Flood forecast intelligent recommendation method based on pattern library

Abstract

The invention discloses a flood forecast intelligent recommendation method based on a pattern library. The method comprises the following steps: firstly, mining a rainfall-flood association mode from historical rainfall data and historical flood data, and constructing a rainfall-flood association mode library; through the steps of target rainfall feature extraction, target rainfall similarity search, construction of a multi-level rainfall-flood incidence matrix, recommendation degree calculation and the like, a flood process recommendation list is obtained, and the flood process recommendation list serves as forecasting and early warning information of a flood development trend possibly caused by current rainfall. According to the method, flood forecast intelligent recommendation based on the rainfall-flood association mode library is realized, the flood process possibly caused by current rainfall can be predicted according to historical scenes, and powerful technical support is provided for flood control decision making.

Description

Technical Field

[0001] The invention relates to the technical field of hydrology, meteorology and intelligent flood forecasting, and in particular to an intelligent flood forecasting recommendation method based on a pattern library. Background Art

[0002] In recent years, the continuous outbreak of flood disasters has posed a serious threat to people's life safety and the natural environment, and has also caused huge property losses. The occurrence of floods is affected by many factors such as hydrology, meteorology, and topography, and real-time flood forecasting is highly uncertain. Rainfall is one of the main factors that cause floods, and there are certain connections and regularities between rainfall and floods in different evolution stages. Therefore, how to transform massive rainfall data into useful information and knowledge, use similarity analysis and clustering methods to extract and classify similar rainfall / flood processes, explore the correlation between rainfall and floods, and predict future flood development trends through intelligent recommendation algorithms has important research significance.

[0003] Hydrological patterns are defined as a collection of hydrological processes with similar characteristics. Using historical hydrological data, identifying and extracting similar rainfall patterns and flood patterns, and then establishing a pattern library, can better analyze the correlation and response relationship between rainfall events and flood events, and play an important role in water resources management, flood disaster risk assessment and other fields. Regarding the correlation between rainfall and floods, some studies have constructed a rainfall-flood pattern library to analyze and explore the spatiotemporal variation laws and causal mechanisms between rainfall and floods, thereby predicting the changing trend of floods, providing new ideas for flood forecasting.

[0004] Although the existing flood forecasting and prediction application research based on the pattern library has achieved good results, most of the methods rely on traditional structured data, fail to fully consider the spatial characteristics of rainfall, and have certain limitations. With the continuous development of technology, rainfall data has gradually expanded from traditional time series data to unstructured data such as rainfall maps. Based on image features, rich spatiotemporal features can be extracted from historical rainfall data sets, which is of great significance for the identification of similar rain and flood processes and the prediction of flood events. Therefore, based on rainfall image features and flood features, the present invention uses machine learning, data mining and other technical methods to target the correlation between rainfall events and flood events, and mines and queries similar rainfall patterns in history by constructing a rainfall-flood pattern library, and then mines the flood process that may be caused by rainfall; and intelligently recommends the historical flood development process to users as a similar flood process forecast result, providing strong technical support for flood control and flood prevention decision-making. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to solve the problem that most existing technical solutions do not fully consider the spatial characteristics of rainfall, and there are few related studies on the use of recommendation algorithms for prediction in the field of flood forecasting, and there are limitations such as cold start and data sparsity. In view of the shortcomings of the existing technology, the present invention provides an intelligent recommendation method for flood forecasting based on a pattern library, which predicts the development trend of floods through the extraction of spatiotemporal characteristics of rainfall and floods, clustering and analysis of similar rainfall / flood processes, mining of rainfall-flood correlation patterns, and intelligent recommendation algorithms under hydrological scenarios, thereby providing new ideas and method references for real-time flood forecasting in river basins.

[0006] Technical solution: A flood forecast intelligent recommendation method based on a pattern library of the present invention is characterized by comprising the following steps:

[0007] Step S1, constructing a rainfall-flood correlation pattern library: First, preprocess the historical rainfall data and historical flood data, and construct a rainfall pattern library RPD = {{R1:RP 11 , RP 12 , ...}, {R2: RP 21 , RP 22 ,...},...,{R n :RP n1 , RP n2 ,...}} and the flood pattern library FPD = {{F1: FP u , FP 12 ,...}, {F2: FP 21 , FP 22 ,...},...,{F m :FP m1 , FP m2 ,...}}, where n is the number of rainfall patterns in the rainfall pattern library, and m is the number of flood patterns in the flood pattern library;

[0008] Then, the rainfall-flood item set is constructed according to the spatiotemporal response relationship between rainfall and flood, and the Apriori algorithm is used to mine rainfall→flood association rules from the pattern library to construct the rainfall-flood association pattern library RFAD = {R i →F j , R i = {RP i1 , Rp i2 , ...}, F j ={FP j1 , Fp j2 ,...}, 1≤i≤n, 1≤j≤m}.

[0009] Step S2, extracting the real-time spatiotemporal characteristics of the target rainfall: input the target rainfall event, extract its spatiotemporal characteristics based on image analysis, and obtain the target rainfall event rainfall feature vector RF obj .

[0010] Step S3, rainfall event similarity search: search the rainfall event pattern library RPD, use the multi-feature fusion distance measurement algorithm to calculate the similarity between the target rainfall and each rainfall in the pattern library, and sort the results to obtain the similar rainfall set SR of the target rainfall.

[0011] Step S4, construct a multi-level rainfall-flood association matrix: First, construct a pattern association matrix based on the association relationship between the overall patterns. Through the rainfall → flood association rules mined from the association pattern library in step 1, define the degree of association between each rainfall pattern and flood pattern, and obtain the pattern association matrix PM; then, construct a historical association matrix based on the historical rainfall-flood spatiotemporal relationship, and calculate the rainfall pattern R in PM that corresponds to the target rainfall. obj The flood mode with the highest correlation is F obj , according to the rainfall-flood relationship between specific events in history, the historical correlation matrix HM is constructed, thus forming a multi-level rainfall-flood correlation matrix MAM.

[0012] Step S5, calculate the recommendation degree of the flood process: define a method for calculating the recommendation degree. In the multi-level rainfall-flood association matrix MAM, the recommendation degree value is obtained by weighted summing the association degree and similarity of each item. The results are sorted to obtain a topK ranked flood recommendation list, including recommended information of the flood process (flood duration, peak flow, increase, etc.), to provide decision support for the forecast and warning of the flood process that may be caused by the target rainfall.

[0013] Step S6: According to the changing trend of the current rainfall, the rainfall pattern library is updated in real time, and the recommended results of the flood process are optimized, that is, the process of steps S2 to S5 is repeated until the rainfall process ends.

[0014] Furthermore, the detailed process of constructing the rainfall-flood correlation pattern library in step S1 is as follows:

[0015] Step S11: based on the rainfall threshold δ and the sliding window, the historical rainfall image sequence is divided into a historical rainfall set RS; each rainfall RP in the rainfall set is extracted k The cumulative rainfall CP, the number of rainy days N, the rainfall time distribution SD, the rainfall spatial distribution TD and the rainfall center motion trajectory TC characteristics form the rainfall feature vector RF k ={CP k , N k ,SD k ,TDk , TC k}; Use the multi-feature fusion distance measurement algorithm to calculate the similarity between different rainfall processes, and cluster similar rainfall processes using the K-means algorithm based on the Davies-Bouldin index to complete the rainfall pattern library RPD = {{R1:RP 11 , RP 12 ,...}, {R2: RP 21 , RP 22 , ...}, ..., {R n :RP n1， RP n2 ,...}} construction.

[0016] Step S12: Segment the historical flood flow data into flood events by time span and flow span to obtain a flood event set FS; for each flood FP in the flood event set k , extract the flood duration FT, peak flow PF, rising water rate RS, peak type N and other characteristics, and obtain the flood feature vector FF k ={FT k , PF k , R.S. k , N k , ...}; Similarly, based on the distance measurement algorithm of multi-feature fusion, similar flood processes are clustered to complete the flood pattern library FPD = {{F1: FP 11 , FP 12 ,...}, {F2: FP 21 , FP 22 ,...},...,{F m :FP m1 , FP m2 ,...}} construction.

[0017] Step S13: Based on the spatiotemporal relationship between rainfall and flood, the sliding window technology and DDTW algorithm are used to construct the rainfall-flood item set IS as input; then, according to the characteristics and statistical laws of the target watershed data set, the minimum support and minimum confidence are set, and the Apriori algorithm is used to mine the rainfall-flood association pattern between the rainfall pattern library RPD and the flood pattern library FPD, and the strong association rule of rainfall → flood is obtained, thereby constructing the rainfall-flood association pattern library RFAD = {R i →F j , R i = {RP i1 , R pi2 , ...}, F j ={FP j1 , Fp j2, ...}, 1≤i≤n, 1≤j≤m}.

[0018] Furthermore, the detailed process of extracting the real-time spatiotemporal characteristics of target rainfall in step S2 is as follows:

[0019] Step S21: For the current date t, if the rainfall on that day CP t Greater than the rainfall threshold δ, and t is consistent with the most recent rainfall process RP in the rainfall pattern library last If the time interval between them does not exceed 2 days, the rainfall on that day is considered to belong to RP last As part of the rainfall process, the rainfall model library RP last Feature information RF obj Update; otherwise, execute step S22.

[0020] Step S22: Target rainfall is a new rainfall event RP obj , the color histogram technology is used to extract the spatiotemporal characteristics of the target rainfall, including the cumulative rainfall CP, the number of rainy days N, the rainfall time distribution SD, the rainfall spatial distribution TD and the rainfall center motion trajectory TC, and the real-time rainfall feature vector RF is obtained obj ={CP obj , N obj , S.D. obj , TD obj , TC obj}, and the rainfall pattern R to which the target rainfall belongs obj , and add the above characteristic information to the rainfall pattern library.

[0021] Furthermore, the detailed process of constructing the multi-level rainfall-flood correlation matrix in step S4 is as follows:

[0022] Step S41: mining the historical rainfall pattern set HS from the rainfall pattern library R = {R1, R2, ..., R n Similarly, the historical flood pattern set HS is mined from the flood pattern library F ={F1, F2, ..., F m}; Mining the historical rainfall-flood correlation pattern set HS from the rainfall-flood correlation pattern library obtained in step S13 R→F ={{R i →F j , Prob Ri→Fj}, 1≤i≤n, 1≤j≤m}.

[0023] Step S42: construct a pattern association matrix PM. R is the horizontal axis, HS FAs the ordinate, establish an n*m pattern association matrix PM. For the elements PM in the matrix ij , which represents the rainfall pattern R i With flood mode F j The degree of correlation between them can be determined by the probability Prob in the rainfall-flood correlation pattern set Ri→Fj To express, the calculation method is as follows:

[0024]

[0025] Step S43: Search the pattern association matrix PM for the rainfall pattern R to which the target rainfall belongs. obj The flood mode with the highest correlation is F p ; Get F from the flood pattern database p The corresponding flood process set SF = {FP p1 , FP p2 , ..., FP pk}; Traverse the rainfall-flood item set IS = {{RP1, FP1}, {RP2, FP2}, ..., {RP n , FP n}}, get the corresponding relationship between each rainfall and flood in history, such as the rainfall process RP i Triggered the flood process FP i .

[0026] Step S44: Construct the historical association matrix HM. Construct the historical association matrix HM with SR as the horizontal coordinate and SF as the vertical coordinate. For the elements HM in the matrix ij , which represents the rainfall process RP i FP with flood process j The degree of correlation between them can be expressed by the corresponding relationship between rainfall and flood in the historical rainfall-flood item set. The calculation formula is as follows:

[0027]

[0028] After the above steps, a multi-level rainfall-flood association matrix MAM can be obtained.

[0029] Furthermore, the detailed process of calculating the flood process recommendation degree in step S5 is as follows:

[0030] Step S51: Calculate the recommendation degree for each flood in the rainfall-flood association matrix MAM. First, query the flood-causing process FP from the historical rainfall-flood item set IS. i The rainfall process RP i , use the multi-feature distance metric algorithm to calculate the target rainfall RP obj With RP iThe similarity between them is expressed as sim(obj,i).

[0031] Step S52: Search the rainfall-flood correlation matrix to obtain the flood process FP i and rainfall pattern R obj The correlation degree between each rainfall process in forms a correlation vector {a1, a2, ..., a M}, M is the number of similar rainfall events.

[0032] Step S53: define the method for calculating the recommendation score, and convert FP i The correlation between each rainfall in SR and RP i With RP obj The similarity between them is weighted and summed, and the calculation formula is as follows:

[0033]

[0034] Among them, ω1 and ω2 are the weights of association and similarity respectively.

[0035] Step S54: Calculate the recommendation score for each flood process in SF according to the above calculation method. FPi , filter out the topK floods with the highest recommendation as the flood recommendation list, and recommend their corresponding feature information, including flood duration, peak flow, flood rise rate, etc. to users. Flood recommendation list rec_list = {{FP1, FF1}, {FP2, FF2}, ..., {FP topK , FF topK}}It will serve as forecast and warning information on the flood development trend most likely to be caused by the target rainfall.

[0036] Beneficial effect: The application scenario of the present invention is intelligent recommendation of flood forecast. The focus of the present invention is to mine rainfall-flood correlation patterns from historical rainfall and flood data, establish a correlation pattern library, and combine the intelligent recommendation algorithm with the hydrological forecast scenario. Through the steps of target rainfall feature extraction, rainfall similarity search, construction of a multi-level rainfall-flood correlation matrix, and recommendation degree calculation, the intelligent recommendation of flood forecast based on the rainfall-flood correlation pattern library is realized. It can make recommendations on flood development trends according to historical scenarios, and provide a fast and intelligent solution for flood forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is an overall flow chart of a flood forecast intelligent recommendation method based on a pattern library of the present invention;

[0038] Figure 2 It is an example of a daily rainfall map of the Jialing River Basin in the embodiment;

[0039] Figure 3 is an example of a historical rainfall-flood correlation pattern in the embodiment;

[0040] Figure 4 is the result of similarity search on target rainfall in the embodiment;

[0041] Figure 5 It is the result of intelligent recommendation on the flood process that may be caused by the target rainfall. DETAILED DESCRIPTION

[0042] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the scope of the embodiments.

[0043] like Figure 1 As shown, the flood forecast intelligent recommendation method based on the pattern library of this embodiment includes the following steps:

[0044] Step S1, constructing a rainfall-flood correlation pattern library: First, preprocessing, feature extraction, and similarity measurement are performed on historical rainfall data and historical flood data to complete the classification of similar rainfall / flood processes, thereby constructing a rainfall pattern library RPD = {{R1:RP 11 , RP 12 ,...}, {R2: RP 21 , RP 22 ,...},...,{R n :RP n1 , RP n2 ,...}} and the flood pattern library FPD = {{F1: FP 11 , FP 12 ,...}, {F2: FP 21 , FP 22 ,...},...,{F m :FP m1 , FP m2 ,...}}, where n represents the number of rainfall patterns and m represents the number of flood patterns;

[0045] Then, the rainfall-flood item set is constructed based on the rainfall-flood spatiotemporal response relationship, and the Apriori algorithm is used to mine the association relationship between the two pattern libraries to construct the rainfall-flood association pattern library RFAD = {R i →F j , R i = {RP il , Rp i2 , ...}, F j ={FP j1 , Fp j2 , ...}, 1≤i≤n, 1≤j≤m}.

[0046] Step S2, dynamically extracting the spatiotemporal characteristics of target rainfall: input the target rainfall event, extract its spatiotemporal characteristics based on image analysis, and obtain the target rainfall event rainfall feature vector RF obj .

[0047] Step S3, rainfall event similarity search: Based on the multi-feature fusion distance measurement algorithm, a similar rainfall set SR of the target rainfall is searched from the rainfall event pattern library.

[0048] Step S4, construct a multi-level rainfall-flood association matrix: First, consider the association between rainfall patterns and flood patterns, define the association degree between each rainfall pattern and flood pattern according to the rainfall → flood association rules mined from the association pattern library, and construct the pattern association matrix PM; then, calculate the rainfall pattern R in PM that corresponds to the target rainfall. obj The flood mode with the highest correlation is F obj , and based on the historical rainfall-flood spatiotemporal response relationship, the historical correlation matrix HM is constructed to form a multi-level rainfall-flood correlation matrix MAM.

[0049] Step S5, calculate the recommendation degree of the flood process: in the rainfall-flood association matrix MAM, define the recommendation degree calculation method for each flood process; perform weighted summation of the association degree and similarity to obtain the recommendation degree value of each item in the matrix and sort them, and recommend the characteristic information (flood duration, peak flow, increase, etc.) of the topK flood processes as a flood recommendation list to users, so as to provide decision support for the forecast and warning of flood processes that may be caused by the current rainfall.

[0050] Step S6: dynamically optimize and update the rainfall pattern library and flood forecast recommendation results according to the current rainfall change trend, that is, repeat the process of step S2 to step S5 until the rainfall process ends.

[0051] Figure 2 This is an example of rainfall image data used in this embodiment, namely, the daily rainfall map of the Jialing River Basin on September 8, 2015.

[0052] Figure 3 The spatiotemporal response relationship between rainfall patterns and flood patterns in historical scenarios is shown. The rainfall-flood item set can be obtained by matching the rainfall pattern library and the rainfall-flood pattern library through the sliding window technology and DDTW algorithm. For example, the rainfall in the Jialing River Basin from June 22 to June 30, 2015 belongs to rainfall pattern 1, which triggered the flood process numbered 201501, which belongs to flood pattern 5. The historical rainfall-flood item set is used as input, and the rainfall→flood association rules are mined through the association rule mining algorithm to construct the rainfall-flood association pattern library.

[0053] Figure 4 The results of similarity search for target rainfall are shown. Assume that the target rainfall is the rainfall from September 4 to September 6, 2010, which belongs to rainfall pattern 0. The multi-feature fusion distance measurement algorithm is used to calculate the similarity between the target rainfall and each rainfall in the rainfall pattern library, and the most similar rainfall process is searched as the similar rainfall set of the target rainfall.

[0054] Figure 5 is the recommended result of the flood process most likely to be caused by the target rainfall. A multi-level rainfall-flood association matrix is ​​constructed to calculate the recommendation degree of each flood process; first, in the pattern association matrix, the recommendation degree of each flood mode is calculated and sorted, and the flood mode with the highest recommendation degree is flood mode 0, that is, rainfall mode 0→flood mode 0 is the rainfall-flood association mode with the highest probability of occurrence; then, a historical association matrix is ​​constructed, and the recommendation degree of each flood in flood mode 0 is calculated by weighted summation of the association degree and similarity, and the results are sorted to obtain the three flood processes with the highest recommendation degrees in historical scenarios, namely 201201 flood, 201701 flood and 201603 flood, as the results of intelligent recommendation for flood forecasting.

Claims

1. A flood forecast intelligent recommendation method based on a pattern library, characterized in that: The following steps are involved: Step S1, constructing a rainfall-flood correlation pattern library: preprocessing, feature extraction, similarity measurement, and cluster analysis are performed on historical rainfall data and historical flood data to complete the rainfall pattern library RPD = {{R1: RP 11 , RP 12 ,…}, {R2: RP 21 , RP 22 ,…},...,{R n :RP n1 , RP n2 ,…}} and the flood pattern library FPD = {{F1: FP 11 , FP 12 ,…}, {F2: FP 21 , FP 22 ,…},...,{F m :FP m1 , FP m2 ,…}}; n represents the number of rainfall patterns in the rainfall pattern library, and m represents the number of flood patterns in the flood pattern library; based on the spatiotemporal characteristics of rainfall and flood, the rainfall-flood item set is constructed, and the Apriori algorithm is used to mine the association relationship between the two pattern libraries, so as to construct the rainfall-flood association pattern library RFAD={R i →F j ,R i = {RP i1 , Rp i2 , ...}, F j ={FP j1 , Fp j2 , ...}, 1≤i≤n, 1≤j≤m}. Step S2, dynamically extracting the spatiotemporal characteristics of target rainfall: input the target rainfall event, extract its spatiotemporal characteristics based on image analysis, and obtain the target rainfall event rainfall feature vector RF obj . Step S3, rainfall event similarity search: Based on the multi-feature fusion distance measurement algorithm, search from the rainfall event pattern library RPD obtained in step 1 to obtain a similar rainfall set SR of the target rainfall. Step S4, construct a multi-level rainfall-flood association matrix: First, consider the association relationship between the overall patterns. According to the rainfall → flood association rules mined in the association pattern library in step 1, define the association degree between each rainfall pattern and flood pattern, and construct the pattern association matrix PM. By calculation, obtain the rainfall pattern R to which the target rainfall belongs in PM. obj The flood mode with the highest correlation is F obj ,According to the rainfall-flood relationship between historical events, the historical correlation matrix HM is constructed, thus forming a multi-level rainfall-flood correlation matrix MAM. Step S5, calculate the recommendation degree of the flood process: define the recommendation degree calculation method in the multi-level rainfall-flood association matrix MAM, perform weighted summation of the association degree and similarity, obtain the recommendation degree value of each item in the matrix, sort the results, and recommend the characteristic information (flood duration, peak flow, increase, etc.) of the topK flood processes to users, so as to provide decision support for the forecast and warning of flood processes that may be caused by the current rainfall. Step S6: dynamically optimize and update the rainfall pattern library and flood forecast recommendation results according to the changing trend of the current rainfall process, that is, repeat the process of steps S2 to S5 until the rainfall process ends.

2. The flood forecast intelligent recommendation method based on pattern library according to claim 1 is characterized in that: The detailed process of constructing the rainfall-flood correlation model library in step S1 is as follows: Step S11, segment the historical rainfall image data into rainfall events by using the rainfall threshold δ and the sliding window to obtain the historical rainfall event set RS; extract the image features of each rainfall event in RS, including the cumulative rainfall CP, the number of rainy days N, the rainfall time distribution SD, the rainfall spatial distribution TD and the rainfall center motion trajectory TC, and for the kth rainfall event, form the rainfall event feature vector RF k ={CP k ,N k , S.D. k , TD k , TC k Based on the multi-feature fusion distance measurement algorithm, the K-means algorithm based on the Davies-Bouldin index is used to cluster similar rainfall processes, and the rainfall pattern library RPD is obtained. 11 , RP 12 ,…}, {R2: RP 21 , RP 22 , ...}, ..., {R n :RP n1 , RP n2 ,...}}. Step S12: Segment the historical flood flow data based on the time span and flow span to obtain the flood set FS; extract the features of each flood in FS, including flood duration FT, peak flow PF, rising water rate RS, peak type N, etc. For the kth flood, form the flood feature vector FF k ={FT k , PF k ,RS k , N k ,…}; Similarly, based on the distance measurement algorithm of multi-feature fusion, similar flood processes are clustered to obtain the flood pattern library FPD={{F1:FP 11 , FP 12 ,...}, {F2: FP 21 , FP 22 ,...},...,{F m :FP m1 , FP m2,… }}. Step S13: Based on the rainfall pattern library RPD and flood pattern library FPD constructed in steps S11 and S12, combined with the spatiotemporal relationship between rainfall and flood, a rainfall-flood item set IS is constructed as input based on a sliding window and DDTW algorithm. According to the characteristics and statistical laws of the current watershed dataset, the minimum support and minimum confidence are set, and the Apriori algorithm is used to mine the strong association rules of rainfall → flood that meet the minimum support and confidence, and to construct the rainfall-flood association pattern library RFAD = {R i →F j ,R i = {RP i1 , Rp i2 , ...}, F j ={FP j1 , Fp j2 , …}, 1≤i≤n, 1≤j≤m}.

3. The flood forecast intelligent recommendation method based on pattern library according to claim 1 is characterized by: The detailed process of step S2 dynamically extracting the spatiotemporal characteristics of target rainfall is as follows: Step S21: Assuming that the current time is the tth day, determine whether the rainfall on the t-1st or t-2nd day belongs to the most recent rainfall process RP in the rainfall pattern library. last If yes, and the rainfall on that day exceeds the rainfall threshold δ, then the rainfall on that day is considered to belong to RP last As part of the rainfall process, the rainfall process is merged and the RP in the rainfall model library is updated. last Feature information RF obj ; Otherwise, execute step S22. Step S22: The current rainfall is a new rainfall RP obj , use color histogram technology to extract its image features, Including the cumulative rainfall CP, the number of rainy days N, the rainfall time distribution SD, the rainfall spatial distribution TD and the rainfall center motion trajectory TC, the real-time rainfall feature vector RF of the target rainfall is obtained obj ={CP obj ,N obj ,SD obj ,TD obj ,TC obj }, and obtain the rainfall pattern R corresponding to the target rainfall event obj , and add the above characteristic information to the rainfall pattern library.

4. The flood forecast intelligent recommendation method based on pattern library according to claim 1 is characterized by: The detailed process of constructing a multi-level rainfall-flood correlation matrix in step S4 is as follows: Step S41: mining the historical rainfall pattern set HS from the rainfall pattern library RPD and the flood pattern library FPD respectively R = {R1, R2, ..., R n } and flood mode set HS F ={F1, F2, ..., F m }; Mining the historical rainfall-flood correlation pattern set HS from the rainfall-flood correlation pattern library RFAD R→F ={{R i →F j ,Prob Ri→Fj }, 1≤i≤n, 1≤j≤m}. Step S42: construct a pattern association matrix PM. R is the horizontal axis, HS F As the ordinate, establish an n*m pattern association matrix PM. Any element PM in the matrix ij Indicates rainfall pattern R i With flood mode F j The degree of correlation between PM ij The calculation formula is as follows: Step S43, traverse the pattern association matrix PM to obtain the rainfall pattern R to which the target rainfall belongs obj The flood mode with the highest correlation is F p ; Search flood mode F from flood mode library p The corresponding flood process set SF = {FP p1 , FP p2 , ..., FP pk }; For each rainfall in the rainfall pattern library, search the rainfall-flood item set IS obtained in step S13 = {{RP1, FP1}, {RP2, FP2}, ..., {RP n ,FP n }}, the corresponding relationship between the historical rainfall process and the flood process is obtained, taking the i-th rainfall RP i For example, FP i The flood process caused by it. Step S44: construct the historical association matrix HM. The historical association matrix HM is established with SR as the horizontal coordinate and SF as the vertical coordinate. Any element in the matrix HM ij Represents the rainfall process RP i FP with flood process j The degree of correlation between HM ij The calculation formula is as follows: After the above steps, a multi-level rainfall-flood association matrix MAM can be obtained.

5. The flood forecast intelligent recommendation method based on pattern library according to claim 1 is characterized by: The detailed process of calculating the flood process recommendation degree in step S5 is as follows: Step S51, calculate the recommendation degree of each flood process in the rainfall-flood association matrix MAM, the calculation method is as follows: for the i-th flood FP i , search from the rainfall-flood item set IS to find out which rainfall PP it is caused by i The target rainfall RP is calculated by multi-feature distance measurement algorithm. obj With RP i The similarity between them is sim(obj,i). Step S52: Search flood process FP from rainfall-flood association matrix MAM i and rainfall pattern R obj The correlation degree between each rainfall process in is obtained by M }, M is the number of rainfall events in SR. Step S53: define the method for calculating the recommendation score, and convert FP i Correlation with each rainfall, and RP i Follow RP obj The similarity between them is weighted and summed, and the calculation formula is as follows: ω1 and ω2 are the weights of association and similarity respectively. Step S54: Sort the flood processes in SF according to the recommendation degree from large to small, select the topK flood processes with the highest recommendation degree, extract the corresponding flood duration, peak flow, flood rise rate and other characteristic information, and form a flood recommendation list rec_list = {{FP1, FF1}, {FP2, FF2}, ..., {FP topK ,FF topK }} and recommend it to users to form forecast and warning information on the flood development trend that is most likely to be caused by the current rainfall.

Images