A method and system for automatic rainfall estimation based on neural network

Abstract

The present invention proposes a method and system for automatically estimating rainfall based on a neural network, wherein the method comprises the following steps: using an algorithm to extract spatial and temporal correspondence data of radar reflectivity and ground rainfall observation sites; using a neural network algorithm to train the extracted data, update weights, and estimate rainfall; optimizing the neural network algorithm by using a weighted loss function, so that the model improves the ability to estimate moderate rain, heavy rain, and rainstorm; calculating and outputting the weights of each node, and estimating the specific value of rainfall; the method effectively solves the technical problems that the existing empirical formula is derived under multiple assumptions, resulting in low estimation accuracy of rainfall, and even lower estimation accuracy of moderate rain, heavy rain, and rainstorm, meets the requirements of automation, and improves the estimation accuracy of rainfall at the same time.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological quantitative estimation of rainfall, and in particular, relates to a method and system for automatic estimation of rainfall based on a neural network. Background Art

[0002] Rainfall is the most frequent weather condition in meteorology. It not only affects people's daily work and travel, but also has an important impact on industry, agriculture, commerce, and transportation. Excessive rainfall can affect people's normal lives and even cause natural disasters. Therefore, doing a good job in heavy rainfall warning and providing accurate rainfall information in a timely manner for people's lives has become the focus and difficulty of meteorological researchers and scientific researchers.

[0003] At present, the rainfall information is obtained through the empirical formula ZR relationship between radar reflectivity and rainfall. That is, the radar reflectivity factor Z unit: mm 6 / m 3 and rainfall R unit: mm / h,

[0004] Under some reasonable assumptions, the derived ZR relationship is assumed to be as follows:

[0005] 1. Rainfall is evenly distributed and does not change over time;

[0006] 2. The airflow perpendicular to the ground can be ignored;

[0007] 3. Raindrop scattering satisfies Rayleigh scattering

[0008] 4. The falling speed of raindrops is assumed to be in a static atmosphere.

[0009] This type of traditional method is derived under many assumptions and does not take into account factors that affect rainfall, such as air flow, raindrop distribution, air pressure, wind speed, temperature and humidity. Therefore, this method has a large error in rainfall estimation. The empirical formula is relatively rough and cannot accurately estimate rainfall information based on the actual environment. Summary of the invention

[0010] The present invention proposes a method and system for automatically estimating rainfall based on a neural network, which not only improves the overall accuracy of rainfall estimation, but also improves the estimation accuracy of moderate rain, heavy rain and rainstorm.

[0011] The present invention is achieved through the following technical solutions:

[0012] A method for automatic rainfall estimation based on neural network:

[0013] The method specifically comprises the following steps:

[0014] Step S101, extracting the spatial and temporal correspondence data of radar reflectivity and ground rainfall observation stations;

[0015] Step S102, using a neural network algorithm to train the data extracted in step S101, updating the weights of the neural network nodes, and estimating rainfall;

[0016] Step S103, using a weighted loss function to optimize the neural network algorithm described in step S102, so as to improve the model's ability to estimate moderate rain, heavy rain, and rainstorm;

[0017] Step S104: output the estimated rainfall through weight calculation of each network node.

[0018] Further, in step S101,

[0019] Extract radar reflectivity from radar echo data;

[0020] Wind direction, wind speed, humidity, dew point, temperature, air pressure and station altitude parameter data are extracted from ground observation stations.

[0021] Further, in step S102,

[0022] The neural network algorithm is trained on the input radar reflectivity, wind speed, humidity, dew point, temperature, air pressure and station height data, and the output is the estimated rainfall;

[0023] The weights of each node in the network are determined through neural network training, and the influence of each parameter on rainfall is obtained, thereby obtaining the estimated rainfall value.

[0024] Further, in step S103,

[0025] Use weighted loss function to optimize the neural network algorithm so that the neural network algorithm pays more attention to moderate rain, heavy rain and rainstorm during training;

[0026] And by increasing the loss of the neural network algorithm, each network node will pay more attention to the loss value of moderate rain, heavy rain and torrential rain when updating parameters, thereby improving the model's estimation ability for moderate rain, heavy rain and torrential rain.

[0027] Further,

[0028] The calculation formula of the weighted loss function is:

[0029]

[0030] where y i is the actual rainfall, To estimate rainfall, w(y) iThe weight is used to weight the error, increase the influence of moderate rain, heavy rain and rainstorm errors in the back propagation process, and thus improve the estimation accuracy of moderate rain, heavy rain and rainstorm.

[0031] A device for automatically estimating rainfall based on neural network:

[0032] The device includes a data extraction module 301, a neural network module 302 and a calculation output module 303;

[0033] The data extraction module 301 is used to extract the spatial and temporal correspondence data of the radar reflectivity and the ground rainfall observation station;

[0034] The neural network module 302 is used to train the neural network using the data extracted by the data extraction module, update the weights of the neural network nodes, and estimate the rainfall; and optimize the neural network module through the weighted loss function to improve the model's ability to estimate moderate rain, heavy rain, and rainstorm;

[0035] The calculation output module 303 is used for weight calculation of each network node and outputs the estimated rainfall.

[0036] Further,

[0037] The data extraction module 301 is also used for:

[0038] Extract radar reflectivity from radar echo data;

[0039] Wind direction, wind speed, humidity, dew point, temperature, air pressure and station altitude parameter data are extracted from ground observation stations.

[0040] Further,

[0041] The neural network module 302 is also used for:

[0042] The input radar reflectivity, wind speed, humidity, dew point, temperature, air pressure and station height data are trained, and the output is the estimated rainfall;

[0043] The training determines the weights of each node in the network, obtains the impact of each parameter on rainfall, and thus obtains the estimated rainfall value.

[0044] Further,

[0045] The neural network module 302 is also used for:

[0046] Optimize the neural network so that the neural network module pays more attention to moderate rain, heavy rain and rainstorm during training;

[0047] And by increasing the loss of the neural network module 302, each network node pays more attention to the loss value of moderate rain, heavy rain and rainstorm when updating parameters, thereby improving the module's estimation ability for moderate rain, heavy rain and rainstorm.

[0048] Further,

[0049] The weighted loss function calculation formula in the neural network module 302 is:

[0050]

[0051] where y i is the actual rainfall, To estimate rainfall, w(y) i The error is weighted by setting the weight to increase the influence of the error of moderate rain, heavy rain and torrential rain in the back propagation process, thereby improving the estimation accuracy of moderate rain, heavy rain and torrential rain.

[0052] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0053] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the steps of any of the above methods are implemented.

[0054] Beneficial effects of the present invention

[0055] The present invention solves the technical problems that the empirical formulas obtained under multiple assumptions in the prior art cannot accurately estimate the actual rainfall, and do not consider information such as temperature, humidity, wind speed, air pressure, etc. that affect rainfall, resulting in low estimation accuracy of rainfall, and at the same time, the estimation accuracy of moderate rain, heavy rain, and rainstorm is too low. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A flowchart of a method for automatically estimating rainfall based on a neural network according to an embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of the principle of a method for automatic rainfall estimation based on a neural network according to an embodiment of the present invention;

[0058] Figure 3 The figure is a schematic diagram of the structure of a neural network-based automatic rainfall estimation system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] For rainfall estimation, rainfall information is obtained through the empirical formula ZR relationship between radar reflectivity and rainfall. That is, radar reflectivity factor Z unit: mm 6 / m 3 and rainfall R unit: mm / h satisfy the following formula:

[0061] Z=aR b

[0062] Where a and b are empirical constants, a is about 200, b is between 1.5 and 2, and is determined according to various factors such as different times, different regions, and different radar products. The radar reflectivity is dBZ, which needs to be converted into the radar reflectivity factor Z through the following formula.

[0063] dBZ=10lg(Z)

[0064] The derivation process of the ZR empirical formula is as follows.

[0065] The radar reflectivity factor Z is defined as follows:

[0066]

[0067] Where n represents the density of rainwater particles, D represents the diameter of water droplets, and n(D)dD represents the number of particles with diameters between D and D+dD per unit volume.

[0068] Ignoring the vertical changes in ground airflow, the rainfall R can be expressed as:

[0069]

[0070] where n(D) and ν(D) are the density and terminal velocity of the falling particle, respectively.

[0071] From the above formula, we can see that the reflectivity factor and rainfall R are related to the raindrop spectrum n(D)dD, and R is also related to the particle mass and the final falling velocity. The raindrop spectrum can be obtained by the following formula:

[0072] n(D)d(D)=c1Due -^D dD

[0073] Among them, c1, u, and ∧ are constants, and their values ​​are usually determined based on actual data.

[0074] The falling speed of raindrops in a static atmosphere can be calculated by the following formula:

[0075] V(D)=c1-c2e -6D

[0076] Where V(D) is in cm / s, and c1 and c2 are constants.

[0077] Combination Figures 1 to 3 ,The present invention proposes a method and system for automatic rainfall estimation based on neural network.

[0078] A method for automatic rainfall estimation based on neural network:

[0079] The method specifically comprises the following steps:

[0080] Step S101, extracting the spatial and temporal correspondence data of radar reflectivity and ground rainfall observation stations;

[0081] Step S102, using a neural network algorithm to train the data extracted in step S101, updating the weights of the neural network nodes, and estimating rainfall;

[0082] Step S103, using a weighted loss function to optimize the neural network algorithm described in step S102, so as to improve the model's ability to estimate moderate rain, heavy rain, and rainstorm;

[0083] Step S104: output the estimated rainfall through weight calculation of each network node.

[0084] In step S101,

[0085] Automatic data extraction through algorithms:

[0086] Mainly extract radar reflectivity related data from radar echo data;

[0087] Data on 20 rainfall-related parameters, including wind direction, wind speed, humidity, dew point, temperature, air pressure, and station altitude, were extracted from ground observation stations.

[0088] In step S102,

[0089] The neural network algorithm is used to train the input radar reflectivity, wind speed, humidity, dew point, temperature, air pressure and station height data, and the output is the estimated rainfall; it learns the correlation between the input and rainfall,

[0090] The weights of each node in the network are determined through neural network training, and the influence of each parameter on rainfall is obtained, thereby obtaining the estimated rainfall value.

[0091] In step S103,

[0092] Use weighted loss function to optimize the neural network algorithm so that the neural network algorithm pays more attention to moderate rain, heavy rain and rainstorm during training;

[0093] And by increasing the loss of the neural network algorithm, each network node will pay more attention to the loss value of moderate rain, heavy rain and torrential rain when updating parameters, thereby improving the model's estimation ability for moderate rain, heavy rain and torrential rain.

[0094] The calculation formula of the weighted loss function is:

[0095]

[0096] where y i is the actual rainfall, To estimate rainfall, w(y) i The weight is used to weight the error, increase the influence of moderate rain, heavy rain and rainstorm errors in the back propagation process, and thus improve the estimation accuracy of moderate rain, heavy rain and rainstorm.

[0097] A device for automatically estimating rainfall based on neural network:

[0098] The device includes a data extraction module 301, a neural network module 302 and a calculation output module 303;

[0099] The data extraction module 301 is used to extract the spatial and temporal correspondence data of the radar reflectivity and the ground rainfall observation station;

[0100] The neural network module 302 is used to train the neural network using the data extracted by the data extraction module, update the weights of the neural network nodes, and estimate the rainfall; and optimize the neural network module through the weighted loss function to improve the model's ability to estimate moderate rain, heavy rain, and rainstorm;

[0101] The calculation output module 303 is used for weight calculation of each network node and outputs the estimated rainfall.

[0102] The data extraction module 301 is also used for:

[0103] Automatic data extraction through algorithms:

[0104] Mainly extract radar reflectivity related data from radar echo data;

[0105] Data on 20 rainfall-related parameters, including wind direction, wind speed, humidity, dew point, temperature, air pressure, and station altitude, were extracted from ground observation stations.

[0106] Radar echo is the reflectivity data of an aerial area obtained by rotating and scanning the radar station, which is updated every 6 minutes. The ground observation station is a device somewhere on the ground used to detect wind direction, wind speed, humidity, dew point, temperature, air pressure, rainfall and other data, which is updated every 6 minutes. Therefore, it is necessary to extract the radar reflectivity data corresponding to time and space through the longitude and latitude of the ground observation station and the longitude and latitude of each point of the radar echo, as well as the corresponding time. In this way, parameter data for estimating the rainfall at the current moment can be obtained.

[0107] The neural network module 302 is also used for:

[0108] The input data of 20 related parameters, including radar reflectivity, wind speed, humidity, dew point, temperature, air pressure and station height, are trained, and the output is the estimated rainfall;

[0109] The training determines the weights of each node in the network, obtains the impact of each parameter on rainfall, and thus obtains the estimated rainfall value.

[0110] The neural network module 302 is also used for:

[0111] Optimize the neural network so that the neural network module pays more attention to moderate rain, heavy rain and rainstorm during training;

[0112] And by increasing the loss of the neural network module 302, each network node pays more attention to the loss value of moderate rain, heavy rain and rainstorm when updating parameters, thereby improving the module's estimation ability for moderate rain, heavy rain and rainstorm.

[0113] The weighted loss function calculation formula in the neural network module 302 is:

[0114]

[0115] where y i is the actual rainfall, To estimate rainfall, w(y) i The error is weighted by setting the weight to increase the influence of the error of moderate rain, heavy rain and torrential rain in the back propagation process, thereby improving the estimation accuracy of moderate rain, heavy rain and torrential rain.

[0116] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0117] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the steps of any of the above methods are implemented.

[0118] The above is a detailed introduction to the neural network-based rainfall automatic estimation method and system proposed in the present invention, and the principles and implementation methods of the present invention are explained. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for automatic rainfall estimation based on a neural network, characterized in that: The method specifically comprises the following steps: Step S101, extracting the spatial and temporal correspondence data of radar reflectivity and ground rainfall observation stations; Extract radar reflectivity from radar echo data; Extract wind direction, wind speed, humidity, dew point, temperature, air pressure and station height parameter data from ground observation stations; Step S102, using a neural network algorithm to train the data extracted in step S101, updating the weights of the neural network nodes, and estimating rainfall; The neural network algorithm is trained on the input radar reflectivity, wind speed, humidity, dew point, temperature, air pressure and station height data, and the output is the estimated rainfall; Through neural network training, the weights of each node in the network are determined, and the influence of each parameter on rainfall is obtained, thereby obtaining the estimated value of rainfall; Step S103, using a weighted loss function to optimize the neural network algorithm described in step S102, so as to improve the model's ability to estimate moderate rain, heavy rain, and rainstorm; Use weighted loss function to optimize the neural network algorithm so that the neural network algorithm pays more attention to moderate rain, heavy rain and rainstorm during training; And by increasing the loss of the neural network algorithm, each network node pays more attention to the loss value of moderate rain, heavy rain and rainstorm when updating parameters, thereby improving the model's estimation ability for moderate rain, heavy rain and rainstorm; The calculation formula of the weighted loss function is: where y i is the actual rainfall, To estimate rainfall, w(y) i The weight is used to increase the influence of the errors of moderate rain, heavy rain and rainstorm in the back propagation process by weighting the errors, thereby improving the estimation accuracy of moderate rain, heavy rain and rainstorm; Step S104: output the estimated rainfall through weight calculation of each network node.

2. An estimation device for executing the neural network-based rainfall automatic estimation method according to claim 1, characterized in that: The device comprises a data extraction module (301), a neural network module (302) and a calculation output module (303); A data extraction module (301) is used to extract the temporal and spatial correspondence data of radar reflectivity and ground rainfall observation sites; A neural network module (302) is used to train a neural network using the data extracted by the data extraction module, update the weights of the neural network nodes, and estimate the rainfall; and optimize the neural network module through a weighted loss function to improve the model's ability to estimate moderate rain, heavy rain, and rainstorm; The calculation output module (303) is used for weight calculation of each network node and outputs the estimated rainfall.

3. The device according to claim 2, characterized in that: The data extraction module (301) is also used for: Extract radar reflectivity from radar echo data; Wind direction, wind speed, humidity, dew point, temperature, air pressure and station altitude parameter data are extracted from ground observation stations.

4. The device according to claim 3, characterized in that: The neural network module (302) is also used for: The input radar reflectivity, wind speed, humidity, dew point, temperature, air pressure and station height data are trained, and the output is the estimated rainfall; The training determines the weight of each node in the network, obtains the influence of each parameter on rainfall, and thus obtains the estimated value of rainfall.

5. The device according to claim 4, characterized in that: The neural network module (302) is also used for: Optimize the neural network so that the neural network module pays more attention to moderate rain, heavy rain and rainstorm during training; By increasing the loss of the neural network module (302), each network node pays more attention to the loss value of moderate rain, heavy rain and rainstorm when updating parameters, thereby improving the module's estimation ability for moderate rain, heavy rain and rainstorm.

6. The device according to claim 5, characterized in that: The weighted loss function calculation formula in the neural network module (302) is: where y i is the actual rainfall, To estimate rainfall, w(y) i The error is weighted by setting the weight, which increases the influence of the error of moderate rain, heavy rain and torrential rain in the back propagation process, thereby improving the estimation accuracy of moderate rain, heavy rain and torrential rain.

Images