coefficient Π in step 101, the specific process is as follows:

[0034] Step A, using GAMIT / GLOBK software to solve the Beidou station data by computer to obtain the zenith total delay ZTD;

[0035] Step B: Using the Saastamoinen model using a computer The zenith dry delay ZHD is obtained; where P represents the surface pressure at the Beidou station location, represents the latitude of the Beidou station, and H represents the geodetic height of the Beidou station;

[0036] Step C, using a computer to obtain the zenith wet delay ZWD according to the formula ZWD=ZTD-ZHD;

[0037] Step D: Use a computer to calculate the The atmospheric water vapor conversion coefficient Π is obtained; where ρ w represents the density of liquid water, R v is the water vapor gas constant, T m represents the weighted average temperature of the atmosphere, k2′ represents the first constant of the atmospheric refractive index, and k3 represents the second constant of the atmospheric refractive index.

[0038] The above-mentioned spatial-temporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor, step 302, the specific process is as follows:

[0039] The average atmospheric precipitable water content in s hours corresponding to consecutive i years is used as input, and the average atmospheric precipitable water content in s hours corresponding to the next year is used as output, and input into the first CNN and LSTM network models for training to obtain the trained first CNN and LSTM network models;

[0040] The average temperature of s hours corresponding to the accumulation of consecutive i years is used as input, and the average temperature of s hours corresponding to the accumulation of the next year is used as output, which is input into the second CNN and LSTM network model for training to obtain the trained second CNN and LSTM network model;

[0041] The average air pressure for s hours corresponding to consecutive i years is used as input, and the average air pressure for s hours corresponding to the next year is used as output, which is input into the third CNN and LSTM network model for training to obtain the trained third CNN and LSTM network model;

[0042] The average relative humidity of s hours corresponding to consecutive i years is used as input, and the average relative humidity of s hours corresponding to the next year is used as output, which is input into the fourth CNN and LSTM network model for training to obtain the trained fourth CNN and LSTM network model;

[0043] The mean rainfall of s hours corresponding to consecutive i years is taken as input, and the mean rainfall of s hours corresponding to the next year is taken as output, which are input into the fifth CNN and LSTM network models for training to obtain the trained fifth CNN and LSTM network models.

[0044] The above-mentioned spatial-temporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor, step 402, the specific process is as follows:

[0045] The average atmospheric precipitable water in s hours corresponding to consecutive i years in the training set is input into the trained first CNN and LSTM network models, and the predicted value of the average atmospheric precipitable water in s hours corresponding to the next year is output;

[0046] The mean temperature of s hours corresponding to the i consecutive years in the training set is input into the trained second CNN and LSTM network model, and the predicted value of the mean temperature of s hours corresponding to the next year is output;

[0047] The s-hour mean pressure values ​​corresponding to consecutive i-year accumulations in the training set are input into the trained third CNN and LSTM network models, and the predicted value of the s-hour mean pressure value corresponding to the next year accumulation is output;

[0048] The average relative humidity of s hours corresponding to consecutive i years in the training set is input into the trained fourth CNN and LSTM network model, and the predicted value of the average relative humidity of s hours corresponding to the next year is output;

[0049] The s-hour rainfall corresponding to consecutive i-year accumulations in the training set is input into the trained fourth CNN and LSTM network model, and the predicted value of s-hour rainfall corresponding to the next year accumulation is output.

[0050] The above-mentioned spatial-temporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor, step 403, the specific process is as follows:

[0051] It is judged whether the s-hour rainfall corresponding to an annual accumulation time in the training set is greater than 0.3 mm. If it is greater than or equal to 0.3 mm, the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the first i-1 annual accumulation time consecutive to the annual accumulation time, and the s-hour atmospheric precipitable water average predicted value, s-hour mean temperature predicted value, s-hour mean air pressure predicted value, s-hour mean relative humidity predicted value and s-hour rainfall predicted value corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time are used as the input of the BPNN mixed model with rain, and the true value of the s-hour rainfall corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time is used as the output of the BPNN mixed model with rain, and training is performed to obtain the trained BPNN mixed model with rain;

[0052] If it is less than 0.3 mm, then the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the first i-1 annual accumulation times consecutive to the annual accumulation time, and the s-hour mean atmospheric precipitable water forecast value, s-hour mean temperature forecast value, s-hour mean air pressure forecast value, s-hour mean relative humidity forecast value and s-hour rainfall forecast value corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time are taken as the input of the BPNN rain-free mixed model, and the true value of the s-hour rainfall corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time is taken as the output of the BPNN rain-free mixed model for training to obtain the trained BPNN rain-free mixed model.

[0053] The above-mentioned spatial-temporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor, step 501, the specific process is as follows:

[0054] In the subsequent collection time of the area to be studied, if the s-hour rainfall corresponding to the current annual accumulation time is greater than or equal to 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation corresponding to the current annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation corresponding to the previous i-1 annual accumulation times consecutive to the current annual accumulation time are respectively input into N trained BPNN rain models to obtain the first predicted value of s-hour rainfall in N subsequent annual accumulation times;

[0055] If the s-hour rainfall corresponding to the current annual accumulation time is less than 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the current annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the previous i-1 annual accumulation times consecutive to the current annual accumulation time are input into N trained BPNN rain-free models to obtain the first predicted value of the s-hour rainfall in N subsequent annual accumulation times.

[0056] The above-mentioned spatial-temporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor, step 502, the specific process is as follows:

[0057] Step 5021, input the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation value corresponding to the current annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation value corresponding to the previous i-1 annual accumulation times consecutive to the current annual accumulation time into the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model and the trained fifth CNN and LSTM network model, respectively, to obtain the s-hour mean atmospheric precipitable water univariate prediction value, s-hour mean temperature univariate prediction value, s-hour mean air pressure univariate prediction value, s-hour mean relative humidity univariate prediction value and s-hour rainfall univariate prediction value corresponding to the subsequent annual accumulation time;

[0058] Step 5022: If the s-hour rainfall corresponding to the current annual accumulation time is greater than or equal to 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the previous i-1 annual accumulation times of the current annual accumulation time, and the s-hour mean atmospheric precipitable water univariate prediction value, s-hour mean temperature univariate prediction value, s-hour mean air pressure univariate prediction value, s-hour mean relative humidity univariate prediction value and s-hour rainfall univariate prediction value corresponding to the subsequent annual accumulation time are input into N BPNN rain mixed models to obtain the s-hour rainfall second prediction value corresponding to N subsequent annual accumulation times;

[0059] If the s-hour rainfall corresponding to the current annual accumulation time is less than 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the previous i-1 annual accumulation times of the current annual accumulation time, and the s-hour mean atmospheric precipitable water univariate prediction value, s-hour mean temperature univariate prediction value, s-hour mean air pressure univariate prediction value, s-hour mean relative humidity univariate prediction value and s-hour rainfall univariate prediction value corresponding to the subsequent annual accumulation time are input into N BPNN rain-free mixed models to obtain the second prediction value of the s-hour rainfall corresponding to N subsequent annual accumulation times.

[0060] Compared with the prior art, the present invention has the following advantages:

[0061] 1. The method of the present invention has simple steps and reasonable design, solves the current problem of quantitative forecasting of surface rainfall, and ensures the accuracy and reliability of rainfall forecast results.

[0062] 2. The present invention uses the atmospheric precipitable water inverted by Beidou and the meteorological historical data to train the constructed N BPNN rainless models and N BPNN rainy models respectively, and obtains N trained BPNN rainless models and N BPNN rainy models; on the one hand, considering the randomness of the BPNN model grid initialization, therefore, N BPNN rainless models and N BPNN rainy models are established for random training to obtain the accuracy of the rainfall prediction value; on the other hand, based on whether the corresponding s-hour rainfall in the annual accumulation time is greater than 0.3 mm, it is judged that the input N BPNN rainless models and N BPNN rainy models are predicted, thereby realizing the classification of rainy or not conditions and improving the prediction accuracy.

[0063] 3. The present invention trains the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour cumulative rainfall through CNN and LSTM network models respectively to realize the prediction of each single variable meteorological parameter.

[0064] 4. The present invention trains N BPNN rain-free mixed models and N BPNN rain-with mixed models based on the s-hour average atmospheric precipitable water, s-hour average temperature, s-hour average air pressure, s-hour average relative humidity and s-hour cumulative rainfall corresponding to i consecutive annual accumulations, as well as the predicted values ​​of each univariate meteorological parameter corresponding to a subsequent annual accumulation, to obtain N trained BPNN rain-free mixed models and N trained BPNN rain-with mixed models, which further assist in improving the accuracy of prediction by inputting prediction values.

[0065] 5. In the subsequent data collection time of the area to be studied, the present invention obtains a first prediction value of s-hour rainfall in N subsequent annual accumulation times based on N trained BPNN rain models and N trained BPNN no-rain models; obtains a second prediction value of s-hour rainfall corresponding to N subsequent annual accumulation times based on N BPNN rain mixed models and N BPNN no-rain mixed models; and weighted average of the mean of the second prediction value of s-hour rainfall corresponding to N subsequent annual accumulation times and the mean of the second prediction value of s-hour rainfall, thereby realizing quantitative forecast of rainfall in the subsequent area to be studied and improving the accuracy of short-term rainfall forecast.

[0066] In summary, the method of the present invention has simple steps and reasonable design. First, the atmospheric precipitable water inverted by Beidou and historical meteorological data are used to train the BPNN rainless model and the BPNN rainy model, and the meteorological parameters predicted by the CNN and LSTM network models are combined with the real values ​​to obtain the BPNN rainless mixed model and the BPNN rainy mixed model. Finally, weighting is performed based on the predicted values ​​of the BPNN rainless model and the BPNN rainy model and the predicted values ​​of the BPNN rainless mixed model and the BPNN rainy mixed model, so as to realize the quantitative forecast of rainfall in the subsequent study area and improve the accuracy of short-term rainfall forecast.

[0067] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The figure is a flowchart of the method of the present invention. DETAILED DESCRIPTION

[0069] like Figure 1 As shown, the spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor of the present invention comprises the following steps:

[0070] Step 1: Obtain atmospheric precipitable water and historical meteorological data inverted by Beidou:

[0071] Obtain the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour cumulative rainfall corresponding to each annual accumulation time in the area to be studied to form a training set; where s is a positive integer;

[0072] Step 2: Construction and training of BPNN rainless model and BPNN rainy model:

[0073] Step 201, constructing N BPNN rainless models and N BPNN rainy models;

[0074] Step 202, determine whether the s-hour rainfall corresponding to an annual accumulation time in the training set is greater than 0.3 mm. If it is greater than or equal to 0.3 mm, then the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the previous i-1 annual accumulation times consecutive to the annual accumulation time are used as input, and the s-hour rainfall in the subsequent annual accumulation time is used as the true output, and are respectively input into N BPNN rain models for training to obtain N trained BPNN rain models; wherein s is a positive integer;

[0075] If it is less than 0.3 mm, then the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the previous i-1 annual accumulation time are used as input, and the s-hour rainfall in the subsequent annual accumulation time is used as the true output, and are respectively input into N BPNN rain-free models for training to obtain N trained BPNN rain-free models; where i is a positive integer;

[0076] Step 3: Training based on single variable meteorological parameters combined with CNN and LSTM network models:

[0077] Step 301, construct a first CNN and LSTM network model, a second CNN and LSTM network model, a third CNN and LSTM network model, a fourth CNN and LSTM network model and a fifth CNN and LSTM network model;

[0078] Step 302: input the s-hour average atmospheric precipitable water, s-hour average temperature, s-hour average air pressure, s-hour average relative humidity and s-hour cumulative rainfall corresponding to the annual accumulation into the first CNN and LSTM network model, the second CNN and LSTM network model, the third CNN and LSTM network model, the fourth CNN and LSTM network model and the fifth CNN and LSTM network model for training, to obtain the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model and the trained fifth CNN and LSTM network model;

[0079] Step 4: Based on the CNN and LSTM network model input, build the BPNN rainless mixed model and the BPNN rainy mixed model and train them:

[0080] Step 401, constructing N BPNN rainless mixed models and N BPNN rainy mixed models;

[0081] Step 402: based on the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model, and the trained fifth CNN and LSTM network model, obtain the s-hour average atmospheric precipitable water prediction value, s-hour average temperature prediction value, s-hour average air pressure prediction value, s-hour average relative humidity prediction value, and s-hour rainfall prediction value corresponding to the subsequent annual accumulation time;

[0082] Step 403: According to the method of step 202, the predicted value of atmospheric precipitable water for s hours, the predicted value of temperature for s hours, the predicted value of air pressure for s hours, the predicted value of relative humidity for s hours, and the predicted value of rainfall for s hours corresponding to a subsequent annual accumulation are added as inputs, and the N BPNN mixed models without rain and the N BPNN mixed models with rain are trained respectively to obtain N trained BPNN mixed models without rain and N trained BPNN mixed models with rain.

[0083] Step 5: Subsequent surface rainfall classification weighted hybrid quantitative forecast:

[0084] Step 501: In subsequent data collection moments in the area to be studied, based on N trained BPNN rain models and N trained BPNN no-rain models, obtain the first predicted values ​​of rainfall in s hours for N subsequent annual accumulation times;

[0085] Step 502: In subsequent collection moments in the area to be studied, based on the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model, and the trained fifth CNN and LSTM network model, as well as N BPNN mixed models with rain and N BPNN mixed models without rain, obtain second predicted values ​​of rainfall in s hours corresponding to N subsequent annual accumulation times;

[0086] Step 503, according to S=S1×0.5+S2×0.5, obtain the weighted quantitative value of rainfall corresponding to the next annual accumulation in the area to be studied; wherein S1 is the average of the second predicted values ​​of rainfall in s hours corresponding to N next annual accumulations, and S2 is the average of the second predicted values ​​of rainfall in s hours corresponding to N next annual accumulations.

[0087] In this embodiment, step one, the specific process is as follows:

[0088] Step 101, using a computer to obtain the atmospheric precipitable water PWV at each measurement time of the Beidou inversion in the area to be studied according to the formula PWV=Π·ZWD; wherein ZWD represents the zenith wet delay, and Π represents the atmospheric water vapor conversion coefficient;

[0089] The atmospheric precipitable water PWV inverted by Beidou at each measurement time in the study area within one hour is averaged to obtain the 1-hour atmospheric precipitable water mean;

[0090] Step 102: Use a temperature and humidity sensor to monitor the area to be studied, obtain the temperature, air pressure, and relative humidity of the area to be studied at each sampling time within one hour, and perform mean processing on the temperature, air pressure, and relative humidity of the area to be studied at each sampling time within one hour to obtain a 1-hour temperature mean, a 1-hour air pressure mean, and a 1-hour relative humidity mean;

[0091] Step 103: Use a rainfall sensor to monitor the area to be studied, obtain the rainfall at each sampling time in the area to be studied within one hour, and accumulate and sum the rainfall at each sampling time in the area to be studied within one hour to obtain a 1-hour rainfall cumulative value;

[0092] Step 104, repeating steps 101 to 103 multiple times, obtaining the average value of atmospheric precipitable water for each hour, the average value of temperature for each hour, the average value of air pressure for each hour, the average value of relative humidity for each hour, and the cumulative value of rainfall for each hour;

[0093] Step 105: According to Get the annual cumulative time T corresponding to the jth time point j ; Where t(j) represents the cumulative day corresponding to the date at the jth time point; j is a natural number, and the value of j ranges from 0 to 23;

[0094] according to Get the annual cumulative time T corresponding to the j-(s-1)th time point j-(s-1) ;

[0095] The annual accumulation time T j-(s-1) To T j The corresponding 1-hour temperature averages are then processed to obtain the annual cumulative temperature T j The corresponding s hourly mean temperature; the annual accumulation time T j-(s-1) To T j The corresponding 1-hour pressure averages are then averaged to obtain the annual cumulative time T j The corresponding s-hour average pressure; the annual accumulation time T j-(s-1) To T j The corresponding 1-hour relative humidity averages are then averaged to obtain the annual cumulative time T j The corresponding s hour relative humidity average; the annual accumulation time T j-(s-1) To T j The corresponding 1-hour rainfall accumulation values ​​are then accumulated and summed to obtain the annual accumulation time T j The corresponding s-hour rainfall accumulation value; the annual accumulation time T j-(s-1) To T j The corresponding hourly atmospheric precipitation water averages are then processed to obtain the annual cumulative time T j The corresponding hourly average of atmospheric precipitable water.

[0096] In this embodiment, the specific process of obtaining the zenith wet delay ZWD and the atmospheric water vapor conversion coefficient Π in step 101 is as follows:

[0097] Step A, using GAMIT / GLOBK software to solve the Beidou station data by computer to obtain the zenith total delay ZTD;

[0098] Step B: Using the Saastamoinen model using a computer The zenith dry delay ZHD is obtained; where P represents the surface pressure at the Beidou station location, represents the latitude of the Beidou station, and H represents the geodetic height of the Beidou station;

[0099] Step C, using a computer to obtain the zenith wet delay ZWD according to the formula ZWD=ZTD-ZHD;

[0100] Step D: Use a computer to calculate the The atmospheric water vapor conversion coefficient Π is obtained; where ρ w represents the density of liquid water, R v is the water vapor gas constant, T m represents the weighted average temperature of the atmosphere, k2′ represents the first constant of the atmospheric refractive index, and k3 represents the second constant of the atmospheric refractive index.

[0101] In this embodiment, step 302, the specific process is as follows:

[0102] The average atmospheric precipitable water content in s hours corresponding to consecutive i years is used as input, and the average atmospheric precipitable water content in s hours corresponding to the next year is used as output, and input into the first CNN and LSTM network models for training to obtain the trained first CNN and LSTM network models;

[0103] The average temperature of s hours corresponding to the accumulation of consecutive i years is used as input, and the average temperature of s hours corresponding to the accumulation of the next year is used as output, which is input into the second CNN and LSTM network model for training to obtain the trained second CNN and LSTM network model;

[0104] The average air pressure for s hours corresponding to consecutive i years is used as input, and the average air pressure for s hours corresponding to the next year is used as output, which is input into the third CNN and LSTM network model for training to obtain the trained third CNN and LSTM network model;

[0105] The average relative humidity of s hours corresponding to consecutive i years is used as input, and the average relative humidity of s hours corresponding to the next year is used as output, which is input into the fourth CNN and LSTM network model for training to obtain the trained fourth CNN and LSTM network model;

[0106] The mean rainfall of s hours corresponding to consecutive i years is taken as input, and the mean rainfall of s hours corresponding to the next year is taken as output, which are input into the fifth CNN and LSTM network models for training to obtain the trained fifth CNN and LSTM network models.

[0107] In this embodiment, step 402, the specific process is as follows:

[0108] The average atmospheric precipitable water in s hours corresponding to consecutive i years in the training set is input into the trained first CNN and LSTM network models, and the predicted value of the average atmospheric precipitable water in s hours corresponding to the next year is output;

[0109] The mean temperature of s hours corresponding to the i consecutive years in the training set is input into the trained second CNN and LSTM network model, and the predicted value of the mean temperature of s hours corresponding to the next year is output;

[0110] The s-hour mean pressure values ​​corresponding to consecutive i-year accumulations in the training set are input into the trained third CNN and LSTM network models, and the predicted value of the s-hour mean pressure value corresponding to the next year accumulation is output;

[0111] The average relative humidity of s hours corresponding to consecutive i years in the training set is input into the trained fourth CNN and LSTM network model, and the predicted value of the average relative humidity of s hours corresponding to the next year is output;

[0112] The s-hour rainfall corresponding to consecutive i-year accumulations in the training set is input into the trained fourth CNN and LSTM network model, and the predicted value of s-hour rainfall corresponding to the next year accumulation is output.

[0113] In this embodiment, step 403, the specific process is as follows:

[0114] It is judged whether the s-hour rainfall corresponding to an annual accumulation time in the training set is greater than 0.3 mm. If it is greater than or equal to 0.3 mm, the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the first i-1 annual accumulation time consecutive to the annual accumulation time, and the s-hour atmospheric precipitable water average predicted value, s-hour mean temperature predicted value, s-hour mean air pressure predicted value, s-hour mean relative humidity predicted value and s-hour rainfall predicted value corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time are used as the input of the BPNN mixed model with rain, and the true value of the s-hour rainfall corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time is used as the output of the BPNN mixed model with rain, and training is performed to obtain the trained BPNN mixed model with rain;

[0115] If it is less than 0.3 mm, then the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the first i-1 annual accumulation times consecutive to the annual accumulation time, and the s-hour mean atmospheric precipitable water forecast value, s-hour mean temperature forecast value, s-hour mean air pressure forecast value, s-hour mean relative humidity forecast value and s-hour rainfall forecast value corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time are taken as the input of the BPNN rain-free mixed model, and the true value of the s-hour rainfall corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time is taken as the output of the BPNN rain-free mixed model for training to obtain the trained BPNN rain-free mixed model.

[0116] In this embodiment, step 501, the specific process is as follows:

[0117] In the subsequent collection time of the area to be studied, if the s-hour rainfall corresponding to the current annual accumulation time is greater than or equal to 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation corresponding to the current annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation corresponding to the previous i-1 annual accumulation times consecutive to the current annual accumulation time are respectively input into N trained BPNN rain models to obtain the first predicted value of s-hour rainfall in N subsequent annual accumulation times;

[0118] If the s-hour rainfall corresponding to the current annual accumulation time is less than 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the current annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the previous i-1 annual accumulation times consecutive to the current annual accumulation time are input into N trained BPNN rain-free models to obtain the first predicted value of the s-hour rainfall in N subsequent annual accumulation times.

[0119] In this embodiment, step 502, the specific process is as follows:

[0120] Step 5021, input the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation value corresponding to the current annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation value corresponding to the previous i-1 annual accumulation times consecutive to the current annual accumulation time into the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model and the trained fifth CNN and LSTM network model, respectively, to obtain the s-hour mean atmospheric precipitable water univariate prediction value, s-hour mean temperature univariate prediction value, s-hour mean air pressure univariate prediction value, s-hour mean relative humidity univariate prediction value and s-hour rainfall univariate prediction value corresponding to the subsequent annual accumulation time;

[0121] Step 5022: If the s-hour rainfall corresponding to the current annual accumulation time is greater than or equal to 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the previous i-1 annual accumulation times of the current annual accumulation time, and the s-hour mean atmospheric precipitable water univariate prediction value, s-hour mean temperature univariate prediction value, s-hour mean air pressure univariate prediction value, s-hour mean relative humidity univariate prediction value and s-hour rainfall univariate prediction value corresponding to the subsequent annual accumulation time are input into N BPNN rain mixed models to obtain the s-hour rainfall second prediction value corresponding to N subsequent annual accumulation times;

[0122] If the s-hour rainfall corresponding to the current annual accumulation time is less than 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the previous i-1 annual accumulation times of the current annual accumulation time, and the s-hour mean atmospheric precipitable water univariate prediction value, s-hour mean temperature univariate prediction value, s-hour mean air pressure univariate prediction value, s-hour mean relative humidity univariate prediction value and s-hour rainfall univariate prediction value corresponding to the subsequent annual accumulation time are input into N BPNN rain-free mixed models to obtain the second prediction value of the s-hour rainfall corresponding to N subsequent annual accumulation times.

[0123] In this embodiment, when it is implemented, the unit of the surface pressure P corresponding to the Beidou station position in step B is hPa, and the latitude of the Beidou station is The unit of is rad, and the unit of the Beidou station geodetic height H is km.

[0124] In this embodiment, the weighted average temperature of the atmosphere T m It can be obtained by referring to conventional methods in the art.

[0125] In this embodiment, during the specific implementation, the density of liquid water in step D is w , the value is 1.0×10 3 kg / m 3 , R v The value of T is 461.495 J / (kg·K), m The unit is K, and the value of k2′ is 17±10K / hPa, and the value of k3 is (3.776±0.04)×10 5 K 2 / hPa.

[0126] In this embodiment, during the specific implementation, the sampling time of the temperature, air pressure, and relative humidity in step 102 and the sampling time of the rainfall in step 103 are set to 5 minutes;

[0127] The measurement time of atmospheric precipitable water in step 101 is set to 30 seconds, so that the measurement times of atmospheric precipitable water, temperature, air pressure, relative humidity and rainfall are all converted to the same 1-hour time.

[0128] In this embodiment, during the specific implementation, the value of N is 100, taking into account the randomness of the BPNN model grid initialization.

[0129] In this embodiment, during specific implementation, the structures of the N BPNN rainless models, the N BPNN rainy models, the N BPNN rainless mixed models and the N BPNN rainy mixed models are the same, and all include an input layer, a hidden layer and an output layer;

[0130] In this embodiment, i is set to 4, and the number of neurons in the input layer of the BPNN rainless model and the BPNN rainy model is N in is 21, and the number of neurons in the hidden layer is 2×N in +1, the number of neurons in the output layer is 1. The number of neurons in the input layer in the BPNN mixed model without rain and the BPNN mixed model with rain is 26.

[0131] In this embodiment, during specific implementation, the structures of the first CNN and LSTM network model to the fifth CNN and LSTM network model are the same, and all include an input layer, a convolutional layer, a pooling layer, an LSTM layer, a fully connected layer and an output layer;

[0132] In this embodiment, during specific implementation, the input layer in the first CNN and LSTM network model to the fifth CNN and LSTM network model has 4 neurons, the convolution kernel in the convolution layer is 1 and the size of the convolution kernel is 2×1, the pooling layer is an average pooling layer, the number of neurons in the hidden layer in the LSTM layer is 64, the number of neurons in the fully connected layer is 64, and the output layer is 1 neuron.

[0133] In this embodiment, during the specific implementation, the training of N BPNN rainless models and N BPNN rainy models, the training of the first CNN and LSTM network model and the fifth CNN and LSTM network model, and the training of N BPNN rainless mixed models and N BPNN rainy mixed models can refer to the conventional training method of neural network.

[0134] In summary, the method of the present invention has simple steps and reasonable design. First, the atmospheric precipitable water inverted by Beidou and historical meteorological data are used to train the BPNN rainless model and the BPNN rainy model, and the meteorological parameters predicted by the CNN and LSTM network models are combined with the real values ​​to obtain the BPNN rainless mixed model and the BPNN rainy mixed model. Finally, weighting is performed based on the predicted values ​​of the BPNN rainless model and the BPNN rainy model and the predicted values ​​of the BPNN rainless mixed model and the BPNN rainy mixed model, so as to realize the quantitative forecast of rainfall in the subsequent study area and improve the accuracy of short-term rainfall forecast.

[0135] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor, characterized in that: The method comprises the following steps: Step 1: Obtain atmospheric precipitable water and historical meteorological data inverted by Beidou: Obtain the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour cumulative rainfall corresponding to each annual accumulation time in the area to be studied to form a training set; where s is a positive integer; Step 2: Construction and training of BPNN rainless model and BPNN rainy model: Step 201, constructing N BPNN rainless models and N BPNN rainy models; Step 202, determine whether the s-hour rainfall corresponding to an annual accumulation time in the training set is greater than 0.3 mm. If it is greater than or equal to 0.3 mm, then the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the previous i-1 annual accumulation times consecutive to the annual accumulation time are used as input, and the s-hour rainfall in the subsequent annual accumulation time is used as the true output, and are respectively input into N BPNN rain models for training to obtain N trained BPNN rain models; wherein s is a positive integer; If it is less than 0.3 mm, then the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the previous i-1 annual accumulation time are used as input, and the s-hour rainfall in the subsequent annual accumulation time is used as the true output, and are respectively input into N BPNN rain-free models for training to obtain N trained BPNN rain-free models; where i is a positive integer; Step 3: Training based on single variable meteorological parameters combined with CNN and LSTM network models: Step 301, construct a first CNN and LSTM network model, a second CNN and LSTM network model, a third CNN and LSTM network model, a fourth CNN and LSTM network model and a fifth CNN and LSTM network model; Step 302: input the s-hour average atmospheric precipitable water, s-hour average temperature, s-hour average air pressure, s-hour average relative humidity and s-hour cumulative rainfall corresponding to the annual accumulation into the first CNN and LSTM network model, the second CNN and LSTM network model, the third CNN and LSTM network model, the fourth CNN and LSTM network model and the fifth CNN and LSTM network model for training, to obtain the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model and the trained fifth CNN and LSTM network model; Step 4: Based on the CNN and LSTM network model input, build the BPNN rainless mixed model and the BPNN rainy mixed model and train them: Step 401, constructing N BPNN rainless mixed models and N BPNN rainy mixed models; Step 402: based on the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model, and the trained fifth CNN and LSTM network model, obtain the s-hour average atmospheric precipitable water prediction value, s-hour average temperature prediction value, s-hour average air pressure prediction value, s-hour average relative humidity prediction value, and s-hour rainfall prediction value corresponding to the subsequent annual accumulation time; Step 403: According to the method of step 202, the predicted value of atmospheric precipitable water for s hours, the predicted value of temperature for s hours, the predicted value of air pressure for s hours, the predicted value of relative humidity for s hours, and the predicted value of rainfall for s hours corresponding to a subsequent annual accumulation are added as inputs, and the N BPNN mixed models without rain and the N BPNN mixed models with rain are trained respectively to obtain N trained BPNN mixed models without rain and N trained BPNN mixed models with rain. Step 5: Subsequent surface rainfall classification weighted hybrid quantitative forecast: Step 501: In subsequent data collection moments in the area to be studied, based on N trained BPNN rain models and N trained BPNN no-rain models, obtain the first predicted values ​​of rainfall in s hours for N subsequent annual accumulation times; Step 502: In subsequent collection moments in the area to be studied, based on the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model, and the trained fifth CNN and LSTM network model, as well as N BPNN mixed models with rain and N BPNN mixed models without rain, obtain second predicted values ​​of rainfall in s hours corresponding to N subsequent annual accumulation times; Step 503, according to S=S1×0.5+S2×0.5, obtain the weighted quantitative value of rainfall corresponding to the next annual accumulation in the area to be studied; wherein S1 is the average of the second predicted values ​​of rainfall in s hours corresponding to N next annual accumulations, and S2 is the average of the second predicted values ​​of rainfall in s hours corresponding to N next annual accumulations.

2. A spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor according to claim 1, characterized in that: Step 1: The specific process is as follows: Step 101, using a computer to obtain the atmospheric precipitable water PWV at each measurement time of the Beidou inversion in the area to be studied according to the formula PWV=Π·ZWD; wherein ZWD represents the zenith wet delay, and Π represents the atmospheric water vapor conversion coefficient; The atmospheric precipitable water PWV inverted by Beidou at each measurement time in the study area within one hour is averaged to obtain the 1-hour atmospheric precipitable water mean; Step 102: Use a temperature and humidity sensor to monitor the area to be studied, obtain the temperature, air pressure, and relative humidity of the area to be studied at each sampling time within one hour, and perform mean processing on the temperature, air pressure, and relative humidity of the area to be studied at each sampling time within one hour to obtain a 1-hour temperature mean, a 1-hour air pressure mean, and a 1-hour relative humidity mean; Step 103: Use a rainfall sensor to monitor the area to be studied, obtain the rainfall at each sampling time in the area to be studied within one hour, and accumulate and sum the rainfall at each sampling time in the area to be studied within one hour to obtain a 1-hour rainfall cumulative value; Step 104, repeating steps 101 to 103 multiple times, obtaining the average value of atmospheric precipitable water for each hour, the average value of temperature for each hour, the average value of air pressure for each hour, the average value of relative humidity for each hour, and the cumulative value of rainfall for each hour; Step 105: According to Get the annual cumulative time T corresponding to the jth time point j ; Where t(j) represents the cumulative day corresponding to the date at the jth time point; j is a natural number, and the value of j ranges from 0 to 23; according to Get the annual cumulative time T corresponding to the j-(s-1)th time point j-(s-1) ; The annual accumulation time T j-(s-1) To T j The corresponding 1-hour temperature averages are then processed to obtain the annual cumulative temperature T j The corresponding s hourly mean temperature; the annual accumulation time T j-(s-1) To T j The corresponding 1-hour pressure averages are then averaged to obtain the annual cumulative time T j The corresponding s-hour average pressure; the annual accumulation time T j-(s-1) To T j The corresponding 1-hour relative humidity averages are then averaged to obtain the annual cumulative time T j The corresponding s hour relative humidity average; the annual accumulation time T j-(s-1) To T j The corresponding 1-hour rainfall accumulation values ​​are then accumulated and summed to obtain the annual accumulation time T j The corresponding s-hour rainfall accumulation value; the annual accumulation time T j-(s-1) To T j The corresponding hourly atmospheric precipitation water averages are then processed to obtain the annual cumulative time T j The corresponding hourly average of atmospheric precipitable water.

3. A spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor according to claim 2, characterized in that: The specific process of obtaining the zenith wet delay ZWD and the atmospheric water vapor conversion coefficient Π in step 101 is as follows: Step A, using GAMIT / GLOBK software to solve the Beidou station data by computer to obtain the zenith total delay ZTD; Step B: Using the Saastamoinen model using a computer The zenith dry delay ZHD is obtained; where P represents the surface pressure at the Beidou station location, represents the latitude of the Beidou station, and H represents the geodetic height of the Beidou station; Step C, using a computer to obtain the zenith wet delay ZWD according to the formula ZWD=ZTD-ZHD; Step D: Use a computer to calculate the The atmospheric water vapor conversion coefficient Π is obtained; where ρ w represents the density of liquid water, R v is the water vapor gas constant, T m represents the weighted average temperature of the atmosphere, k2′ represents the first constant of the atmospheric refractive index, and k3 represents the second constant of the atmospheric refractive index.

4. A spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor according to claim 1, characterized in that: Step 302, the specific process is as follows: The average atmospheric precipitable water content in s hours corresponding to consecutive i years is used as input, and the average atmospheric precipitable water content in s hours corresponding to the next year is used as output, and input into the first CNN and LSTM network models for training to obtain the trained first CNN and LSTM network models; The average temperature of s hours corresponding to the accumulation of consecutive i years is used as input, and the average temperature of s hours corresponding to the accumulation of the next year is used as output, which is input into the second CNN and LSTM network model for training to obtain the trained second CNN and LSTM network model; The average air pressure for s hours corresponding to consecutive i years is used as input, and the average air pressure for s hours corresponding to the next year is used as output, which is input into the third CNN and LSTM network model for training to obtain the trained third CNN and LSTM network model; The average relative humidity of s hours corresponding to consecutive i years is used as input, and the average relative humidity of s hours corresponding to the next year is used as output, which is input into the fourth CNN and LSTM network model for training to obtain the trained fourth CNN and LSTM network model; The mean rainfall of s hours corresponding to consecutive i years is taken as input, and the mean rainfall of s hours corresponding to the next year is taken as output, which are input into the fifth CNN and LSTM network models for training to obtain the trained fifth CNN and LSTM network models.

5. A spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor according to claim 1, characterized in that: Step 402, the specific process is as follows: The average atmospheric precipitable water in s hours corresponding to consecutive i years in the training set is input into the trained first CNN and LSTM network models, and the predicted value of the average atmospheric precipitable water in s hours corresponding to the next year is output; The mean temperature of s hours corresponding to consecutive i years in the training set is input into the trained second CNN and LSTM network model, and the predicted value of the mean temperature of s hours corresponding to the next year is output; The s-hour mean pressure values ​​corresponding to consecutive i-year accumulations in the training set are input into the trained third CNN and LSTM network models, and the predicted value of the s-hour mean pressure value corresponding to the next year accumulation is output; The average relative humidity of s hours corresponding to consecutive i years in the training set is input into the trained fourth CNN and LSTM network model, and the predicted value of the average relative humidity of s hours corresponding to the next year is output; The s-hour rainfall corresponding to consecutive i-year accumulations in the training set is input into the trained fourth CNN and LSTM network model, and the predicted value of s-hour rainfall corresponding to the next year accumulation is output.

6. A spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor according to claim 1, characterized in that: Step 403, the specific process is as follows: It is judged whether the s-hour rainfall corresponding to an annual accumulation time in the training set is greater than 0.3 mm. If it is greater than or equal to 0.3 mm, the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the first i-1 annual accumulation time consecutive to the annual accumulation time, and the s-hour atmospheric precipitable water average predicted value, s-hour mean temperature predicted value, s-hour mean air pressure predicted value, s-hour mean relative humidity predicted value and s-hour rainfall predicted value corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time are used as the input of the BPNN mixed model with rain, and the true value of the s-hour rainfall corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time is used as the output of the BPNN mixed model with rain, and training is performed to obtain the trained BPNN mixed model with rain; If it is less than 0.3 mm, then the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall corresponding to the first i-1 annual accumulation times consecutive to the annual accumulation time, and the s-hour mean atmospheric precipitable water forecast value, s-hour mean temperature forecast value, s-hour mean air pressure forecast value, s-hour mean relative humidity forecast value and s-hour rainfall forecast value corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time are taken as the input of the BPNN rain-free mixed model, and the true value of the s-hour rainfall corresponding to the subsequent annual accumulation time consecutive to the annual accumulation time is taken as the output of the BPNN rain-free mixed model for training to obtain the trained BPNN rain-free mixed model.

7. A spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor according to claim 1, characterized in that: Step 501, the specific process is as follows: In the subsequent collection time of the area to be studied, if the s-hour rainfall corresponding to the current annual accumulation time is greater than or equal to 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation corresponding to the current annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation corresponding to the previous i-1 annual accumulation times consecutive to the current annual accumulation time are respectively input into N trained BPNN rain models to obtain the first predicted value of s-hour rainfall in N subsequent annual accumulation times; If the s-hour rainfall corresponding to the current annual accumulation time is less than 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the current annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the previous i-1 annual accumulation times consecutive to the current annual accumulation time are input into N trained BPNN rain-free models to obtain the first predicted value of the s-hour rainfall in N subsequent annual accumulation times.

8. A spatiotemporal intelligent surface rainfall classification weighted hybrid quantitative forecasting method based on Beidou water vapor according to claim 1, characterized in that: Step 502, the specific process is as follows: Step 5021, input the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation value corresponding to the current annual accumulation time, and the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall accumulation value corresponding to the previous i-1 annual accumulation times consecutive to the current annual accumulation time into the trained first CNN and LSTM network model, the trained second CNN and LSTM network model, the trained third CNN and LSTM network model, the trained fourth CNN and LSTM network model and the trained fifth CNN and LSTM network model, respectively, to obtain the s-hour mean atmospheric precipitable water univariate prediction value, s-hour mean temperature univariate prediction value, s-hour mean air pressure univariate prediction value, s-hour mean relative humidity univariate prediction value and s-hour rainfall univariate prediction value corresponding to the subsequent annual accumulation time; Step 5022: If the s-hour rainfall corresponding to the current annual accumulation time is greater than or equal to 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the previous i-1 annual accumulation times of the current annual accumulation time, and the s-hour mean atmospheric precipitable water univariate prediction value, s-hour mean temperature univariate prediction value, s-hour mean air pressure univariate prediction value, s-hour mean relative humidity univariate prediction value and s-hour rainfall univariate prediction value corresponding to the subsequent annual accumulation time are input into N BPNN rain mixed models to obtain the s-hour rainfall second prediction value corresponding to N subsequent annual accumulation times; If the s-hour rainfall corresponding to the current annual accumulation time is less than 0.3 mm, the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the current annual accumulation time, the s-hour mean atmospheric precipitable water, s-hour mean temperature, s-hour mean air pressure, s-hour mean relative humidity and s-hour rainfall cumulative value corresponding to the previous i-1 annual accumulation times of the current annual accumulation time, and the s-hour mean atmospheric precipitable water univariate prediction value, s-hour mean temperature univariate prediction value, s-hour mean air pressure univariate prediction value, s-hour mean relative humidity univariate prediction value and s-hour rainfall univariate prediction value corresponding to the subsequent annual accumulation time are input into N BPNN rain-free mixed models to obtain the second prediction value of the s-hour rainfall corresponding to N subsequent annual accumulation times.

Images