CN-121978779-A - Rainfall magnitude integrated forecasting method based on Bayesian formula and evidence theory

CN121978779ACN 121978779 ACN121978779 ACN 121978779ACN-121978779-A

Abstract

The invention relates to a weather forecast method, in particular to a rainfall magnitude integrated forecast method based on a Bayesian formula and an evidence theory. The rainfall magnitude integrated forecasting method based on the Bayesian formula and the evidence theory provides a firm and reliable technical support for further improving the rainfall forecasting precision and promoting the application of the rainfall magnitude integrated forecasting method in the early warning of the flood and drought disasters and the scheduling of water resources. The technical scheme is that the rainfall magnitude integrated forecasting method based on the Bayesian formula and the evidence theory comprises the following steps of 1, rainfall forecasting and actual measurement data collection, 2, rainfall magnitude determination, 3, bayesian formula distribution determination and calculation, 4, evidence fusion and 5, integrated forecasting evaluation.

Inventors

Cai Chenkai
WANG JING
JIN YITONG
HUANG YUQING

Assignees

浙江同济科技职业学院

Dates

Publication Date: 20260505
Application Date: 20260107

Claims (3)

1. A rainfall magnitude integrated forecasting method based on a Bayesian formula and an evidence theory comprises the following steps: The method comprises the steps of step 1, collecting rainfall forecast and actual measurement data, namely, investigating and collecting rainfall forecast data issued by a plurality of meteorological institutions in a certain historical period of a research area and rainfall actual measurement data corresponding to the rainfall forecast data; Determining rainfall orders, namely formulating corresponding rainfall order dividing standards, and calculating the orders of rainfall forecast data of different mechanisms and corresponding rainfall actual measurement values collected in the last step; step 3, bayesian formula distribution determination and calculation, namely calculating the occurrence probability of different actually measured rainfall orders by using a Bayesian formula under the condition of the rainfall forecast orders issued by a certain meteorological institution; specifically, assume a certain weather forecast The random variable of the rainfall forecast magnitude of the period is The corresponding measured magnitude random variable is According to the Bayesian formula, there are: (1) wherein: Is shown in When it occurs Conditional probability of occurrence, i.e., posterior distribution; Representation of The probability distribution, i.e. the prior distribution, of occurrence, in particular the uniform distribution; Is shown in When it occurs The occurrence conditional probability, namely likelihood function, is estimated according to different actually measured rainfall orders and corresponding rainfall forecast orders in the collected historical data; Representation of The probability distribution of occurrence is used for normalization, and is estimated specifically according to the collected historical rainfall forecast magnitude distribution situation; Step 4, evidence fusion, namely taking the probability calculated by the Bayesian formula as evidence, respectively determining the credibility allocation functions of all the mechanisms through a formula 6, and carrying out evidence fusion calculation through a formula 7 to finally obtain the possibility of occurrence of different precipitation levels; confidence allocation function The method comprises the following steps: (6) (7) wherein: Fusing operators for evidence; Distributing a function for the credibility of the event after fusion; Evidence respectively The corresponding confidence allocation function, K is a normalization constant, calculated by equation 8: (8)。
2. the rainfall magnitude integrated forecasting method based on the Bayesian formula and the evidence theory according to claim 1, wherein the specific calculation formula in the step 3 is as follows: Rainfall is divided into Magnitude, and is written as I.e. And The range of the values of (a) is all ; (2) (3) (4) In the formula, For the total number of samples, i.e. the number of rainfall prediction orders or the measured magnitude order, the two are in one-to-one correspondence; The rainfall forecast magnitude is the first An order of magnitude; Is the actual measured rainfall level of Magnitude of rainfall forecast is the first magnitude An order of magnitude.
3. The rainfall magnitude integration forecasting method based on the Bayesian formula and the evidence theory according to claim 2, wherein the credibility allocation function determining method of the step 4 is as follows: Assuming that rainfall is divided into two orders, namely no rainfall and no rainfall, and when a certain forecasting mechanism forecasts that the rainfall is in existence, the posterior distribution of the actually measured rainfall orders is in existence and no rainfall is respectively And According to evidence theory, the frame is identified at this time Comprises two exclusive events, namely the measured rainfall level is rainy And the measured rainfall level is rain-free ; Then the confidence allocation function The method comprises the following steps: (5) according to this, if the precipitation is divided into Orders of magnitude, respectively denoted as Then the confidence allocation function The method comprises the following steps: (6)。

Description

Rainfall magnitude integrated forecasting method based on Bayesian formula and evidence theory Technical Field The invention relates to a weather forecast method, in particular to a rainfall magnitude integrated forecast method based on a Bayesian formula and an evidence theory. Background Rainfall forecast is one of the most important products of numerical weather forecast, and is also an important reference for early warning of flood and drought disasters and scheduling of water resources. At present, the precision of rainfall forecast still hardly meets the actual demands of the application of flood and drought disaster forecast and water resource scheduling in the hydrology field, and the effect of rainfall forecast in the application of hydrology is severely restricted. For this reason, a weather student proposes to combine rainfall forecasts of different modes or different initial values of the same mode, called integration forecast. Since the idea of integrated forecasting was proposed, more and more statistical regression analysis methods are applied to integrated forecasting of different meteorological variables from simple equal weight set averaging to complex statistical learning methods. Integrated forecasting tends to provide a better forecasting result than individual members. According to the integrated forecasting concept, it essentially willRainfall forecast for different institutions or modesBy a linear or non-linear functionCombined into a new forecast value. Thus, the forecasting results of each member can be integrally forecastedAnd (3) regarding the evidence as evidence from different sources, and fusing the forecast results of each member by a multi-source evidence fusion method in the evidence theory, so as to generate an integrated forecast result. Theoretically, it is feasible to implement rainfall integration forecast by adopting evidence theory, but there are two problems to be solved. First, evidence theory requires defining an identification framework, i.e., a collection of objects or objects under investigation, defined as a non-empty collectionWhich comprisesTwo mutually exclusive events. For example, the number of the cells to be processed,By two mutually exclusive eventsAndThe composition is that there isThen have a power set. For rainfall forecast, the value can beAny value in the interval range has infinite numbers of mutual exclusion events, so that the evidence theory is difficult to calculate due to infinite numbers of power sets. Another problem is how to make the confidence allocation. Disclosure of Invention The invention aims to overcome the defects of the technical background, provides a rainfall magnitude integrated forecasting method based on a Bayesian formula and an evidence theory, and provides a firm and reliable technical support for further improving the rainfall forecasting precision and promoting the application of the rainfall magnitude integrated forecasting method in early warning of a flood and drought disaster and scheduling of water resources. The technical scheme provided by the invention is that the rainfall magnitude integrated forecasting method based on a Bayesian formula and an evidence theory comprises the following steps: The method comprises the steps of step 1, collecting rainfall forecast and actual measurement data, namely, investigating and collecting rainfall forecast data issued by a plurality of meteorological institutions in a certain historical period of a research area and rainfall actual measurement data corresponding to the rainfall forecast data; Determining rainfall orders, namely setting corresponding rainfall order dividing standards by taking national standard precipitation grade (GB/T28592-2012) as a reference, and calculating the orders of rainfall forecast data of different mechanisms and corresponding rainfall actual measurement values collected in the last step; step 3, bayesian formula distribution determination and calculation, namely calculating the occurrence probability of different actually measured rainfall orders through a Bayesian formula under the condition of the rainfall forecast orders issued by a certain meteorological institution, and specifically, assuming that a certain weather forecast is the first The random variable of the rainfall forecast magnitude of the period isThe corresponding measured magnitude random variable is,AndAre all discrete random variables, and according to a Bayesian formula, the method comprises the following steps: (1) In the formula, Is shown inWhen it occursConditional probability of occurrence, i.e., posterior distribution; Representation of The probability distribution of occurrence, i.e., a priori distribution; Is shown in When it occursThe conditional probability of occurrence, i.e., likelihood function; Representation of The probability distribution of occurrence is used for normalization. According to the above formula, it is necessary to first determine,And. Wherein, t