A multi-source rainfall forecast fusion method based on typhoon similarity

Abstract

The invention discloses a multi-source rainfall forecast fusion method based on typhoon similarity, which relates to the field of rainfall forecasting. The method selects key factors that affect the formation and development of typhoons, and obtains multi-dimensional vectors corresponding to each typhoon path point for the current typhoon and each historical typhoon, thereby ensuring a comprehensive description of typhoon characteristics; by constructing a historical typhoon data set, and based on the one-dimensional vectors corresponding to the set and each forecasting agency, the comprehensive correction coefficient of each forecasting agency is obtained, thereby effectively correcting the system errors of each forecasting agency; at the same time, the comprehensive correction coefficient is used to calculate the optimal weight coefficient and standard deviation, and the latest future rainfall forecast values ​​of each forecasting agency are obtained by obtaining the comprehensive correction coefficient for correction, thereby obtaining the forecast correction values ​​corresponding to each forecasting agency, and combining the optimal weight coefficients, standard deviations and forecast correction values ​​of each forecasting agency to obtain the forecast fusion values ​​and the corresponding variance estimation values, thereby greatly improving the prediction accuracy.

Description

Technical Field

[0001] The invention relates to the field of rainfall forecasting, and in particular to a multi-source rainfall forecast fusion method based on typhoon similarity. Background Art

[0002] During typhoons, rainfall is an active factor in the hydrological cycle and the most common factor causing flood disasters. Its data accuracy and rainfall type are of great significance for flood forecasting and flood prevention and disaster reduction. However, the current rainfall forecast shows significant spatiotemporal variability due to the influence of factors such as basin topography, geographical location, and atmospheric movement, which makes the future rainfall forecast data provided by single-source agencies obviously insufficient. First, the future rainfall and rainfall types of single-source agencies are usually forecasted in a time series with a step length of 3 hours, and it is impossible to provide hourly rainfall time series, resulting in insufficient refinement of rainfall type distribution, which is difficult to meet the needs of flood forecast models for high-resolution input data. Secondly, the future rainfall forecast accuracy of single-source agencies is insufficient. Due to technical limitations and uncertainties in external conditions, these forecast values ​​are often too large to accurately reflect the actual rainfall conditions, thus affecting the reliability and accuracy of flood forecast models. Therefore, in order to improve the accuracy and reliability of flood forecasts, a method that can fuse multi-source rainfall forecast data is urgently needed to overcome the limitations of single-source forecasts and improve the data accuracy of rainfall forecasts. Summary of the invention

[0003] In order to achieve the fusion of multi-source rainfall forecast data and improve the data accuracy of rainfall forecast, the present invention proposes a multi-source rainfall forecast fusion method based on typhoon similarity, including:

[0004] Select the key factors that affect the formation and development of typhoons, obtain the multidimensional vectors formed by the key factors corresponding to each typhoon path point in the typhoon path for the current typhoon and each historical typhoon, and construct the typhoon factor time series corresponding to each typhoon based on the multidimensional vectors corresponding to each typhoon path point;

[0005] The current typhoon is combined with each historical typhoon one-to-one, and for each combination, the corresponding typhoon factor time series is used to calculate the similarity distance between the current typhoon and the corresponding historical typhoon; historical typhoons with similarity distances less than a preset threshold are selected to form a historical typhoon data set;

[0006] A one-dimensional vector is constructed based on the measured rainfall corresponding to each typhoon event in the historical typhoon data set, and a corresponding one-dimensional vector is constructed for each forecasting agency based on the rainfall forecast values ​​of multiple forecasting agencies corresponding to each typhoon event;

[0007] Through the historical typhoon data set and the one-dimensional vector corresponding to each forecasting agency, the comprehensive correction coefficient corresponding to each forecasting agency is obtained;

[0008] A dynamic Bayesian optimization model is constructed through the historical typhoon data set, the one-dimensional vector corresponding to each forecasting agency, and the comprehensive correction coefficient corresponding to each forecasting agency. Based on the dynamic Bayesian optimization model, the optimal weight coefficient and standard deviation corresponding to each forecasting agency are obtained.

[0009] The latest future rainfall forecast values ​​from each forecasting agency are obtained, and the future rainfall forecast values ​​are corrected by the comprehensive correction coefficient to obtain the corresponding forecast correction values ​​of each forecasting agency. The forecast fusion values ​​and their corresponding variance estimates are obtained by using the corresponding optimal weight coefficients, standard deviations and forecast correction values ​​of each forecasting agency.

[0010] Furthermore, the key factors include: typhoon wind speed, atmospheric pressure, typhoon moving speed, seasonal factor value, and central longitude and central latitude of the typhoon path point.

[0011] Furthermore, for each combination, using the corresponding typhoon factor time series, calculating the similarity distance between the current typhoon and the historical typhoon specifically includes:

[0012] The points in the first sequence and the second sequence, i.e., the multidimensional vectors, are combined in pairs, and the multidimensional Euclidean distance of each point pair is calculated; the first sequence represents the typhoon factor time series of the current typhoon in the combination; the second sequence represents the typhoon factor time series corresponding to the historical typhoon in the corresponding combination;

[0013] Construct a distance matrix based on the multidimensional Euclidean distances of all point pairs;

[0014] Based on the distance matrix, a dynamic time warping algorithm is used to obtain a cumulative distance matrix M for recording the cumulative distance between the first sequence and the second sequence;

[0015] The similarity distance between the first sequence and the second sequence is calculated by accumulating the last element in the distance matrix M.

[0016] Furthermore, the formula expression of the first sequence is:

[0017] ;in, ;

[0018] In the formula, i=1…n, n represents the number of points in the first sequence; Represents the multidimensional vector corresponding to the i-th point in the first sequence; Respectively represent the central longitude, central latitude, typhoon wind speed, atmospheric pressure, typhoon moving speed and seasonal factor value of the i-th point in the first sequence;

[0019] The formula expression of the second sequence is:

[0020] ;in, ;

[0021] In the formula, j=1…m, m represents the number of points in the second sequence; Represents the multidimensional vector corresponding to the j-th point in the second sequence; Respectively represent the central longitude, central latitude, typhoon wind speed, atmospheric pressure, typhoon moving speed and seasonal factor value of the jth point in the second sequence;

[0022] The calculation formula for obtaining the multidimensional Euclidean distance of each point pair is:

[0023] ;

[0024] In the formula, Indicate point and The multidimensional Euclidean distance between ; , , , , , The sum of the addition is 1, which represents the preset weight values ​​corresponding to the central longitude, central latitude, typhoon wind speed, atmospheric pressure, typhoon moving speed and seasonal factor value respectively; It represents the direct distance between the i-th point in the first sequence and the j-th point in the second sequence, that is, the multidimensional Euclidean distance.

[0025] Furthermore, the method of obtaining a cumulative distance matrix M for recording the cumulative distance between the first sequence and the second sequence by using a dynamic time warping algorithm based on the distance matrix specifically includes:

[0026] Set the boundary conditions for obtaining the cumulative distance matrix M;

[0027] Using the boundary conditions and the elements in the distance matrix, each element in the cumulative distance matrix M is gradually solved in a recursive manner to obtain the cumulative distance matrix M; the solution formula is:

[0028] ; and ;

[0029] In the formula, represents the minimum cumulative distance among all possible paths from the first point of the first sequence to the i-th point and from the first point of the second sequence to the j-th point;

[0030] Indicates diagonal movement, that is, from the first sequence The first point to the ith point and the second point from the The minimum cumulative distance among all possible paths between the point to the jth point;

[0031] Indicates horizontal movement, that is, from the first sequence The minimum cumulative distance from the i-th point to the second sequence among all possible paths that remain at the j-th point;

[0032] represents vertical movement, that is, from the first sequence to the second sequence, The minimum cumulative distance among all possible paths from point j to point j.

[0033] Furthermore, the boundary conditions include:

[0034] , indicating that the cumulative distance between the first sequence and the starting point of the second sequence is zero;

[0035] ; In the formula, , represents the cumulative distance when aligning the first i points of the first sequence with the first point of the second sequence;

[0036] ; In the formula, , represents the cumulative distance when aligning the first point of the first sequence with the first j points of the second sequence.

[0037] Furthermore, the similarity distance between the first sequence and the second sequence is calculated by accumulating the last element in the distance matrix M, and the calculation formula is:

[0038] In the formula, The last element in the cumulative distance matrix M represents the minimum cumulative distance of the global optimal matching path between the first sequence and the second sequence; Represent the number of points in the first and second sequences respectively; Represents the similarity distance between the first sequence and the second sequence.

[0039] Furthermore, the method of obtaining the comprehensive correction coefficient corresponding to each forecasting agency through the historical typhoon data set and the one-dimensional vector corresponding to each forecasting agency specifically includes:

[0040] For each element in the measured one-dimensional vector, the ratio between the element and the corresponding element in the one-dimensional vector corresponding to each forecasting agency is obtained to obtain a correction coefficient set corresponding to each forecasting agency; the measured one-dimensional vector represents a one-dimensional vector corresponding to the historical typhoon data set;

[0041] The correction coefficient sets corresponding to each forecasting agency are averaged to obtain the comprehensive correction coefficients corresponding to each forecasting agency.

[0042] Furthermore, the dynamic Bayesian optimization model is constructed by using the historical typhoon data set and the one-dimensional vector corresponding to each forecasting agency and the comprehensive correction coefficient corresponding to each forecasting agency, and the optimal weight coefficient and standard deviation corresponding to each forecasting agency are obtained based on the dynamic Bayesian optimization model, specifically including:

[0043] Construct a posterior probability distribution; the calculation formula of the posterior probability distribution is:

[0044] ;

[0045] In the formula, It represents the forecast fusion value obtained by fusing the rainfall forecast values ​​of various forecast agencies. represents the measured one-dimensional vector, s represents the number of forecasting agencies, represents the weight coefficient of the i-th forecasting agency, represents the comprehensive correction coefficient of the i-th forecast agency, represents the one-dimensional vector of the i-th forecasting agency, Represents the posterior probability distribution of the measured one-dimensional vector and the corresponding one-dimensional vector of different forecast agencies; represents the posterior probability distribution;

[0046] set up Normal distribution: ; In the formula, Indicates the number of forecast agencies in the i-th The standard deviation of

[0047] Construct the calculation formula for the deterministic estimate:

[0048] ;

[0049] In the formula, , represents the forecast expectation value of the i-th forecast agency; represents a deterministic estimate;

[0050] Construct a prediction variance calculation formula to reflect the numerical uncertainty of forecast fusion:

[0051] ;

[0052] In the formula, represents the forecast variance reflecting the numerical uncertainty of forecast fusion;

[0053] Construct the log-likelihood function using the maximum likelihood estimation method:

[0054] ;

[0055] Through the calculation formula corresponding to the posterior probability distribution, deterministic estimate, prediction variance and log-likelihood function, and The dynamic Bayesian optimization model formed by the formula expression obeying the normal distribution is used to obtain For each parameter in , the optimal weight coefficient and standard deviation corresponding to each forecasting agency are obtained.

[0056] Further, the method of obtaining the latest future rainfall forecast values ​​of each forecast agency, and correcting the future rainfall forecast values ​​by a comprehensive correction coefficient to obtain the forecast correction values ​​corresponding to each forecast agency, and obtaining the forecast fusion values ​​and their corresponding variance estimation values ​​by using the optimal weight coefficients, standard deviations and forecast correction values ​​corresponding to each forecast agency, specifically includes:

[0057] The latest future rainfall forecast values ​​of each forecasting agency are obtained, and the corresponding future rainfall forecast values ​​are corrected by the comprehensive correction coefficients of each forecasting agency; the correction formula is:

[0058] ; In the formula, represents the future rainfall forecast value of the i-th forecast agency, represents the forecast correction value corresponding to the i-th forecast agency;

[0059] According to the forecast correction values ​​and optimal weight coefficients corresponding to each forecast agency, the forecast fusion value is calculated using the following formula: ; In the formula, and ;

[0060] Based on the calculated forecast fusion value, the variance estimation value used to reflect the uncertainty of the forecast fusion value is obtained. The formula is:

[0061] , where Represents the variance estimate.

[0062] Compared with the prior art, the present invention has at least the following beneficial effects:

[0063] (1) The present invention selects key factors that affect the formation and development of typhoons, and obtains multidimensional vectors corresponding to each typhoon path point for the current typhoon and each historical typhoon; this method ensures a comprehensive description of typhoon characteristics, improves the refinement of typhoon paths and factor time series, and provides high-quality basic data for subsequent calculations of similar distances; by constructing a historical typhoon data set, and based on the one-dimensional vectors corresponding to the set and each forecasting agency, the present invention obtains the comprehensive correction coefficient of each forecasting agency, effectively corrects the systematic errors of each forecasting agency, and makes future rainfall forecasts closer to reality; on this basis, the present invention further uses the comprehensive correction coefficient to obtain the comprehensive correction coefficient of each forecasting agency. The optimal weight coefficient and standard deviation are calculated by combining the correction coefficient, which significantly improves the calculation accuracy; finally, by obtaining the latest future rainfall forecast values ​​of each forecast agency and correcting them using the comprehensive correction coefficient, the forecast correction values ​​corresponding to each forecast agency are obtained. Combining the optimal weight coefficient, standard deviation and forecast correction value of each forecast agency, the present invention obtains the forecast fusion value and its corresponding variance estimation value; this process not only realizes the targeted prediction of the latest future rainfall forecast values ​​(that is, targeted prediction of the current typhoon through a similar historical typhoon data set), but also greatly improves the prediction accuracy, providing a reliable basis for flood forecasting and flood prevention and disaster reduction;

[0064] (2) The present invention combines the current typhoon with each historical typhoon in a one-to-one manner, and uses the corresponding typhoon factor time series to calculate the similarity distance. By selecting historical typhoons whose similarity distance is less than a preset threshold, a historical typhoon data set is formed, which ensures that the historical typhoons used for reference have a high degree of similarity with the current typhoon, thereby improving the accuracy of the forecast;

[0065] (3) The present invention constructs a one-dimensional vector based on the measured rainfall corresponding to each typhoon event in the historical typhoon data set, and constructs a corresponding one-dimensional vector for each forecasting agency based on the rainfall forecast values ​​of multiple forecasting agencies corresponding to each typhoon event. This approach not only integrates multi-source data, but also further improves data accuracy through a comprehensive correction coefficient, reducing the uncertainty of single-source forecasts;

[0066] (4) The present invention obtains the comprehensive correction coefficient corresponding to each forecasting agency through the historical typhoon data set and the one-dimensional vector corresponding to each forecasting agency. This step effectively corrects the systematic error of each forecasting agency, making the future rainfall forecast closer to reality and improving the reliability and consistency of the forecast;

[0067] (5) The present invention constructs a dynamic Bayesian optimization model based on the historical typhoon data set and the one-dimensional vector corresponding to each forecasting agency and the comprehensive correction coefficient corresponding to each forecasting agency; the model can obtain the optimal weight coefficient and standard deviation corresponding to each forecasting agency, realize the intelligent fusion of multi-source forecast data, and improve the accuracy and stability of the forecast results;

[0068] (6) The present invention obtains the latest future rainfall forecast values ​​of each forecasting agency, and corrects these values ​​through the comprehensive correction coefficient to obtain the forecast correction value corresponding to each forecasting agency. Finally, the final forecast fusion value and its corresponding variance estimation value are obtained through the optimal weight coefficient, standard deviation and forecast correction value corresponding to each forecasting agency. This method not only improves the accuracy of the forecast, but also provides uncertainty assessment of the forecast results, thereby enhancing the credibility and practicality of the forecast.

[0069] (7) The present invention overcomes the problems of large spatiotemporal variability, inaccurate rainfall pattern distribution and insufficient forecast accuracy of single-source rainfall forecast data, realizes the efficient fusion of multi-source rainfall forecast data, and significantly improves the accuracy and reliability of flood forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a flow chart of a multi-source rainfall forecast fusion method based on typhoon similarity. DETAILED DESCRIPTION

[0071] The following are specific embodiments of the present invention and the accompanying drawings to further describe the technical solution of the present invention, but the present invention is not limited to these embodiments.

[0072] Since the future rainfall patterns of single-source agencies are usually forecasted in time series with a step length of 3 hours, it is impossible to provide hourly rainfall time series, resulting in insufficient refinement of rainfall pattern distribution, making it difficult to meet the needs of flood forecasting models for high-resolution input data. Secondly, the accuracy of future rainfall forecasts by single-source agencies is insufficient. Due to technical limitations and uncertainties in external conditions, these forecast values ​​are often too large and cannot accurately reflect actual rainfall conditions, thus affecting the reliability and accuracy of flood forecasting models. Therefore, in order to achieve the fusion of multi-source rainfall forecast data, improve the data accuracy of rainfall forecasts, and thereby improve the accuracy and reliability of flood forecasts, such as Figure 1 As shown, the present invention proposes a multi-source rainfall forecast fusion method based on typhoon similarity, comprising:

[0073] Select the key factors that affect the formation and development of typhoons, obtain the multidimensional vectors formed by the key factors corresponding to each typhoon path point in the typhoon path for the current typhoon and each historical typhoon, and construct the typhoon factor time series corresponding to each typhoon based on the multidimensional vectors corresponding to each typhoon path point;

[0074] The key factors include: typhoon wind speed, atmospheric pressure, typhoon moving speed, seasonal factor value, and the central longitude and central latitude of the typhoon path point.

[0075] The current typhoon is combined with each historical typhoon one-to-one, and for each combination, the corresponding typhoon factor time series is used to calculate the similarity distance between the current typhoon and the corresponding historical typhoon; historical typhoons with similarity distances less than a preset threshold are selected to form a historical typhoon data set;

[0076] For each combination, the similarity distance between the current typhoon and the historical typhoon is calculated using the corresponding typhoon factor time series, specifically including:

[0077] The points in the first sequence and the second sequence, i.e., the multidimensional vectors, are combined in pairs, and the multidimensional Euclidean distance of each point pair is calculated; the first sequence represents the typhoon factor time series of the current typhoon in the combination; the second sequence represents the typhoon factor time series corresponding to the historical typhoon in the corresponding combination;

[0078] The formula expression of the first sequence is:

[0079] ;

[0080] in, ;

[0081] In the formula, i=1…n, n represents the number of points in the first sequence; Represents the multidimensional vector corresponding to the i-th point in the first sequence; Respectively represent the central longitude, central latitude, typhoon wind speed, atmospheric pressure, typhoon moving speed and seasonal factor value of the i-th point in the first sequence;

[0082] The formula expression of the second sequence is:

[0083] ;

[0084] in, ;

[0085] In the formula, j=1…m, m represents the number of points in the second sequence; Represents the multidimensional vector corresponding to the j-th point in the second sequence; Respectively represent the central longitude, central latitude, typhoon wind speed, atmospheric pressure, typhoon moving speed and seasonal factor value of the jth point in the second sequence;

[0086] The calculation formula for obtaining the multidimensional Euclidean distance of each point pair is:

[0087] ;

[0088] In the formula, Indicate point and The multidimensional Euclidean distance between ; , , , , , The sum of the addition is 1, which represents the preset weight values ​​corresponding to the central longitude, central latitude, typhoon wind speed, atmospheric pressure, typhoon moving speed and seasonal factor value respectively; It represents the direct distance between the i-th point in the first sequence and the j-th point in the second sequence, that is, the multidimensional Euclidean distance.

[0089] Construct a distance matrix based on the multidimensional Euclidean distances of all point pairs;

[0090] Based on the distance matrix, a dynamic time warping algorithm is used to obtain a cumulative distance matrix M for recording the cumulative distance between the first sequence and the second sequence; the elements in the cumulative distance matrix M are Indicates the first sequence The first point in the second sequence The minimum cumulative distance when points are aligned;

[0091] The method of obtaining a cumulative distance matrix M for recording the cumulative distance between the first sequence and the second sequence by using a dynamic time warping algorithm based on the distance matrix specifically includes:

[0092] Set the boundary conditions for obtaining the cumulative distance matrix M;

[0093] The boundary conditions include:

[0094] , indicating that the cumulative distance between the first sequence and the starting point of the second sequence is zero;

[0095] ; In the formula, , represents the cumulative distance when aligning the first i points of the first sequence with the first point of the second sequence;

[0096] ; In the formula, , represents the cumulative distance when aligning the first point of the first sequence with the first j points of the second sequence.

[0097] Using the boundary conditions and the elements in the distance matrix, each element in the cumulative distance matrix M is gradually solved in a recursive manner to obtain the cumulative distance matrix M; the solution formula is:

[0098] ; and ;

[0099] In the formula, represents the minimum cumulative distance among all possible paths from the first point of the first sequence to the i-th point and from the first point of the second sequence to the j-th point;

[0100] Indicates diagonal movement, that is, from the first sequence The first point to the ith point and the second point from the The minimum cumulative distance among all possible paths between the point to the jth point;

[0101] Indicates horizontal movement, that is, from the first sequence The minimum cumulative distance from the i-th point to the second sequence among all possible paths that remain at the j-th point;

[0102] represents vertical movement, that is, from the first sequence to the second sequence, The minimum cumulative distance among all possible paths from point j to point j.

[0103] The following is a detailed description of the data In order to simplify the calculation process, this embodiment uses the first sequence With the second sequence Both include only 3 point pairs Calculation instructions:

[0104] Assumptions: , , and the distance matrix D is:

[0105]

[0106] Represents the first sequence The i-th point and the second sequence The direct distance between the j-th points in .

[0107] Based on this, the calculation process is as follows:

[0108] 1. Boundary conditions:

[0109] ;

[0110] ;

[0111] ;

[0112] .

[0113] 2. Fill the elements in the cumulative distance matrix M step by step in a recursive manner:

[0114] 1) Calculation :

[0115] ;

[0116] 2) Calculation :

[0117] ;

[0118] And so on, until the last element in the cumulative distance matrix M is solved (since the calculation of the current element depends on the value of the previous element, all elements need to be solved step by step to calculate the last element).

[0119] The similarity distance between the first sequence and the second sequence is calculated by accumulating the last element in the distance matrix M; the calculation formula is:

[0120] In the formula, The last element in the cumulative distance matrix M represents the minimum cumulative distance of the global optimal matching path between the first sequence and the second sequence; Represent the number of points in the first and second sequences respectively; Represents the similarity distance between the first sequence and the second sequence.

[0121] The present invention combines the current typhoon with each historical typhoon on a one-to-one basis, and uses the corresponding typhoon factor time series to calculate the similarity distance. By selecting historical typhoons whose similarity distance is less than a preset threshold, a historical typhoon data set is formed (i.e., historical typhoon rain types similar to the current typhoon are obtained), ensuring that the historical typhoons used for reference are highly similar to the current typhoon, thereby improving the accuracy of the forecast.

[0122] It should be noted that, in this embodiment, the similarity distance may be normalized by a normalization function, and historical typhoons whose normalized values ​​are less than a set threshold are selected to form a historical typhoon data set; the formula expression of the normalization function is:

[0123] ;Where k=100.

[0124] A one-dimensional vector is constructed based on the measured rainfall corresponding to each typhoon event in the historical typhoon data set, and a corresponding one-dimensional vector is constructed for each forecasting agency based on the rainfall forecast values ​​of multiple forecasting agencies corresponding to each typhoon event;

[0125] In this embodiment, the forecasting agencies include: GFS forecasting agency, EC forecasting agency and China Meteorological Administration.

[0126] The formula expressions of the one-dimensional vectors corresponding to the historical typhoon data set, GFS forecast agency, EC forecast agency and China Meteorological Administration are:

[0127] ;

[0128] In the formula, Represents the one-dimensional vector corresponding to the historical typhoon data set, that is, the set of measured rainfall corresponding to each typhoon event, represents the measured rainfall corresponding to the i-th historical typhoon; Represents the one-dimensional vector corresponding to the GFS forecast agency, that is, the set of rainfall forecast values ​​corresponding to each typhoon event predicted by the agency, It represents the rainfall forecast value of the i-th historical typhoon predicted by the GFS forecast agency; represents the one-dimensional vector corresponding to the EC forecast agency, that is, the set of rainfall forecast values ​​corresponding to each typhoon event predicted by the agency, It represents the rainfall forecast value of the i-th historical typhoon predicted by the EC forecast agency; represents the one-dimensional vector corresponding to the China Meteorological Administration, that is, the set of rainfall forecast values ​​corresponding to each typhoon event predicted by the Meteorological Administration. represents the rainfall forecast value of the i-th historical typhoon predicted by the China Meteorological Administration; among them, , Represents the number of typhoons in the historical typhoon data set.

[0129] The present invention constructs a one-dimensional vector according to the measured rainfall corresponding to each typhoon event in the historical typhoon data set, and constructs a corresponding one-dimensional vector for each forecasting agency according to the rainfall forecast values ​​of multiple forecasting agencies corresponding to each typhoon event. This approach not only integrates multi-source data, but also further improves data accuracy through a comprehensive correction coefficient, reducing the uncertainty of single-source forecasts.

[0130] Through the historical typhoon data set and the one-dimensional vector corresponding to each forecasting agency, the comprehensive correction coefficient corresponding to each forecasting agency is obtained; specifically, it includes:

[0131] For each element in the measured one-dimensional vector, the ratio between the element and the corresponding element in the one-dimensional vector corresponding to each forecasting agency is obtained to obtain a correction coefficient set corresponding to each forecasting agency; the measured one-dimensional vector represents a one-dimensional vector corresponding to the historical typhoon data set;

[0132] In this embodiment, the formula expression of the ratio is:

[0133] ; In the formula, represents the rainfall forecast value of the jth historical typhoon predicted by the i-th forecast agency, represents the measured rainfall corresponding to the jth historical typhoon, It represents the ratio (or single-event ratio coefficient) of the i-th forecast agency for the j-th historical typhoon;

[0134] The formula expressions of the correction coefficient sets corresponding to the GFS forecast agency, EC forecast agency and China Meteorological Administration are:

[0135] ;

[0136] In the formula, represents the set of correction coefficients corresponding to the GFS forecast agency, represents the set of correction coefficients corresponding to the EC forecasting agency, Represents the set of correction coefficients corresponding to the China Meteorological Administration.

[0137] The correction coefficient sets corresponding to each forecasting agency are averaged to obtain the comprehensive correction coefficients corresponding to each forecasting agency.

[0138] The present invention obtains the comprehensive correction coefficient corresponding to each forecasting agency through the historical typhoon data set and the one-dimensional vector corresponding to each forecasting agency. This step effectively corrects the systematic errors of each forecasting agency, making future rainfall forecasts closer to reality and improving the reliability and consistency of the forecast.

[0139] In this embodiment, the formula for averaging is:

[0140] ; In the formula, represents the comprehensive correction coefficient of the i-th forecasting agency.

[0141] A dynamic Bayesian optimization model is constructed through the historical typhoon data set, the one-dimensional vector corresponding to each forecasting agency, and the comprehensive correction coefficient corresponding to each forecasting agency. Based on the dynamic Bayesian optimization model, the optimal weight coefficient and standard deviation corresponding to each forecasting agency are obtained; specifically, the following are included:

[0142] Construct a posterior probability distribution; the calculation formula of the posterior probability distribution is:

[0143] ;

[0144] In the formula, It represents the forecast fusion value obtained by fusing the rainfall forecast values ​​of various forecast agencies. represents the measured one-dimensional vector, s represents the number of forecasting agencies, represents the weight coefficient of the i-th forecasting agency, represents the comprehensive correction coefficient of the i-th forecast agency, represents the one-dimensional vector of the i-th forecasting agency, Represents the posterior probability distribution of the measured one-dimensional vector and the corresponding one-dimensional vector of different forecast agencies; represents the posterior probability distribution;

[0145] set up Normal distribution: ; In the formula, Indicates the number of forecast agencies in the i-th The standard deviation of

[0146] Construct the calculation formula for the deterministic estimate:

[0147] ;

[0148] In the formula, , represents the forecast expectation value of the i-th forecast agency; represents a deterministic estimate;

[0149] Construct a prediction variance calculation formula to reflect the numerical uncertainty of forecast fusion:

[0150] ;

[0151] In the formula, represents the forecast variance reflecting the numerical uncertainty of forecast fusion;

[0152] Construct the log-likelihood function using the maximum likelihood estimation method:

[0153] ;

[0154] Through the calculation formula corresponding to the posterior probability distribution, deterministic estimate, prediction variance and log-likelihood function, and The dynamic Bayesian optimization model formed by the formula expression obeying the normal distribution is used to obtain For each parameter in , the optimal weight coefficient and standard deviation corresponding to each forecasting agency are obtained.

[0155] The present invention constructs a dynamic Bayesian optimization model based on the historical typhoon data set and the one-dimensional vectors corresponding to each forecasting agency and the comprehensive correction coefficient corresponding to each forecasting agency; the model can obtain the optimal weight coefficient and standard deviation corresponding to each forecasting agency, realize the intelligent fusion of multi-source forecast data, and improve the accuracy and stability of the forecast results.

[0156] Obtain the latest future rainfall forecast values ​​from each forecast agency, and correct the future rainfall forecast values ​​through the comprehensive correction coefficient to obtain the forecast correction values ​​corresponding to each forecast agency. Obtain the forecast fusion value and its corresponding variance estimate through the optimal weight coefficient, standard deviation and forecast correction value corresponding to each forecast agency, including:

[0157] The latest future rainfall forecast values ​​of each forecasting agency are obtained, and the corresponding future rainfall forecast values ​​are corrected by the comprehensive correction coefficients of each forecasting agency; the correction formula is:

[0158] ; In the formula, represents the future rainfall forecast value of the i-th forecast agency, represents the forecast correction value corresponding to the i-th forecast agency;

[0159] According to the forecast correction values ​​and optimal weight coefficients corresponding to each forecast agency, the forecast fusion value is calculated using the following formula: ; In the formula, and ;

[0160] Based on the calculated forecast fusion value, the variance estimation value used to reflect the uncertainty of the forecast fusion value is obtained. The formula is:

[0161] , where Represents the variance estimate.

[0162] The present invention obtains the latest future rainfall forecast values ​​of each forecasting agency, and corrects these values ​​through a comprehensive correction coefficient to obtain the forecast correction value corresponding to each forecasting agency. Finally, the final forecast fusion value and its corresponding variance estimation value are obtained through the optimal weight coefficient, standard deviation and forecast correction value corresponding to each forecasting agency. This method not only improves the accuracy of the forecast, but also provides an uncertainty assessment of the forecast results, thereby enhancing the credibility and practicality of the forecast.

[0163] Specifically, the present invention selects key factors that affect the formation and development of typhoons, and obtains multidimensional vectors corresponding to each typhoon path point for the current typhoon and each historical typhoon; this method ensures a comprehensive description of typhoon characteristics, improves the refinement of typhoon paths and factor time series, and provides high-quality basic data for subsequent calculations of similar distances; by constructing a historical typhoon data set, and based on the one-dimensional vectors corresponding to the set and each forecasting agency, the present invention obtains the comprehensive correction coefficient of each forecasting agency, effectively corrects the systematic errors of each forecasting agency, and makes future rainfall forecasts closer to reality; on this basis, the present invention further utilizes The comprehensive correction coefficient is used to calculate the optimal weight coefficient and standard deviation, which significantly improves the calculation accuracy. Finally, by obtaining the latest future rainfall forecast values ​​of each forecast agency and using the comprehensive correction coefficient for correction, the corresponding forecast correction values ​​of each forecast agency are obtained. Combining the optimal weight coefficient, standard deviation and forecast correction value of each forecast agency, the present invention obtains the forecast fusion value and its corresponding variance estimate. This process not only realizes the targeted prediction of the latest future rainfall forecast values ​​(i.e., targeted prediction of the current typhoon through similar historical typhoon data sets), but also greatly improves the prediction accuracy, providing a reliable basis for flood forecasting and flood prevention and disaster reduction.

[0164] In addition, the present invention overcomes the problems of large spatiotemporal variability of single-source rainfall forecast data, inaccurate rainfall pattern distribution and insufficient forecast accuracy, realizes the efficient fusion of multi-source rainfall forecast data, and significantly improves the accuracy and reliability of flood forecasting.

[0165] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0166] In addition, in the present invention, descriptions such as "first", "second", "one", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0167] In the present invention, unless otherwise clearly specified and limited, the terms "connection", "fixation", etc. should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0168] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. A multi-source rainfall forecast fusion method based on typhoon similarity, characterized in that: include: Select the key factors that affect the formation and development of typhoons, obtain the multidimensional vectors formed by the key factors corresponding to each typhoon path point in the typhoon path for the current typhoon and each historical typhoon, and construct the typhoon factor time series corresponding to each typhoon based on the multidimensional vectors corresponding to each typhoon path point; The key factors include: typhoon wind speed, atmospheric pressure, typhoon moving speed, seasonal factor value, and central longitude and central latitude of the typhoon path point; The current typhoon is combined with each historical typhoon one-to-one, and for each combination, the corresponding typhoon factor time series is used to calculate the similarity distance between the current typhoon and the corresponding historical typhoon; historical typhoons with similarity distances less than a preset threshold are selected to form a historical typhoon data set; For each combination, the similarity distance between the current typhoon and the corresponding historical typhoon is calculated using the corresponding typhoon factor time series, specifically including: The points in the first sequence and the second sequence, i.e., the multidimensional vectors, are combined in pairs, and the multidimensional Euclidean distance of each point pair is calculated; the first sequence represents the typhoon factor time series of the current typhoon in the combination; the second sequence represents the typhoon factor time series corresponding to the historical typhoon in the corresponding combination; Construct a distance matrix based on the multidimensional Euclidean distances of all point pairs; Based on the distance matrix, a dynamic time warping algorithm is used to obtain a cumulative distance matrix M for recording the cumulative distance between the first sequence and the second sequence; The similarity distance between the first sequence and the second sequence is calculated by accumulating the last element in the distance matrix M; A one-dimensional vector is constructed based on the measured rainfall corresponding to each typhoon event in the historical typhoon data set, and a corresponding one-dimensional vector is constructed for each forecasting agency based on the rainfall forecast values ​​of multiple forecasting agencies corresponding to each typhoon event; Through the historical typhoon data set and the one-dimensional vector corresponding to each forecasting agency, the comprehensive correction coefficient corresponding to each forecasting agency is obtained, including: For each element in the measured one-dimensional vector, the ratio between the element and the corresponding element in the one-dimensional vector corresponding to each forecasting agency is obtained to obtain a correction coefficient set corresponding to each forecasting agency; the measured one-dimensional vector represents a one-dimensional vector corresponding to the historical typhoon data set; The correction coefficient sets corresponding to each forecasting agency are averaged to obtain the comprehensive correction coefficients corresponding to each forecasting agency; A dynamic Bayesian optimization model is constructed through the historical typhoon data set, the one-dimensional vector corresponding to each forecasting agency, and the comprehensive correction coefficient corresponding to each forecasting agency. Based on the dynamic Bayesian optimization model, the optimal weight coefficient and standard deviation corresponding to each forecasting agency are obtained. The latest future rainfall forecast values ​​from each forecasting agency are obtained, and the future rainfall forecast values ​​are corrected by the comprehensive correction coefficient to obtain the corresponding forecast correction values ​​of each forecasting agency. The forecast fusion values ​​and their corresponding variance estimates are obtained by using the corresponding optimal weight coefficients, standard deviations and forecast correction values ​​of each forecasting agency.

2. The multi-source rainfall forecast fusion method based on typhoon similarity according to claim 1 is characterized in that: The formula expression of the first sequence is: ;in, ; In the formula, i=1…n, n represents the number of points in the first sequence; Represents the multidimensional vector corresponding to the i-th point in the first sequence; Respectively represent the central longitude, central latitude, typhoon wind speed, atmospheric pressure, typhoon moving speed and seasonal factor value of the i-th point in the first sequence; The formula expression of the second sequence is: ;in, ; In the formula, j=1…m, m represents the number of points in the second sequence; Represents the multidimensional vector corresponding to the j-th point in the second sequence; Respectively represent the central longitude, central latitude, typhoon wind speed, atmospheric pressure, typhoon moving speed and seasonal factor value of the jth point in the second sequence; The calculation formula for obtaining the multidimensional Euclidean distance of each point pair is: ； In the formula, Indicate point and The multidimensional Euclidean distance between ; , , , , , The sum of the addition is 1, which represents the preset weight values ​​corresponding to the central longitude, central latitude, typhoon wind speed, atmospheric pressure, typhoon moving speed and seasonal factor value respectively; It represents the direct distance between the i-th point in the first sequence and the j-th point in the second sequence, that is, the multidimensional Euclidean distance.

3. The multi-source rainfall forecast fusion method based on typhoon similarity according to claim 2 is characterized in that: The method of obtaining a cumulative distance matrix M for recording the cumulative distance between the first sequence and the second sequence by using a dynamic time warping algorithm based on the distance matrix specifically includes: Set the boundary conditions for obtaining the cumulative distance matrix M; Using the boundary conditions and the elements in the distance matrix, each element in the cumulative distance matrix M is gradually solved in a recursive manner to obtain the cumulative distance matrix M; the solution formula is: ; and ; In the formula, represents the minimum cumulative distance among all possible paths from the first point of the first sequence to the i-th point and from the first point of the second sequence to the j-th point; Indicates diagonal movement, that is, from the first sequence The first point to the ith point and the second point from the The minimum cumulative distance among all possible paths between the point to the jth point; Indicates horizontal movement, that is, from the first sequence The minimum cumulative distance from the i-th point to the j-th point in all possible paths of the second sequence; represents vertical movement, that is, from the first sequence to the second sequence, The minimum cumulative distance among all possible paths from point j to point j.

4. The multi-source rainfall forecast fusion method based on typhoon similarity according to claim 3 is characterized in that: The boundary conditions include: , indicating that the cumulative distance between the first sequence and the starting point of the second sequence is zero; ; In the formula, , represents the cumulative distance when aligning the first i points of the first sequence with the first point of the second sequence; ; In the formula, , represents the cumulative distance when aligning the first point of the first sequence with the first j points of the second sequence.

5. The multi-source rainfall forecast fusion method based on typhoon similarity according to claim 4 is characterized in that: The similarity distance between the first sequence and the second sequence is calculated by accumulating the last element in the distance matrix M, and the calculation formula is: In the formula, The last element in the cumulative distance matrix M represents the minimum cumulative distance of the global optimal matching path between the first sequence and the second sequence; Represent the number of points in the first and second sequences respectively; Represents the similarity distance between the first sequence and the second sequence.

6. The multi-source rainfall forecast fusion method based on typhoon similarity according to claim 1 is characterized in that: The method constructs a dynamic Bayesian optimization model through a historical typhoon data set and a one-dimensional vector corresponding to each forecasting agency and a comprehensive correction coefficient corresponding to each forecasting agency, and obtains the optimal weight coefficient and standard deviation corresponding to each forecasting agency based on the dynamic Bayesian optimization model, specifically including: Construct a posterior probability distribution; the calculation formula of the posterior probability distribution is: ； In the formula, It represents the forecast fusion value obtained by fusing the rainfall forecast values ​​of various forecast agencies. represents the measured one-dimensional vector, s represents the number of forecasting agencies, represents the weight coefficient of the i-th forecasting agency, represents the comprehensive correction coefficient of the i-th forecast agency, represents the one-dimensional vector of the i-th forecasting agency, Represents the posterior probability distribution of the measured one-dimensional vector and the corresponding one-dimensional vector of different forecast agencies; represents the posterior probability distribution; set up Normal distribution: ; In the formula, Indicates the number of forecast agencies in the i-th The standard deviation of Construct the calculation formula for the deterministic estimate: ； In the formula, , represents the forecast expectation value of the i-th forecast agency; represents a deterministic estimate; Construct a prediction variance calculation formula to reflect the numerical uncertainty of forecast fusion: ； In the formula, represents the forecast variance reflecting the numerical uncertainty of forecast fusion; Construct the log-likelihood function using the maximum likelihood estimation method: ； Through the calculation formula corresponding to the posterior probability distribution, deterministic estimate, prediction variance and log-likelihood function, and The dynamic Bayesian optimization model formed by the formula expression obeying the normal distribution is used to obtain For each parameter in , the optimal weight coefficient and standard deviation corresponding to each forecasting agency are obtained.

7. The multi-source rainfall forecast fusion method based on typhoon similarity according to claim 6 is characterized in that: The method of obtaining the latest future rainfall forecast values ​​of each forecast agency, and correcting the future rainfall forecast values ​​by using a comprehensive correction coefficient to obtain the forecast correction values ​​corresponding to each forecast agency, and obtaining the forecast fusion values ​​and their corresponding variance estimation values ​​by using the optimal weight coefficients, standard deviations and forecast correction values ​​corresponding to each forecast agency, specifically includes: The latest future rainfall forecast values ​​of each forecasting agency are obtained, and the corresponding future rainfall forecast values ​​are corrected by the comprehensive correction coefficients of each forecasting agency; the correction formula is: ; In the formula, represents the future rainfall forecast value of the i-th forecast agency, represents the forecast correction value corresponding to the i-th forecast agency; According to the forecast correction values ​​and optimal weight coefficients corresponding to each forecast agency, the forecast fusion value is calculated using the following formula: ; In the formula, and ; Based on the calculated forecast fusion value, the variance estimation value used to reflect the uncertainty of the forecast fusion value is obtained. The formula is: , where Represents the variance estimate.

Images