A flood similarity prediction method based on rainstorm flood knowledge graph

Abstract

The present invention discloses a flood similarity prediction method based on a knowledge graph of rainstorms and floods, comprising the following steps: step 1, establishing a knowledge graph of rainstorms and floods in the study area; step 2, establishing a rainstorm and flood characteristic index system; step 3, constructing a rainstorm and flood scenario library based on a distributed model; step 4, performing similarity discrimination of rainstorm and flood weighting classification; step 5, real-time flood rolling correction and extrapolation prediction. The method described in the present invention completes the real-time flood key scenario retrieval and reasoning based on the knowledge graph of rainstorms and floods, realizes the intelligent forecasting of rainstorms and floods, improves the accuracy of flood forecasting, promotes the development of flood process prediction technology based on historical rainstorm and flood knowledge, and can also promote the automation and intelligent processing of water conservancy data, thereby promoting the modernization and precision of water conservancy in the water conservancy industry, and has strong application value.

Description

Technical Field

[0001] The present invention belongs to the technical field of flood forecasting and prediction, and in particular relates to a flood similarity estimation method based on a rainstorm flood knowledge graph. Background Art

[0002] Flood forecasting is an important part of flood control engineering measures. Flood forecasting methods can generally be divided into two categories: hydrological model approach and data-driven approach. In the past, the hydrological model approach was the main approach. With the continuous accumulation of hydrological data and the continuous advancement of big data analysis technology, the data-driven approach has received more and more attention. In particular, with the successful application and continuous deepening of artificial intelligence technology in the water conservancy industry, forecasting methods based on hydrological data mining have gradually become a research hotspot. Hydrological knowledge mining and forecasting based on the similarity principle is one of the important research directions.

[0003] At present, there are many forecasting methods based on similarity analysis of rainstorm and flood knowledge, such as principal component analysis, grey correlation analysis, correlation coefficient method and weighted distance coefficient method. Theoretically, if there is no natural climate change and human activities, the hydrological sequence meets the consistency. The longer the hydrological data series, the more rainstorm and flood knowledge that can be mined, and the higher the forecast accuracy based on similarity analysis. However, due to the change in the underlying surface conditions under the changing environment, the hydrological series with a long time span shows inconsistency, which affects the prediction accuracy of flood forecasts based on similar knowledge. Therefore, how to improve the prediction accuracy of flood forecasts based on similar knowledge is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to provide a flood similarity prediction method based on a rainstorm and flood knowledge graph to solve the above technical problems.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] The present invention discloses a flood similarity prediction method based on a rainstorm and flood knowledge graph, the method comprising the following steps:

[0007] Step 1: Establish a knowledge graph of rainstorm and flood in the study area:

[0008] Collect and organize historical rainstorm and flood data of the basin in the study area, and express the rainstorm and flood data into structured data, semi-structured data and unstructured data for computer storage; based on the above data, establish a rainstorm and flood knowledge map of the basin in the study area;

[0009] Step 2: Establish a rainstorm and flood characteristic index system:

[0010] Based on the rainstorm and flood data of the study area, combined with the topography and spatial scale characteristics, characteristic indicators are extracted to establish a rainstorm and flood characteristic indicator system based on knowledge graph;

[0011] Step 3: Build a rainstorm and flood scenario database based on a distributed model:

[0012] A flood forecasting hydrological model suitable for the basin characteristics of the study area was selected, and the previous rainfall P, the pre-rainfall water level H of large and medium-sized reservoirs, the rainfall type T, and the rainfall amplification ratio R were selected as knowledge elements. The knowledge elements were used as the input of the flood forecasting hydrological model, and the model was run to generate the corresponding scenarios.

[0013] Using the database as the carrier for establishing the scenario library, Scenario is recorded as the scenario, and the scenario composition can be expressed as:

[0014] Scenario = {T, P, R, H}

[0015] According to different combinations of T, P, R and H values, multiple different scenarios are generated, and a rainstorm flood scenario library is constructed accordingly;

[0016] Step 4: Conduct similarity determination of rainstorm and flood weighting system classification:

[0017] Based on the weighting of rainstorm flood characteristic indicators, combined with the Euclidean distance discrimination method, dynamic time bending discrimination method, and Borda counting method, the current rainstorm flood and the rainstorm flood in the scenario library are classified and similarly judged by weighting system, and the most similar historical rainstorm flood is calculated;

[0018] Step 5: Real-time flood rolling correction and extrapolation forecast:

[0019] A consistency analysis is conducted on the rainstorm and flood data in the study area to determine whether the rainstorm and flood data meet the consistency requirements. For rainstorm and flood processes that meet the consistency requirements, the most similar historical rainstorm and flood extrapolation estimates are directly transferred; for rainstorm and flood processes that do not meet the consistency requirements, they are corrected and then similar flood extrapolation estimates are conducted; by repeating the above process at each forecast time, the current real-time flood process can be continuously rolled out to obtain real-time dynamic prediction results.

[0020] Furthermore, the historical rainstorm and flood data in step 1 include an entity structure consisting of rainfall events, water conservancy projects, and flood events, as well as relationship characteristics between entities and attribute values. The attribute values ​​of the rainfall events include the location of the center of the rainstorm, the average total rainfall on the surface, the previous rainfall, and the rainfall process. The attribute values ​​of the water conservancy projects include the reservoir water level. The attribute values ​​of the flood events include the flood rising flow, flood rise rate, flood volume, and flood process.

[0021] Furthermore, the characteristic indicators described in step 2 include the location of the center of the rainstorm, the average total rainfall, the previous influencing rainfall, the rainfall process, the reservoir water level, the flood rising flow, the flood rise rate, the flood volume, and the flood process.

[0022] Furthermore, the specific method of selecting the flood forecasting hydrological model suitable for the basin characteristics of the study area in step 3 is to select a full-storage runoff hydrological model for humid areas and to select an excess-seepage runoff hydrological model for arid areas.

[0023] Furthermore, the weighted classification similarity determination of the current rainstorm flood and the rainstorm flood in the scenario library described in step 4 specifically includes the following steps:

[0024] Step 4.1, based on the availability of data and the importance of indicators, the extracted rainstorm flood characteristic indicators are classified, and each type of indicator represents a characteristic of the rainstorm flood;

[0025] Step 4.2, similarity discrimination is performed on the current rainstorm flood and all N rainstorm floods in the scenario library, and the discrimination method adopts the Euclidean distance discrimination method or the dynamic time warping discrimination method;

[0026] Step 4.3, based on the results of each type of characteristic index, the similarity of n rainstorm floods in the scenario library is scored by the Borda counting method. The higher the similarity, the higher the score;

[0027] Step 4.4, weight the scores of each type of feature indicators, and assign higher weights to feature indicators with higher importance, and the sum of the total weights is equal to 1;

[0028] Step 4.5, use the Borda counting method to sort and score, and you can get the corresponding Borda score matrix:

[0029]

[0030] in, In the formula, is the wave arrival score of the corresponding attribute sequence of the nth flood in the scenario library when the current flood is q, n = 1, ..., N, and the total wave arrival count score of N rainstorm floods is calculated. The one with the highest total score is the most similar flood obtained by weighted classification similarity judgment.

[0031] Furthermore, the Euclidean distance determination method described in step 4.2 is specifically to calculate the shortest straight line distance between two points in space, and the point a(x 1,1 ,x 2,1 ,…,x m,1 ) and point b(x 1,2 ,x 2,2 ,…,x m,2 ) between12 for:

[0032]

[0033] Calculate the shortest Euclidean distance D=min(d 12 ,d 13 ,…,d 1n ) is the flood with the highest similarity;

[0034] The dynamic time bending discrimination method is specifically to match the data points of the time series by bending the time domain of the time series. For two flood process flow sequences of unequal lengths, let U = (u1, u2, ..., u l ,…,u L |max(t,j h )≤L≤t+j h -1) is a curved path in two sequences, where u l =(t,j h ) t is the lth element of the curved path, represented by s k (q,t) * and k (h,j h ) * The corresponding relationship between them must satisfy the following three constraints:

[0035] (1) Boundary conditions:

[0036] The curved path must start at point [s k (q,1) * ,s k (h,1) * ], finally click [s k (q,t) * ,s k (h,j h ) * ], that is, the path has two boundaries: the starting point and the end point;

[0037] (2) Monotonicity:

[0038] For u l =(t,j h ) and u l+1 =(t′,j′ h ) two points, must satisfy t′≥t,j′ h ≥j h , that is, the two points before and after the path are monotonic in sequence number;

[0039] (3) Continuity:

[0040] For u l =(t,j h ) and u l+1 =(t′,j′ h ) two points, must satisfy t′≤t′+1, j h ≤j h +1, that is, any two adjacent points in the path are continuous;

[0041] D DTW It is the shortest path among all the curved paths, D DTW It is expressed as:

[0042] Calculate D DTW , we can get the flood with the highest similarity to the current flood among all N rainstorm floods.

[0043] Furthermore, the specific method for determining whether the rainstorm and flood data satisfy consistency in step 5 is: statistically analyzing the "rainfall-peak flow" or "rainfall-runoff depth" relationship of all rainstorm and flood data, and observing whether the relationship curve undergoes a sudden change in a certain year. If no sudden change occurs, the consistency is satisfied; if a sudden change occurs, the consistency is not satisfied.

[0044] Furthermore, the specific method of correcting the rainstorm and flood process that does not meet the consistency in step 5 is:

[0045] 1) If the most similar flood in history is the flood before the mutation, then based on the relationship between “rainfall-peak flow” or “rainfall-runoff depth”, the flood reduction or expansion coefficient k before and after the mutation year is calculated, and then the multiple ratio relationship between the current flood and the most similar flood in history is established:

[0046]

[0047] or

[0048] In the formula, k is the reduction or expansion factor, Q m (current) is the peak flow of the current flood, Q m (historical) is the peak discharge of the most similar flood in history, R(current) is the runoff depth of the current flood, and R(historical) is the runoff depth of the most similar flood in history;

[0049] 2) If the most similar historical flood is identified as a flood after a sudden change, the flood process is reduced or expanded based on the relationship between rainfall and runoff depth:

[0050]

[0051] Where P (当前) and Pa (当前) are the rainfall of the current flood and the previous rainfall index, P (历史) and P a (历史) are the rainfall and antecedent rainfall index of the most similar flood in history, W0 (当前) and W2 (当前) are the runoff of the basin outlet section before and after the forecast time t0 of the current flood, W0 (历史) and W1 (历史) are the runoff of the basin outlet section before and after the forecast time t0 of the most similar flood in history, α (当前) is the runoff coefficient of the current flood, α (历史) is the runoff coefficient of the most similar flood in history.

[0052] The beneficial effects of the present invention are as follows: a flood similarity prediction method based on a rainstorm and flood knowledge graph provided by the present invention completes real-time flood key scenario retrieval and reasoning based on the rainstorm and flood knowledge graph, realizes intelligent forecasting of rainstorms and floods, improves the accuracy of flood forecasting, and promotes the development of flood process prediction technology based on historical rainstorm and flood knowledge. At the same time, it can promote the automation and intelligent processing of water conservancy data, thereby promoting the modernization and precision of water conservancy in the water conservancy industry, and has strong application value.

[0053] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the process of the method of the present invention;

[0055] Figure 2 It is a line diagram of 71 historical rainstorm and flood processes in Mengyin Basin in Example 1;

[0056] Figure 3 The rainstorm and flood knowledge graph of Example 1;

[0057] Figure 4 This is a schematic diagram of similarity discrimination of rainstorm and flood weighting classification in Example 1;

[0058] Figure 5 The relationship diagram of "rainfall-peak flow" and "rainfall-runoff depth" of the Mengyin Basin in Example 1;

[0059] Figure 6 This is a diagram showing the similarity extrapolation prediction results of a current rainstorm flood in Example 1. DETAILED DESCRIPTION

[0060] The present invention discloses a flood similarity prediction method based on a rainstorm flood knowledge graph, such as Figure 1 As shown, the method comprises the following steps:

[0061] Step 1: Establish a knowledge graph of rainstorm and flood in the study area:

[0062] Taking the historical rainstorm and floods in the study area as the research object, the historical rainstorm and flood data were collected and organized, including the entity structure composed of rainfall events, water conservancy projects, and flood events, as well as the relationship characteristics between entities and attribute values. Among them, the attribute values ​​of rainfall events include the location of the center of the rainstorm, the average total rainfall on the surface, the previous rainfall, the rainfall process, etc. The attribute values ​​of water conservancy projects include the reservoir water level, etc. The attribute values ​​of flood events include flood rising flow, flood rise rate, flood volume, flood process, etc.

[0063] The collected and organized historical rainstorm and flood data are expressed as structured data, semi-structured data and unstructured data for computer storage.

[0064] The historical rainstorm and flood data currently collected may have missing data for some years and ambiguous data naming, which can be solved through methods such as data interpolation and extension, and merging and eliminating ambiguous naming.

[0065] Based on the above data, a knowledge graph of rainstorm and flood in the study area is established, and effective storage mode, query optimization mode and human-computer interaction mode are designed to support the effective management of the knowledge graph of rainstorm and flood.

[0066] Step 2: Establish a rainstorm and flood characteristic index system:

[0067] According to the rainstorm and flood data of the study area, combined with the characteristics of topography and spatial scale, characteristic indicators are extracted, and a rainstorm and flood characteristic indicator system based on knowledge graph is established. The characteristic indicators include the location of the rainstorm center, the average total rainfall, the previous influencing rainfall, the rainfall process, the reservoir water level, the flood rising flow, the flood rise rate, the flood volume, the flood process, etc.

[0068] Step 3: Build a rainstorm and flood scenario database based on a distributed model:

[0069] Select a flood forecasting hydrological model suitable for the basin characteristics of the study area, such as a full-storage runoff hydrological model for humid areas and an over-seepage runoff hydrological model for arid areas; select previous rainfall P, pre-rainfall water level H of large and medium-sized reservoirs, rainfall type T, and rainfall amplification ratio R as knowledge elements, use the knowledge elements as inputs to the flood forecasting hydrological model, and run the model to generate corresponding scenarios. Different scenarios will be generated when the knowledge element values ​​are different; describe the scenario elements through the description of the knowledge elements, and finally describe the changes in scenario content through the layered associations between the knowledge elements;

[0070] Using the database as the carrier for establishing the scenario library, Scenario is recorded as the scenario, and the scenario composition can be expressed as:

[0071] Scenario = {T, P, R, H}

[0072] According to different combinations of T, P, R and H values, multiple different scenarios are generated, and a rainstorm and flood scenario library is constructed accordingly.

[0073] Step 4: Conduct similarity determination of rainstorm and flood weighting system classification:

[0074] The weighting of rainstorm and flood characteristic indicators is based on rainstorm similarity as the main and flood similarity as the auxiliary, that is, the weight of rainfall characteristic indicators is greater and the weight of flood characteristic indicators is smaller; combined with the Euclidean distance discrimination method, dynamic time bending discrimination method, Borda counting method and other criteria, the current rainstorm and flood are classified and similarly judged by weighted classification with the rainstorm and flood in the scenario library, and the most similar historical rainstorm and flood are calculated.

[0075] The specific steps are as follows:

[0076] Step 4.1: Based on the availability of data and the importance of indicators, the extracted rainstorm flood characteristic indicators are classified. Each type of indicator represents an important characteristic of the rainstorm flood.

[0077] Step 4.2, the current rainstorm flood is similar to all N rainstorm floods in the scenario library, and the discrimination method adopts the Euclidean distance discrimination method or the dynamic time warping discrimination method.

[0078] 1) Euclidean distance is the most intuitive distance measurement method, that is, the shortest straight line distance between two points in space. 1,1 ,x 2,1 ,…,x m,1 ) and point b(x 1,2 ,x 2,2 ,…,x m,2 ) between 12 for:

[0079]

[0080] Calculate the shortest Euclidean distance D=min(d 12 ,d 13 ,…,d 1n ) is the flood with the highest similarity.

[0081] 2) Dynamic time warping discrimination method is a similarity measurement method that combines time regularization and distance measurement calculation. It can match the data points of the time series by warping the time domain of the time series. It can not only obtain better morphological measurement effects, but also measure two time series of unequal lengths.

[0082] For two flood flow sequences of unequal length, let U = (u1, u2, ..., u l ,…,u L |max(t,j h )≤L≤t+j h -1) is a curved path in two sequences, where u l =(t,j h ) t is the lth element of the curved path, represented by s k (q,t) * and k (h,j h ) * The corresponding relationship between them must satisfy the following three constraints:

[0083] (1) Boundary conditions:

[0084] The curved path must start at point [s k (q,1) * ,s k (h,1) * ], finally click [s k (q,t) * ,s k (h,j h ) * ], that is, the path has two boundaries: the starting point and the end point;

[0085] (2) Monotonicity:

[0086] For u l =(t,j h ) and u l+1 =(t′,j′ h ) two points, must satisfy t′≥t,j′ h ≥j h , that is, the two points before and after the path are monotonic in sequence number;

[0087] (3) Continuity:

[0088] For u l =(t,j h ) and u l+1 =(t′,j′ h ) two points, must satisfy t′≤t′+1, j h ≤jh +1, that is, any two adjacent points in the path are continuous;

[0089] D DTW It is the shortest path among all the curved paths, D DTW It is expressed as:

[0090] Therefore, D DTW , we can get the flood with the highest similarity to the current flood among all N rainstorm floods.

[0091] In step 4.3, based on the identification results of each type of characteristic index, the similarity of n rainstorm floods in the scenario library is scored by the Borda counting method. The higher the similarity, the higher the score.

[0092] In step 4.4, weight the scores of each type of feature indicators. Higher weights are assigned to similar indicators with higher importance, and the sum of the total weights must be equal to 1.

[0093] Step 4.5, use the Borda counting method to sort and score, and you can get the corresponding Borda score matrix:

[0094]

[0095] in,

[0096] In the formula, is the wave arrival score of the corresponding attribute sequence of the nth flood in the scenario library when the current flood is q, n = 1, ..., N, and the total wave arrival count score of N rainstorm floods is calculated. The one with the highest total score is the most similar flood obtained by weighted classification similarity judgment.

[0097] Step 5: Real-time flood rolling correction and extrapolation forecast:

[0098] Consider and analyze the "rainfall-peak flow" or "rainfall-runoff depth" relationship before and after the variation point (a certain time position) under the changing environment, based on the most similar rainstorm and flood events identified, after the "peak-volume" joint correction and elimination of inconsistency effects, real-time rolling extrapolation is used to estimate future flood processes.

[0099] A consistency analysis was conducted on the rainstorm and flood data in the study area. The specific method was to statistically analyze the "rainfall-peak flow" or "rainfall-runoff depth" relationship of all rainstorm and flood data, and to determine whether the rainstorm and flood data met the consistency requirements by observing whether the relationship curve changed suddenly in a certain year. If no sudden change occurred, the data met the consistency requirements; if a sudden change occurred, the data did not meet the consistency requirements.

[0100] For the rainstorm and flood processes that meet the consistency, the most similar historical rainstorm and flood extrapolation prediction can be directly transferred. However, in the changing environment, the hydrological series often cannot meet the consistency requirements. After the rainstorm and flood processes that do not meet the consistency are corrected, similar flood extrapolation predictions are performed.

[0101] The specific correction process is:

[0102] 1) If the most similar flood in history is the flood before the mutation, then based on the relationship between “rainfall-peak flow” or “rainfall-runoff depth”, the flood reduction or expansion coefficient k before and after the mutation year is calculated, and then the multiple ratio relationship between the current flood and the most similar flood in history is established:

[0103]

[0104] or

[0105] In the formula, k is the reduction or expansion factor, Q m (current) is the peak flow of the current flood, Q m R(historical) is the peak flow of the most similar flood in history, R(current) is the runoff depth of the current flood, and R(historical) is the runoff depth of the most similar flood in history.

[0106] 2) If the most similar historical flood is identified as a flood after a sudden change, the flood process is reduced or expanded based on the relationship between rainfall and runoff depth:

[0107]

[0108] Where P (当前) and P a (当前) are the rainfall of the current flood and the previous rainfall index (reflecting the previous soil wetness), P (历史) and P a (历史) are the rainfall of the most similar flood in history and the previous rainfall index (reflecting the previous soil wetness), W0 (当前) and W2 (当前) are the runoff of the basin outlet section before and after the forecast time t0 of the current flood, W0 (历史) and W1 (历史) are the runoff of the basin outlet section before and after the forecast time t0 of the most similar flood in history, α (当前) is the runoff coefficient of the current flood, α (历史) is the runoff coefficient of the most similar flood in history.

[0109] By repeating the above process at each forecast time, the current real-time rainstorm and flood process can be continuously extrapolated, the flow at future forecast times can be quickly predicted, and the real-time dynamic prediction results of the flood process at the hydrological station in the study area can be obtained.

[0110] Embodiment 1

[0111] This embodiment is a specific application example of the above method.

[0112] This embodiment discloses a flood similarity prediction method based on a rainstorm and flood knowledge graph, comprising the following steps:

[0113] Step 1: Establish a knowledge graph of rainstorm and flood in the study area:

[0114] This example uses the Mengyin River Basin in Shandong Province as the study area and collects and organizes 71 historical rainstorm and flood data in the Mengyin River Basin. Figure 2 The specific flood events and flood numbers are shown in Table 1.

[0115] Table 1 Numbers and flood numbers of 71 historical rainstorms and floods in Mengyin Basin

[0116]

[0117] like Figure 3 As shown, the rainstorm and flood data include an entity structure consisting of rainfall events, water conservancy projects, and flood events, as well as relationship characteristics between entities and attribute values. Among them, the attribute values ​​of rainfall events include the location of the center of the rainstorm, the average total rainfall on the surface, the previous rainfall, the rainfall process, etc. The attribute values ​​of water conservancy projects include the reservoir water level, etc. The attribute values ​​of flood events include the flood rising flow, flood rise rate, flood volume, flood process, etc.

[0118] The collected rainstorm flood data are expressed as structured data, semi-structured data and unstructured data for computer storage.

[0119] The rainstorm and flood data currently collected may have problems such as missing data for some years and ambiguous data naming, which can be solved through methods such as data interpolation and extension, and merging and eliminating ambiguous naming.

[0120] Based on the above data, a knowledge graph of rainstorm and flood in Mengyin Basin, Shandong Province was established, and effective storage mode, query optimization mode and human-computer interaction mode were designed to support the effective management of the knowledge graph of rainstorm and flood.

[0121] Step 2: Establish a rainstorm and flood characteristic index system:

[0122] Based on the rainstorm and flood data of Mengyin Basin in Shandong Province, and taking into account the topography and spatial scale characteristics of Mengyin Basin, nine characteristic indicators are extracted, including the location of the rainstorm center, the average total rainfall, the previous influencing rainfall, the rainfall process, the reservoir water level, the flood rising flow, the flood rise rate, the flood volume and the flood process, and a rainstorm and flood characteristic indicator system based on the knowledge graph is established.

[0123] Step 3: Build a rainstorm and flood scenario database based on a distributed model:

[0124] In this embodiment, the Xin'anjiang River Basin hydrological model of three water sources considering the influence of small and medium-sized reservoirs is selected to construct the rainstorm and flood scenario library of the Mengyin Basin. The model parameters of the hydrological model are 17, of which there are 6 initial state variables, namely the upper, lower and deep tension water content at the initial moment, the free water content of the basin, the virtual reservoir capacity of small reservoirs, and the capacity of large and medium-sized reservoirs.

[0125] The previous rainfall P, the pre-rainfall water level H of large and medium-sized reservoirs, the rainfall type T and the rainfall amplification ratio R are selected as knowledge elements. The knowledge elements are used as the input of the hydrological model, and the corresponding scenarios are generated by running the model.

[0126] Using the database as the carrier for establishing the scenario library, Scenario is recorded as the scenario, and the scenario composition can be expressed as:

[0127] Scenario = {T, P, R, H}

[0128] Where H is the water level of Huangtushan Reservoir before the rain.

[0129] In this embodiment, according to different combinations of T, P, R and H values, a total of 1.98 million scenarios are generated, based on which a rainstorm and flood scenario library is constructed.

[0130] Step 4: Conduct similarity discrimination of rainstorm and flood weighting system classification, which specifically includes the following steps:

[0131] Step 4.1, this embodiment divides the selected 9 representative rainstorm and flood characteristic indicators into 5 categories, the first category is the rainstorm center location, the second category is the surface average rainfall and the previous impact rainfall, the third category is the rainfall process, the fourth category is the flood index (including the rising flow, flood rising rate, flood volume, and reservoir water level), and the fifth category is the flood process;

[0132] Step 4.2: The current rainstorm flood and all 1.98 million rainstorm floods in the scenario database are similarly judged one by one. For the similarity judgment, the first, second and fourth category characteristic indicators are judged by the Euclidean distance method (the fourth category judges the four flood indicators at the same time), and the third and fifth category characteristic indicators are judged by the dynamic time warping method;

[0133] Step 4.3: Based on the results of each type of indicator, the similarity of the 1.98 million rainstorm floods is scored using the Borda counting method. The more similar the floods, the higher the score. The highest score for the most similar floods is 1,980,001 points, and the lowest score for the least similar floods is 1 point.

[0134] Step 4.4, assign weights to the scores of each characteristic index. The first three types of indicators represent the similarity of "quantity and type" of rainfall process, and the last two types of indicators represent the similarity of "quantity and type" of flood process. This time, the similarity of rainfall process is more important, so the weights assigned to these five types of characteristic indicators are 0.25, 0.25, 0.4, 0.05, and 0.05 respectively (0.25+0.25+0.4+0.05+0.05=1);

[0135] Step 4.5, calculate the total score of the boda counts of 1.98 million rainstorm floods. The third flood with the highest total score (i.e., 1,980,000 points) is the most similar flood based on the weighted classification similarity judgment, as shown in Figure 4 shown.

[0136] Step 5: Real-time flood rolling correction and extrapolation forecast:

[0137] The consistency analysis of the rainstorm and flood data in the Mengyin Basin shows that Figure 5 From the statistical graphs of the relationship between "(P+Pa)-Qm (rainfall-peak flow)" and "(P+Pa)-R (rainfall-runoff depth)", it can be seen that there was an obvious mutation in the consistency around 1980 and the consistency was not met. Therefore, it is necessary to correct the rainstorm flood process in the Mengyin Basin before extrapolating and estimating similar floods.

[0138] The third flood, which is the most similar one, identified in step 4 is the flood before the mutation. Therefore, the multiple ratio relationship between the current flood and the most similar flood in history is established based on the relationship between “rainfall-peak flow” and “rainfall-runoff depth” for correction.

[0139] Repeat the above steps to estimate the actual rainstorm flood process currently occurring, and obtain the following Figure 6 The scenario library shown retrieves similar flood and real-time dynamic forecast results from 6h to 8h.

[0140] Finally, it should be noted that the above is only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred arrangement scheme, a person skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A flood similarity prediction method based on a rainstorm flood knowledge graph, characterized in that: The method comprises the following steps: Step 1: Establish a knowledge graph of rainstorm and flood in the study area: Collect and organize historical rainstorm and flood data of the basin in the study area, and express the rainstorm and flood data into structured data, semi-structured data and unstructured data for computer storage; based on the above data, establish a rainstorm and flood knowledge map of the basin in the study area; Step 2: Establish a rainstorm and flood characteristic index system: Based on the rainstorm and flood data of the study area, combined with the topography and spatial scale characteristics, characteristic indicators are extracted to establish a rainstorm and flood characteristic indicator system based on knowledge graph; Step 3: Build a rainstorm and flood scenario database based on a distributed model: A flood forecasting hydrological model suitable for the basin characteristics of the study area was selected, and the previous rainfall P, the pre-rainfall water level H of large and medium-sized reservoirs, the rainfall type T, and the rainfall amplification ratio R were selected as knowledge elements. The knowledge elements were used as the input of the flood forecasting hydrological model, and the model was run to generate the corresponding scenarios. Using the database as the carrier for establishing the scenario library, Scenario is recorded as the scenario, and the scenario composition can be expressed as: Scenario = {T, P, R, H} According to different combinations of T, P, R and H values, multiple different scenarios are generated, and a rainstorm flood scenario library is constructed accordingly; Step 4: Conduct similarity determination of rainstorm and flood weighting system classification: Based on the weighting of rainstorm flood characteristic indicators, combined with the Euclidean distance discrimination method, dynamic time bending discrimination method, and Borda counting method, the current rainstorm flood and the rainstorm flood in the scenario library are classified and similarly judged by weighting system, and the most similar historical rainstorm flood is calculated; Step 5: Real-time flood rolling correction and extrapolation forecast: A consistency analysis is conducted on the rainstorm and flood data in the study area to determine whether the rainstorm and flood data meet the consistency requirements. For rainstorm and flood processes that meet the consistency requirements, the most similar historical rainstorm and flood extrapolation estimates are directly transferred; for rainstorm and flood processes that do not meet the consistency requirements, they are corrected and then similar flood extrapolation estimates are conducted; by repeating the above process at each forecast time, the current real-time flood process can be continuously rolled out to obtain real-time dynamic prediction results.

2. The flood similarity prediction method based on the rainstorm and flood knowledge graph according to claim 1 is characterized in that: The historical rainstorm and flood data described in step 1 include an entity structure consisting of rainfall events, water conservancy projects, and flood events, as well as relationship characteristics between entities and attribute values. The attribute values ​​of the rainfall events include the location of the center of the rainstorm, the average total rainfall on the surface, the previous rainfall influencing the rainfall, and the rainfall process. The attribute values ​​of the water conservancy projects include the reservoir water level. The attribute values ​​of the flood events include the flood rising flow, flood rise rate, flood volume, and flood process.

3. The flood similarity prediction method based on the rainstorm and flood knowledge graph according to claim 1 is characterized in that: The characteristic indicators described in step 2 include the location of the center of the rainstorm, the average total rainfall, the previous rainfall, the rainfall process, the reservoir water level, the flood flow, the flood rate, the flood volume, and the flood process.

4. The flood similarity prediction method based on the rainstorm and flood knowledge graph according to claim 1 is characterized in that: The specific method of selecting the flood forecasting hydrological model suitable for the basin characteristics of the study area described in step 3 is to select a full-storage runoff hydrological model for humid areas and a super-seepage runoff hydrological model for arid areas.

5. The flood similarity prediction method based on the rainstorm and flood knowledge graph according to claim 1 is characterized in that: The weighted classification similarity determination of the current rainstorm flood and the rainstorm flood in the scenario library described in step 4 specifically includes the following steps: Step 4.1, based on the availability of data and the importance of indicators, the extracted rainstorm flood characteristic indicators are classified, and each type of indicator represents a characteristic of the rainstorm flood; Step 4.2, similarity discrimination is performed on the current rainstorm flood and all N rainstorm floods in the scenario library, and the discrimination method adopts the Euclidean distance discrimination method or the dynamic time warping discrimination method; Step 4.3, based on the identification results of each type of characteristic indicators, the similarity of n rainstorm floods in the scenario library is scored by the Borda counting method. The higher the similarity, the higher the score; Step 4.4, weight the scores of each type of feature indicators, and assign higher weights to feature indicators with higher importance, and the sum of the total weights is equal to 1; Step 4.5, use the Borda counting method to sort and score, and you can get the corresponding Borda score matrix: in, In the formula, is the wave arrival score of the corresponding attribute sequence of the nth flood in the scenario library when the current flood is q, n = 1, ..., N, and the total wave arrival count score of N rainstorm floods is calculated. The one with the highest total score is the most similar flood obtained by weighted classification similarity judgment.

6. The flood similarity prediction method based on the rainstorm and flood knowledge graph according to claim 5 is characterized in that: The Euclidean distance judgment method described in step 4.2 is specifically to calculate the shortest straight line distance between two points in space. 1,1 ,x 2,1 ,…,x m,1 ) and point b(x 1,2 ,x 2,2 ,…,x m,2 ) between 12 for: Calculate the shortest Euclidean distance D=min(d 12 ,d 13 ,…,d 1n ) is the flood with the highest similarity; The dynamic time bending discrimination method is specifically to match the data points of the time series by bending the time domain of the time series. For two flood process flow sequences of unequal lengths, let U = (u1, u2, ..., u l ,…,u L |max(t,j h )≤L≤t+j h -1) is a curved path in two sequences, where u l =(t,j h ) t is the lth element of the curved path, represented by s k (q,t) * and k (h,j h ) * The corresponding relationship between them must satisfy the following three constraints: (1) Boundary conditions: The curved path must start at point [s k (q,1) * ,s k (h,1) * ], finally click [s k (q,t) * ,s k (h,j h ) * ], that is, the path has two boundaries: the starting point and the end point; (2) Monotonicity: For u l =(t,j h ) and u l+1 =(t′,j′ h ) two points, must satisfy t′≥t,j′ h ≥j h , that is, the two points before and after the path are monotonic in sequence number; (3) Continuity: For u l =(t,j h ) and u l+1 =(t′,j′ h ) two points, must satisfy t′≤t′+1, j h ≤j h +1, that is, any two adjacent points in the path are continuous; D DTW It is the shortest path among all the curved paths, D DTW It is expressed as: Calculate D DTW , we can get the flood with the highest similarity to the current flood among all N rainstorm floods.

7. The flood similarity prediction method based on the rainstorm and flood knowledge graph according to claim 1 is characterized in that: The specific method for judging whether the rainstorm and flood data meet the consistency in step 5 is: statistically analyzing the "rainfall-peak flow" or "rainfall-runoff depth" relationship of all rainstorm and flood data, and observing whether the relationship curve changes suddenly in a certain year. If no sudden change occurs, the consistency is met; if a sudden change occurs, the consistency is not met.

8. The flood similarity prediction method based on the rainstorm and flood knowledge graph according to claim 7 is characterized in that: The specific method for correcting the rainstorm flood process that does not meet the consistency described in step 5 is: 1) If the most similar flood in history is the flood before the mutation, then based on the relationship between "rainfall-peak flow" or "rainfall-runoff depth", the flood process reduction or expansion coefficient k before and after the mutation year is calculated, and then the multiple ratio relationship between the current flood and the most similar flood in history is established: or In the formula, k is the reduction or expansion factor, Q m (current) is the peak flow of the current flood, Q m (historical) is the peak discharge of the most similar flood in history, R(current) is the runoff depth of the current flood, and R(historical) is the runoff depth of the most similar flood in history; 2) If the most similar historical flood is identified as a flood after a sudden change, the flood process is reduced or expanded based on the "rainfall-runoff depth" relationship: Where P (当前) and are the rainfall of the current flood and the previous rainfall index, P (历史) and are the rainfall and antecedent rainfall index of the most similar flood in history, W0 (当前) and W2 (当前) are the runoff of the basin outlet section before and after the forecast time t0 of the current flood, W0 (历史) and W1 (历史) are the runoff of the basin outlet section before and after the forecast time t0 of the most similar flood in history, α (当前) is the runoff coefficient of the current flood, α (历史) is the runoff coefficient of the most similar flood in history.

Images