Key parameter extraction method based on flood change influence factor synergistic effect
By using the synergistic effect of flood change influence factors, the LSTM network and lag term analysis are used to solve the subjectivity and uncertainty of the flood event division method, and the efficient and accurate extraction of key flood parameters is achieved, and the understanding and prediction ability of flood change laws is improved.
Patent Information
- Application Number
- CN202510709811.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-29
- Publication Date
- 2025-09-02
AI Technical Summary
The existing flood event classification methods have strong subjectivity, incomplete flood behavior characteristic indicators, inaccurate analysis of impact factors, and large errors in uncertainty analysis methods, resulting in inaccurate extraction of key parameters for flood changes and low efficiency.
Using a method based on the synergistic effect of flood change influence factors, we use the LSTM network to perform timing prediction and hysteresis analysis, obtain the weights of each influence factor, eliminate discrete values, and determine key parameters.
It improves the accuracy and efficiency of the extraction of key parameters of flood change, reduces the error caused by the time span, and enhances the understanding and prediction accuracy of the flood change laws.
Smart Images
Figure CN120579162A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of flood occurrence key parameter extraction, and in particular relates to a key parameter extraction method based on the synergistic effect of flood change influencing factors. Background Art
[0002] Flood damage is becoming increasingly serious. Accurately extracting key parameters of flood changes is crucial for flood monitoring, forecasting, early warning, and flood prevention and disaster reduction. However, there are currently several problems and challenges in this area, as follows:
[0003] Flood event classification methods have flaws: Accurately classifying flood processes within a continuous runoff sequence is fundamental to studying flood variability. However, existing classification methods often rely on rainfall data and expert experience to manually determine flood processes. This approach is highly subjective, inefficient for screening large flood samples, and lacks accuracy.
[0004] The existing flood behavior indicator system is imperfect: The existing flood behavior indicator system is mainly divided into three categories: hydrometeorological indicators, hydrological indicators, and indicators based on flood processes. The first two categories mainly consider the factors that cause floods, while flood process-based indicators currently focus on flood magnitude characteristics, such as flood peaks, flood volumes, and runoff depths, ignoring the characteristic elements of the flood hydrograph. This leads to varying degrees of compression and loss of basin flood characteristic information, making it impossible to fully and accurately depict flood changes.
[0005] Inaccurate analysis of influencing factors: Floods are influenced by numerous factors, including meteorological and geographical factors, human activities, and socioeconomic factors, such as precipitation, temperature, wind speed, topography, land use change, urbanization rate, and population density. Each factor contributes to varying degrees of influence on flood processes, making it difficult to accurately identify the dominant controlling factor and extract its key parameters. This hinders our understanding and precise grasp of flood dynamics.
[0006] Uncertainty analysis methods need improvement: Floods are random, requiring uncertainty analysis. However, this is currently primarily done through non-statistical methods, which suffer from large errors and cannot yield reliable flood uncertainty analysis results, thus affecting the accuracy and reliability of extracting key parameters affecting flood fluctuations.
[0007] Lack of efficient data processing and analysis technologies: With the development of monitoring technology, the amount of flood-related data collected, including hydrological data, meteorological data, and geographic information data, is becoming increasingly massive. Rapidly and accurately extracting useful information from this massive amount of data, as well as efficiently processing and analyzing this data to identify key parameters that influence flood changes, are also key challenges. Traditional data processing and analysis methods often fail to meet practical needs, necessitating the use of advanced information technology and data analysis methods. Summary of the Invention
[0008] In response to the above-mentioned deficiencies in the prior art, the present invention provides a key parameter extraction method based on the synergistic effect of flood change influencing factors, which solves the problem of incorrect key parameter identification caused by the sporadic nature of single flood events in the prior art.
[0009] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a key parameter extraction method based on the synergistic effect of flood change influencing factors, comprising:
[0010] Determine the set of factors influencing flood changes and obtain the set of historical flood events in the study basin;
[0011] Based on the flood change influencing factor set, obtain the historical data set of each influencing factor;
[0012] Determine the analysis window size, and intercept the historical flood event set based on the analysis window to obtain a window data set in descending order of flood occurrence frequency;
[0013] Based on the window data set and the historical data set of each impact factor, the weight of each impact factor under each window data is obtained respectively;
[0014] Determine the key parameters based on the weights of each influencing factor under each window data.
[0015] The beneficial effects of the present invention are as follows: the present invention selects flood events over a large span of time in the basin, and based on the analysis window, focuses from global features to local features, obtains the weights of each influencing factor in each window in turn, and then amplifies from local features (i.e. window data) to global features, reducing the amount of data processing over a large span of time and avoiding fitting errors caused by too long a time span.
[0016] Furthermore, the window data set is obtained as follows:
[0017] According to the occurrence time of each historical flood event in the historical flood event set, a flood event curve is constructed with time as the independent variable and the number of flood events as the dependent variable within the study period of the study basin; the starting time of the study period is the occurrence time of the earliest historical flood event, and the end time is the occurrence time of the latest historical flood event;
[0018] Based on the analysis window, the flood event curve is divided into several non-overlapping segments to obtain the window dataset.
[0019] The beneficial effects of the above further solution are: avoiding directly processing the global characteristics of the historical flood event set, reducing the amount of data processing, and avoiding errors caused by large spans.
[0020] Furthermore, the weight of each impact factor under each window data is obtained, specifically:
[0021] Determine the current window data;
[0022] Extract the impact factor historical data tuple from each impact factor historical data set based on the start and end time of the current window data;
[0023] Use the LSTM network to perform time series prediction on the data that is 0 in the historical data tuple of the impact factor to obtain the complete historical data tuple of each impact factor;
[0024] Based on the complete historical data tuples of each impact factor, the lag term is introduced to obtain the characteristic data of the impact factor at each time;
[0025] Fit the characteristic data of influencing factors to obtain the fitting relationship between each influencing factor and flood occurrence;
[0026] According to the fitting relationship between each influencing factor and flood occurrence, the weight of each influencing factor under the current window data is determined;
[0027] Return to get the weight of each influencing factor under the next window data, until the weight of each influencing factor under all window data is obtained.
[0028] The beneficial effects of the above further scheme are: data completion is performed based on LSTM for time series prediction to avoid data shortage; at the same time, a lag term is introduced to take into account the current influencing factors on the probability of future flood disasters, thereby improving the fitting accuracy.
[0029] Furthermore, the impact factor historical data tuple:
[0030]
[0031] T i,s ≤t i ≤T i,e
[0032] Among them, X i is the historical data tuple of impact factor; t i For time; t i At time T, the historical data of the jth impact factor;i,s is the starting time of the current window data; T i,e is the end time of the current window data; j is the total number of influencing factors.
[0033] The beneficial effect of the above further solution is: preparing data for subsequent fitting.
[0034] Furthermore, the expression of the influencing factor characteristic data at each time is:
[0035]
[0036] Among them, X(t i ) is t i Impact factor characteristic data under time; x j (t i -m) is the jth impact factor at t i -The value at time m; m is the maximum lag order.
[0037] The beneficial effect of the above further solution is that it takes into account the lag terms in different situations and introduces the superposition effect of time on disaster-causing.
[0038] Furthermore, the label of the influencing factor characteristic data at each time is Y(t i ); if t i Time of flood, Y(t i )=1, otherwise, Y(t i )=0.
[0039] The beneficial effect of the above further solution is to prepare for subsequent fitting data.
[0040] Furthermore, the fitting relationship between the influencing factors and flood occurrence is:
[0041]
[0042] Where P(Y(t)=1) is the probability of flood occurrence at time t; Y(t) is the label at time t, which is 1 if a flood occurs and 0 otherwise; t is time; β0 is the intercept term; β j,k is the regression coefficient of the jth influencing factor at lag k; k is the lag order; x j (tk) is the value of the jth influencing factor at time tk.
[0043] The beneficial effect of the above further solution is that the fitting accuracy is improved by taking into account the fitting relationship of different influencing factors and different hysteresis conditions.
[0044] Furthermore, the expression of the weight of each influencing factor under the current window data is:
[0045]
[0046] Among them, S j is the weight of the jth influencing factor under the current window data; β j,k is the regression coefficient of the jth influencing factor at the kth lag; β r,k is the regression coefficient of the rth influencing factor at the kth lag; ε is the significance threshold.
[0047] The beneficial effect of the above further solution is to eliminate non-significant lag terms and avoid redundant bias of lag terms.
[0048] Furthermore, the key parameters are determined according to the weights of the influencing factors under the data of each window, specifically:
[0049] According to the weight of each influencing factor under each window data, the discrete values of the weights of each influencing factor under different window data are eliminated, and the weight mean of each influencing factor is calculated based on the data after eliminating the discrete values;
[0050] The influencing factors whose weight mean is greater than the weight threshold are determined as key parameters.
[0051] The beneficial effects of the above further solution are: amplifying local features to global considerations, eliminating discrete values, and avoiding errors caused by occasional flood events. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION
[0053] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.
[0054] like Figure 1 As shown in Figure 1, a key parameter extraction method based on the synergistic effect of flood change influencing factors includes:
[0055] Determine the set of factors influencing flood changes and obtain the set of historical flood events in the study basin;
[0056] Based on the flood change influencing factor set, obtain the historical data set of each influencing factor;
[0057] Determine the analysis window size, and intercept the historical flood event set based on the analysis window to obtain a window data set in descending order of flood occurrence frequency;
[0058] Based on the window data set and the historical data set of each impact factor, the weight of each impact factor under each window data is obtained respectively;
[0059] Determine the key parameters based on the weights of each influencing factor under each window data.
[0060] In this example, the historical flood event set is preferably selected over a long time period to avoid errors caused by sporadic floods. To reduce the complexity of data processing, the long-span event set is divided into several smaller segments. Each segment is processed separately, focusing on local features and then performing global processing based on the local results.
[0061] The window data set is obtained as follows:
[0062] According to the occurrence time of each historical flood event in the historical flood event set, a flood event curve is constructed with time as the independent variable and the number of flood events as the dependent variable within the study period of the study basin; the starting time of the study period is the occurrence time of the earliest historical flood event, and the end time is the occurrence time of the latest historical flood event;
[0063] Based on the analysis window, the flood event curve is divided into several non-overlapping segments to obtain the window dataset.
[0064] The weight of each impact factor under each window data is obtained as follows:
[0065] Determine the current window data;
[0066] Extract the impact factor historical data tuple from each impact factor historical data set based on the start and end time of the current window data;
[0067] Use the LSTM network to perform time series prediction on the data that is 0 in the historical data tuple of the impact factor to obtain the complete historical data tuple of each impact factor;
[0068] Based on the complete historical data tuples of each impact factor, the lag term is introduced to obtain the characteristic data of the impact factor at each time;
[0069] Fit the characteristic data of influencing factors to obtain the fitting relationship between each influencing factor and flood occurrence;
[0070] According to the fitting relationship between each influencing factor and flood occurrence, the weight of each influencing factor under the current window data is determined;
[0071] Return to get the weight of each influencing factor under the next window data, until the weight of each influencing factor under all window data is obtained.
[0072] In this embodiment, due to the relatively large time span, some influencing factor data are very scarce. In this case, it is possible to choose to exclude such influencing factors from the analysis of their impact on the current window data, that is, exclude them from the fitting process.
[0073] The impact factor historical data tuple:
[0074]
[0075] T i,s ≤t i ≤T i,e
[0076] Among them, X i is the historical data tuple of impact factor; t i For time; t i At time T, the historical data of the jth impact factor; i,s is the starting time of the current window data; T i,e is the end time of the current window data; j is the total number of influencing factors.
[0077] The expression of the influencing factor characteristic data at each time is:
[0078]
[0079] Among them, X(t i ) is t i Impact factor characteristic data under time; x j (t i -m) is the jth impact factor at t i -The value at time m; m is the maximum lag order.
[0080] The label of the influencing factor characteristic data at each time is Y(t i ); if t i Time of flood, Y(t i )=1, otherwise, Y(t i )=0.
[0081] The fitting relationship between the influencing factors and flood occurrence is:
[0082]
[0083] Where P(Y(t)=1) is the probability of flood occurrence at time t; Y(t) is the label at time t, which is 1 if a flood occurs and 0 otherwise; t is time; β0 is the intercept term; β j,k is the regression coefficient of the jth influencing factor at lag k; k is the lag order; x j(tk) is the value of the jth influencing factor at time tk.
[0084] The expression of the weight of each influencing factor under the current window data is:
[0085]
[0086] Among them, S j is the weight of the jth influencing factor under the current window data; β j,k is the regression coefficient of the jth influencing factor at the kth lag; β r,k is the regression coefficient of the rth influencing factor at the kth lag; ε is the significance threshold.
[0087] The key parameters are determined based on the weights of the influencing factors under the data of each window, specifically:
[0088] According to the weight of each influencing factor under each window data, the discrete values of the weights of each influencing factor under different window data are eliminated, and the weight mean of each influencing factor is calculated based on the data after eliminating the discrete values;
[0089] The influencing factors whose weight mean is greater than the weight threshold are determined as key parameters.
Claims
1. A key parameter extraction method based on the synergistic effect of flood change influencing factors, characterized in that: include: Determine the set of factors influencing flood changes and obtain the set of historical flood events in the study basin; Based on the flood change influencing factor set, obtain the historical data set of each influencing factor; Determine the analysis window size, and intercept the historical flood event set based on the analysis window to obtain the window data set; Based on the window data set and the historical data set of each impact factor, the weight of each impact factor under each window data is obtained respectively; Determine the key parameters based on the weights of each influencing factor under each window data.
2. The key parameter extraction method based on the synergistic effect of flood change influencing factors according to claim 1 is characterized in that: The window data set is obtained as follows: According to the occurrence time of each historical flood event in the historical flood event set, a flood event curve is constructed with time as the independent variable and the number of flood events as the dependent variable within the study period of the study basin; the starting time of the study period is the occurrence time of the earliest historical flood event, and the end time is the occurrence time of the latest historical flood event; Based on the analysis window, the flood event curve is divided into several non-overlapping segments to obtain the window dataset.
3. The key parameter extraction method based on the synergistic effect of flood change influencing factors according to claim 1 is characterized in that: The weight of each impact factor under each window data is obtained as follows: Determine the current window data; Extract the impact factor historical data tuple from each impact factor historical data set based on the start and end time of the current window data; Use the LSTM network to perform time series prediction on the data that is 0 in the historical data tuple of the impact factor to obtain the complete historical data tuple of each impact factor; Based on the complete historical data tuples of each impact factor, the lag term is introduced to obtain the characteristic data of the impact factor at each time; Fit the characteristic data of influencing factors to obtain the fitting relationship between each influencing factor and flood occurrence; According to the fitting relationship between each influencing factor and flood occurrence, the weight of each influencing factor under the current window data is determined; Return to get the weight of each influencing factor under the next window data, until the weight of each influencing factor under all window data is obtained.
4. The key parameter extraction method based on the synergistic effect of flood change influencing factors according to claim 3 is characterized in that: The impact factor historical data tuple: T i,s ≤t i ≤T i,e Among them, X i is the historical data tuple of impact factor; t i For time; t i At time T, the historical data of the jth impact factor; i,s is the starting time of the current window data; T i,e is the end time of the current window data; j is the total number of influencing factors.
5. The key parameter extraction method based on the synergistic effect of flood change influencing factors according to claim 3 is characterized in that: The expression of the influencing factor characteristic data at each time is: Among them, X(t i ) is t i Impact factor characteristic data under time; x j (t i -m) is the jth impact factor at t i -The value at time m; m is the maximum lag order.
6. The key parameter extraction method based on the synergistic effect of flood change influencing factors according to claim 5 is characterized in that: The label of the influencing factor characteristic data at each time is Y(t i ); if t i Time of flood, Y(t i )=1, otherwise, Y(t i )=0.
7. The key parameter extraction method based on the synergistic effect of flood change influencing factors according to claim 3 is characterized in that: The fitting relationship between the influencing factors and flood occurrence is: Where P(Y(t)=1) is the probability of flood occurrence at time t; Y(t) is the label at time t, which is 1 if a flood occurs and 0 otherwise; t is time; β0 is the intercept term; β j,k is the regression coefficient of the jth influencing factor at lag k; k is the lag order; x j (tk) is the value of the jth influencing factor at time tk.
8. The key parameter extraction method based on the synergistic effect of flood change influencing factors according to claim 3 is characterized in that: The expression of the weight of each influencing factor under the current window data is: Among them, S j is the weight of the jth influencing factor under the current window data; β j,k is the regression coefficient of the jth influencing factor at the kth lag; β r,k is the regression coefficient of the rth influencing factor at the kth lag; ε is the significance threshold.
9. The key parameter extraction method based on the synergistic effect of flood change influencing factors according to claim 1 is characterized in that: The key parameters are determined based on the weights of the influencing factors under the data of each window, specifically: According to the weight of each influencing factor under each window data, the discrete values of the weights of each influencing factor under different window data are eliminated, and the weight mean of each influencing factor is calculated based on the data after eliminating the discrete values; The influencing factors whose weight mean is greater than the weight threshold are determined as key parameters.