High-precision rainfall forecasting method for medium and small drainage basins

Abstract

The invention relates to the technical field of weather forecasting, in particular to a high-precision rainfall forecasting method for small and medium-sized watersheds, which comprises the following steps of: completing a Chinese historical lattice point multi-meteorological-element database through an own data processing technology; an integrated WRF mode and an artificial intelligence machine learning technology are adopted to research and develop an own small-scale lattice weather forecasting system, the forecasting precision can reach 500 meters, and the precision problem of existing weather forecasting is solved. According to the method, a high-resolution weather prediction machine learning model is established according to the conditions of regions, weather and the like, a model trained by historical data is used for predicting the local effect, the weather prediction accuracy is improved, multiple weather early warning means are adopted, and reliable and stable arrival of information is ensured.

Description

Technical Field

[0001] The invention relates to the technical field of weather forecasting, and in particular to a method for high-precision rainfall forecasting in small and medium-sized river basins. Background Art

[0002] Currently, artificial intelligence technology and big data are in the stage of large-scale research and application. Introducing these two technologies into high-resolution weather forecasting has great technical advantages. Once an artificial intelligence machine learning model is established, it is usually faster than a complete numerical simulation project. A model emulation method can be used to establish a statistical model that is very close to the numerical model. This method has been applied to climate sensitivity research. Some scenarios in weather forecasting need to be implemented through atmospheric physical models, but it is impossible or unreasonable to use models in some scenarios. In these scenarios, using statistical approximation is the best choice, and model emulation will have a good effect in forecasting business. On the other hand, we can collect and analyze meteorological, geographical and environmental data to form big data, and combine prediction models and statistical models to further improve the prediction accuracy. With the continuous improvement of computing power and model algorithms, the application value of prediction results will be further improved. Therefore, the use of artificial intelligence technology and big data in weather forecasting is a hot topic in current research.

[0003] Precipitation is one of the most important factors affecting the operation of hydropower enterprises. Accurate rainfall forecast can scientifically guide the production scheduling of hydropower enterprises, which is of great significance. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the present invention provides a method for high-precision rainfall forecasting in small and medium-sized watersheds, which improves the accuracy of rainfall forecasting through optimized evaluation model design. The specific technical solution is as follows: The present invention has completed the Chinese historical gridded multi-meteorological element database through its own data processing technology. In combination with the characteristics of China's climate and terrain, the present invention has developed its own small-scale gridded weather forecast system using the ensemble WRF model and artificial intelligence machine learning technology. The forecast accuracy can reach 500 meters, which makes up for the accuracy problem of existing meteorological forecasts. At the same time, the concept of surface rainfall is proposed, which provides a solution specifically for high-precision forecasts of small and medium-sized basins.

[0007] 1. Core algorithms and technologies: (1) Calculation of multiple sets of data based on WRF WRF (Weather Research Forecast) is a numerical meteorological forecasting model for high-resolution mesoscale meteorological research and forecasting numerical models and assimilation systems. WRF is mainly used for simulation and forecasting of 0-10km grid spacing on the ground. While calculating, it combines the Geographic Information System (GIS) to correct the forecast value according to the difference in topography. At present, we have calculated 4 groups of result data and then selected the best result.

[0008] (2) Application of artificial intelligence technology in the optimization process Artificial intelligence technology is used to filter multiple sets of data calculated by the WRF model to obtain the optimal data. The result data is compared with the actual data, and then the algorithm model parameters of different time periods are optimized to form a new algorithm model.

[0009] (3) AI downscaling optimization in combination with monitoring stations Monitoring stations and ice sensors are deployed in key areas to obtain normal and extreme meteorological data, which are input into AI models (owned) for small-scale calculations. The preliminary results are optimized to obtain optimized results.

[0010] 2. Precipitation forecast technology for small and medium-sized watersheds A single river basin is divided into several sub-basins according to the location of water conservancy regulation hubs such as cascade power stations or dams and the distribution of river streams and tributaries. Each river basin is composed of several grids with an accuracy of 1 to 3 km. Each grid carries a rainfall forecast value for a period of time in the future. The average rainfall in an area can be obtained by adding up all the grid data in the area and dividing it by the number of grids in the area, which is also called surface rainfall.

[0011] 3. Terms and terms related to the present invention Climate forecast: also known as climate prediction or climate outlook, it is the outlook for the future climate of a certain area. It uses the laws of climate evolution to infer the climate conditions in the next few years or even decades or longer. Its main basis is long-term climate records, solar activity, geomagnetism, sea ice, tree rings, historical records and archaeology, etc., and also considers the impact of human activities. Climate forecast is suitable for medium- and long-term forecasts, that is, weather forecasts for more than 15 days.

[0012] Briefing: Through the description of rainfall in the basin, time and region, users can quickly understand the rainfall situation.

[0013] Numerical display: Display detailed numerical values ​​of a certain area (rainfall data for 1h, 6h, 12h and 24h).

[0014] Rain Map: View a visual map of rainfall for each main stream and tributaries in the basin.

[0015] Numerical weather forecast: Numerical weather forecast refers to a method of predicting the state of atmospheric movement and weather phenomena in a certain period of time in the future by solving the equations of fluid mechanics and thermodynamics that describe the evolution of weather based on the actual atmospheric conditions under certain initial and boundary conditions through large computers. We generally use the numerical weather forecast model to deal with weather forecasts of no more than 15 days, which we also call meteorological forecasts. Meteorological forecasts are more detailed and more accurate than climate forecasts.

[0016] Instruction Manual Figure 1 It is the algorithm logic diagram of the present invention.

[0017] Beneficial Effects Compared with the prior art, the technical effects of the present invention are embodied in: (1) Establish a high-resolution weather forecast machine learning model based on regional and meteorological conditions. Use a model trained with historical data (a machine learning method) to predict local effects, and then establish a mapping relationship between large-scale model predictions and small-scale effects to improve weather forecast accuracy.

[0018] (2) Since the prediction is essentially based on historical data, the computational complexity of statistical methods is several orders of magnitude smaller than that of numerical methods, so the cost of each prediction is lower.

[0019] (3) Meteorological data mining based on big data, through the collection and analysis of meteorological, geographical and environmental data, combined with the above-mentioned prediction models and statistical models to further improve the prediction accuracy.

[0020] (4) Using cloud computing technology, meteorological data is calculated and counted in the cloud data center to improve safety and efficiency.

[0021] Multi-terminal coverage can provide multi-terminal coverage including PC, tablets and mobile phones, and multiple weather warning methods to ensure that information arrives reliably and stably. DETAILED DESCRIPTION

[0022] The technical solution of the present invention is further defined below in conjunction with specific implementation methods, but the scope of protection required is not limited to the description.

[0023] Example 1 This embodiment uses its own data processing technology to complete the Chinese historical gridded multi-meteorological element database. In combination with China's climate and terrain characteristics, the ensemble WRF model and artificial intelligence machine learning technology are used to develop a small-scale gridded weather forecast system, with a forecast accuracy of up to 500 meters, which makes up for the accuracy problem of existing weather forecasts. At the same time, the concept of surface rainfall is proposed, which provides a solution specifically for high-precision forecasts of small and medium-sized basins.

[0024] 1. Core algorithms and technologies: (1) Calculation of multiple sets of data based on WRF WRF (Weather Research Forecast) is a numerical meteorological forecasting model for high-resolution mesoscale meteorological research and forecasting numerical models and assimilation systems. WRF is mainly used for simulation and forecasting of 0-10km grid spacing on the ground. While calculating, it combines the Geographic Information System (GIS) to correct the forecast value according to the difference in topography. At present, we have calculated 4 groups of result data and then selected the best result.

[0025] (2) Application of artificial intelligence technology in the optimization process Artificial intelligence technology is used to filter multiple sets of data calculated by the WRF model to obtain the optimal data. The result data is compared with the actual data, and then the algorithm model parameters of different time periods are optimized to form a new algorithm model.

[0026] (3) AI downscaling optimization in combination with monitoring stations Monitoring stations and ice sensors are deployed in key areas to obtain normal and extreme weather data, input into AI models (own) for small-scale calculations, and optimize the preliminary results to obtain optimized results. See the attached for specific optimization logic. Figure 1 .

[0027] Based on the global model and reanalysis data EC, NCEP, combined with observation data from observation points such as sounding, basic stations, satellites, and high-precision terrain data, objective analysis, quality control inspection, and four-dimensional assimilation data are formed to extract the mesoscale meteorological model, ocean model, and environmental model architecture, and obtain the dynamic model grid data, which is further optimized into patented algorithms for comparison, optimization, and fusion to obtain refined forecasts and historical grid data, forming the final patented climate forecast product service. Local observation data comes from multiple ground stations, and after quality control inspection, they are compared, optimized, and fused.

[0028] The surface rainfall forecast for small and medium-sized river basins described in this embodiment is that a single river basin is divided into several sub-basins according to the location of water conservancy regulation hubs such as cascade power stations or dams and the distribution of river flows and tributaries. Each river basin is composed of several grids with an accuracy of 1 to 3 km. Each grid carries a rainfall forecast value for a period of time in the future. The average rainfall in an area can be obtained by adding up all the grid data in the area and dividing it by the number of grids in the area, which is also called surface rainfall.

[0029] The climate forecast in this embodiment is a prospect for the future climate of a certain area. The climate evolution law is used to infer the climate conditions in the next few years or even decades or longer. Its main basis is long-term climate records, solar activity, geomagnetism, sea ice, tree rings, historical records and archaeology, etc. In addition, the impact of human activities is also considered. Climate forecast is suitable for medium- and long-term forecasts, that is, weather forecasts for more than 15 days.

[0030] This implementation of the climate forecast uses a brief report as one of the presentation methods: through the description of rainfall in the basin, time, and region, users can quickly understand the rainfall situation.

[0031] This implementation of the climate forecast uses numerical values ​​as one of the presentation methods: numerical display is to display detailed numerical values ​​of a certain area (rainfall data for 1h, 6h, 12h and 24h).

[0032] This implementation of the climate forecast uses a rain map as one of the presentation methods: view a visual map of the rainfall in each main stream and tributary of the basin.

[0033] The numerical weather forecast of this embodiment refers to a method of predicting the atmospheric motion state and weather phenomena for a certain period of time in the future by solving the equations of fluid mechanics and thermodynamics describing the weather evolution process based on the actual atmospheric conditions under certain initial and boundary conditions through numerical calculations by large computers. This embodiment adopts the mode of numerical weather forecast to process weather forecasts for no more than 15 days. Meteorological forecasts are more detailed and more accurate than climate forecasts.

Claims

1. A method for high-precision rainfall forecasting in small and medium-sized watersheds, characterized in that: include: (1) Calculation of multiple sets of data based on WRF Combined with the Geographic Information System (GIS), WRF is used to modify the forecast values ​​according to the differences in topography and landforms to calculate multiple sets of forecast values, and then the best result is selected; (2) Artificial intelligence technology in optimizing models Apply artificial intelligence technology to filter multiple sets of data calculated by the WRF model to obtain the optimal data, compare the result data with the actual data, and then optimize the algorithm model parameters in different time periods to form a new algorithm model; (3) AI downscaling optimization combined with monitoring stations Monitoring stations and icing sensors are deployed in key areas to obtain normal and extreme meteorological data. The data are input into the algorithm model in step (2) for small-scale calculations, and the preliminary results are optimized to obtain optimized results.

2. The method for high-precision rainfall forecasting in small and medium-sized watersheds according to claim 1 is characterized in that: The data collection in the data calculation includes: a single basin is divided into several sub-basins according to the location of the water conservancy regulation hub and the distribution of river flows and tributaries. Each basin is composed of several grids with an accuracy of 1 to 3 km. Each grid carries a rainfall forecast value for a period of time in the future. All grid data in an area are added together and then divided by the number of grids in the area to obtain the average rainfall in the area, that is, the surface rainfall.

3. The method for high-precision rainfall forecasting in small and medium-sized watersheds according to claim 2 is characterized in that: The water regulation hub is any one of a cascade power station or a dam.

4. The method for high-precision rainfall forecasting in small and medium-sized watersheds according to claim 1 is characterized in that: The climate forecast is the final reference result obtained after evaluating long-term climate records, solar activity, geomagnetism, sea ice, tree rings, historical records, archaeological materials, and the impact of human activities.

5. The method for high-precision rainfall forecasting in small and medium-sized watersheds according to claim 1 is characterized in that: The optimized results are presented in the form of brief reports, which specifically describe the rainfall in the watershed, time, and region, so that users can quickly understand the rainfall situation.

6. The method for high-precision rainfall forecasting in small and medium-sized watersheds according to claim 5 is characterized in that: The briefing method includes numerical display, specifically displaying detailed numerical values ​​of a certain area, including rainfall data for 1h, 6h, 12h and 24h.

7. The method for high-precision rainfall forecasting in small and medium-sized watersheds according to claim 1 is characterized in that: The briefing method includes the presentation of a rain map, forming an intuitive picture of the rainfall of each main stream and tributary in the basin.

Images