CN111199298B - Flood forecasting method and system based on neural network - Google Patents
Flood forecasting method and system based on neural network Download PDFInfo
- Publication number
- CN111199298B CN111199298B CN201811454308.8A CN201811454308A CN111199298B CN 111199298 B CN111199298 B CN 111199298B CN 201811454308 A CN201811454308 A CN 201811454308A CN 111199298 B CN111199298 B CN 111199298B
- Authority
- CN
- China
- Prior art keywords
- flood
- data
- neural network
- forecast
- rainfall
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/02—Reservations, e.g. for tickets, services or events
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Abstract
The invention discloses a real-time flood forecasting method based on continuous multi-early-period data recursion of a neural network, which is characterized in that on the basis of an established data warehouse taking a study object as a subject, the regional average rainfall calculated by a Thiessen polygon method is added, the original data warehouse is adjusted and expanded, the water rain condition and hydraulic engineering scheduling data of a plurality of continuous flood fields in a plurality of early periods after transformation according to the idea of a flood forecasting synthetic flow method in the data warehouse are innovatively adopted as data analysis units to carry out neural network analysis modeling on at least one of the flood peak flow and the water level of the forecast object in the forecast period, and the flood peak flow/water level of the study object in the forecast period is recursively obtained in real time.
Description
Technical Field
The invention relates to a flood forecasting method and system based on a neural network.
Background
Modern flood forecasting technology is based on the existing weather forecasting and hydrologic forecasting theory, and mainly comprises three aspects: firstly, researching a quantitative precipitation forecasting technology capable of meeting special requirements of flood forecasting; secondly, a mode of organically combining quantitative rainfall forecast and flood forecast is established; and thirdly, a real-time flood forecasting method comprises river basin production convergence and river channel flood forecasting. At present, the real-time flood forecasting technology mainly adopts a physical process and a mathematical method to analyze, model and calculate, and corrects the structure, parameters or model output of the model in real time according to the latest monitored rainfall, water level or flow data, so that the forecasting precision of the process flood is continuously improved.
At present, a national hydrologic database is built, various water level and rainfall measuring stations are built to villages in a refined mode, and an Internet technology which is developed rapidly is combined, so that a good platform is provided for hydrologic information inquiry and data analysis.
However, at present, the hydrologic departments in China mainly rely on a flood forecasting system to forecast flood, and most of the hydrologic departments adopt the traditional hydrologic algorithm to forecast flood, and because the system database structure and the algorithm are relatively fixed, even if factors affecting the flood process are received in analysis and calculation, the complicated relationship and the rain conditions, water conditions and work conditions which are very variable are difficult to balance, partial forecasting is easy to have insufficient precision, the forecasting period is not long enough, and forecasting operation time is not rapid enough, so that the requirements of flood prevention decision are difficult to meet.
In a literature published by the inventor of the application, (literature 1, university of Guangxi, shuoshi paper, research and application of Dairy river mountain station flood forecast based on neural network, authors Su Cheng, 12 months in 2017), a flood forecast method is provided and applied to He Jiang river mountain station flood forecast, a certain breakthrough is made in solving the technical problems of the prior art, but the forecast precision still only reaches the level of third, and the improvement needs are still provided.
Disclosure of Invention
The invention aims to provide a flood forecasting method and system based on a neural network, which improve forecasting precision.
Therefore, the flood forecasting method based on the neural network provided by the invention comprises the following steps of: s1, adding the regional average rainfall calculated by a Thiessen polygon method on the basis of a data warehouse taking a research object as a theme; s2, adjusting and expanding the data warehouse; and S3, carrying out neural network analysis modeling by adopting water rain conditions and hydraulic engineering scheduling data of a plurality of flood fields subjected to transformation according to a flood forecast synthetic flow method in a data warehouse, and recursively obtaining at least one of flood peak flow and water level of a study object in a forecast period in real time.
In some embodiments, the method further comprises the following technical characteristics:
in the step S1, a Thiessen polygon net is formed in a full-river basin or a certain area according to a Thiessen polygon method, each Thiessen polygon corresponds to a rainfall observation point in the net, and the rainfall of the rainfall observation point represents the rainfall in the corresponding Thiessen polygon; therefore, the average rainfall in the area is calculated through the rainfall of all rainfall observation points in the network.
The method for forming the Thiessen polygonal net comprises the following steps: firstly, drawing a rain collecting area of a certain rainfall station of a drainage basin required for research on a topographic map, and marking the position of the rainfall station; then, a region map is exported; and finally, constructing a Thiessen polygon of the region map according to the principle and the characteristics of the Thiessen polygon, and obtaining the Thiessen polygon network of the rain-collecting region required by research.
Regional average rainfall
The method can be obtained by the following formula:
wherein n is the number of Thiessen polygons, and x i And A is the rainfall weight coefficient which is the ratio of each corresponding Thiessen polygon to the area of the area for the rainfall of each rainfall observation point.
In step S2, the adjusting and expanding the data warehouse includes: increasing the average rainfall of the area as a main factor affecting the water level of the rainfall station to be forecasted; the upstream reservoir flood discharge flow is adjusted to the delivery flow, delivery flow = flood discharge flow + power generation flow + irrigation flow.
In step S3, according to the synthetic flow method, flow synthesis is performed on each factor according to the propagation time, that is, column displacement transformation is performed on the propagation time corresponding to the event sample data, and the corresponding regional average rainfall is calculated on the rainfall data in the transformed data unit, and the calculated result is used as a new factor to participate in neural network analysis, so that the data unit to be subjected to the neural network analysis shows the actual influence of the water level of the rainfall station to be predicted, and the data of the data unit to be subjected to the neural network analysis is added with the data of the continuous multiple earlier time periods correspondingly transformed according to the synthetic flow method.
In step S3, the number of flood process shots that can be modeled by neural network analysis at a time is not limited.
When n is time, the forecast water level of n+1 is made for the station to be forecasted, and when n+1 is time, the rainfall station data of the station to be forecasted is obtained through weather forecast, so that the forecast with the forecast period of 1 hour is completed; the minimum value of the propagation time of the upstream station is recorded as m, the data units after adjustment and expansion in the future for m hours, namely, the data of continuous multi-early-period corresponding to the factors affecting the forecast object according to the synthetic flow method can be complete, reasonable and effective, and the water level and the flow of the station to be forecast in the time of n+m can be gradually and reasonably recursively estimated, namely, the forecast period is estimated to be m hours; wherein m and n are positive integers.
The method comprises the steps of adopting water rain conditions and hydraulic engineering scheduling data of a plurality of flood fields subjected to transformation according to a flood forecast synthetic flow method and a plurality of continuous earlier time periods as a data analysis unit to conduct neural network analysis modeling on at least one of flood peak flow and water level of a forecast object, directly using hyper-warning flood event sample data which do not participate in flood forecast scheme compiling to conduct preliminary precision inspection on a model obtained through the neural network analysis, conducting complete standard precision assessment and inspection on the model if the inspection result is accurate, conducting repeated modeling, inspection and error correction, properly adjusting the forecast period according to the precision inspection result, and finally obtaining a set of relatively satisfactory flood forecast scheme.
The invention also provides a flood forecasting system based on the neural network, which adopts the method to forecast the flood.
The invention also proposes a computer medium storing a computer program executable to implement the flood forecast method described above.
The invention relates to a neural network-based flood forecasting method and a neural network-based continuous multi-early-stage data recurrence real-time flood forecasting system, which are characterized in that on the basis of an existing data warehouse taking a research object as a subject, the regional average rainfall calculated by a Thiessen polygon method is added, the original data warehouse is adjusted and expanded according to a flood forecasting synthetic flow method, the neural network analysis modeling is performed by innovatively adopting the water rain condition and hydraulic engineering scheduling data of a plurality of continuous early-stage periods in the data warehouse as a data analysis unit, the flood peak flow or/and the water level of the research object in the forecasting period are obtained in a recurrence mode in real time, and the forecasting precision is improved.
In some embodiments, the following benefits are also included:
the regional average rainfall calculated by the Thiessen polygon method is added to provide important basis for flood forecasting and evaluation, and the inventor proves that the real-time and accurate regional average rainfall calculation result plays a vital role in the accuracy of flood forecasting. The rainfall observed by the rainfall station can only represent the rainfall condition of a smaller range around the rainfall station, so that the rainfall data of the individual rainfall station cannot be completely used as the basis for flood forecast and evaluation, and the rainfall data of all the rainfall stations in the whole river basin or a certain area is required to be adopted to calculate the regional average rainfall.
The data warehouse in the prior art is adjusted and expanded, and the data for neural network analysis in the adjusted and expanded data warehouse is finer and relatively increased, so that a more powerful data support is provided for the neural network analysis.
The method has the advantages that the water rain condition and hydraulic engineering scheduling data of a plurality of flood fields subjected to transformation according to the thought of the flood forecast synthetic flow method are used as a data analysis unit for carrying out neural network analysis modeling, so that the flood development process is represented more comprehensively, and the hidden rules of the flood process can be fully mined by the neural network analysis; the probability of unreasonable fitting of analysis performed by using serial numbers to represent flood development process in document 1 is avoided; the number of flood process shots that can be analyzed at a time is not limited.
In one embodiment of the present application, the verification accuracy of the forecast is successfully improved from level C to level B.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention.
Fig. 2 is a schematic diagram of a river basin water system topography applied in an embodiment of the present invention.
Fig. 2a, 2b, 2c are schematic diagrams of construction process of the Thiessen polygon of the rain-collecting area.
Fig. 3 is a causal graph (fishbone graph) of the variation of the water level at the greetings station.
FIG. 4 is a schematic diagram of a multi-layer feedforward neural network having an hidden layer.
FIG. 5A is a schematic diagram of a synthetic flow rate according to an embodiment of the invention.
Fig. 5B is a corresponding neural network relationship diagram.
FIG. 6 is a plot of event sample accuracy assessment for example 2016.5.20 of the present invention.
FIG. 7 is a plot of event sample accuracy assessment for example 2016.6.13 of the present invention.
Fig. 8 is a schematic diagram of the test result of the forecasting scheme of the present embodiment.
Fig. 9 is a schematic diagram of the result of the forecasting scheme of comparative document 1.
Detailed Description
The following embodiments of the present invention are presented in connection with various disciplines such as data mining, neural networks, and flood forecasting.
Data mining is a multidisciplinary intersection domain. On the one hand, to find useful knowledge in a large data set in a special way, data mining must draw nutrition from other disciplines such as statistics, neural networks, information retrieval, high performance computing, etc. On the other hand, other disciplines also require analysis and understanding of data from different perspectives; data mining also provides new opportunities and challenges for the development of these discipline areas (Zhang Xiongling, yang Guan. Application of data mining in accurate prediction of river floods [ J ]. Computer simulation, 2013 (30)). From an information processing perspective, it is more desirable for computers to help analyze data and understand data, helping them make decisions based on rich data. Thus, data mining (finding useful knowledge from large amounts of data in an extraordinary way) is a natural need.
A neural network is a network that simulates the characteristics of biological nervous system activity, and is a mathematical model of extensive and parallel information processing by computing units (neurons). As a common analysis algorithm, the method is integrated into various data mining tools, and has high fault tolerance and learning capability by adjusting the interconnection relationship among a large number of internal nodes according to the complexity degree of the system and the high-speed computing capability of a computer, and the neural network can fully approximate to a complex function or a nonlinear relationship on the premise of given enough hidden units and enough training samples so as to achieve the purpose of processing information. In theory, neural networks can easily solve the problem of hundreds of parameters, providing a relatively simple and efficient method for solving the highly complex problem.
The traditional models of flood forecast are many, they reflect some laws of hydrology, but because the understanding of human beings on hydrological meteorological laws of the flow field is limited, the laws of nature are complex and changeable, human activities also continuously affect the underlying surface of the flow field, and the establishment of various models cannot get rid of various assumptions on simulation generalization of real hydrological phenomena, so that the best model is difficult to comprehensively reflect objective laws.
The real-time flood forecast method based on the continuous multi-early-stage data recursion of the neural network is characterized in that on the basis of an established data warehouse taking a study object as a subject, the regional average rainfall calculated by a Thiessen polygon method is added, the original data warehouse is adjusted and expanded according to a flood forecast synthetic flow method, the neural network analysis modeling is creatively carried out by adopting the water rain condition and hydraulic engineering scheduling data of a plurality of continuous early-stage periods in the data warehouse, and the flood peak flow (water level) of the study object in the forecast period is recursively obtained in real time. The flow chart is shown in figure 1.
The invention will be described in further detail below with reference to the water level forecast of the state station in the river basin.
In recent years, he Jiang river basins have been changed due to production and construction, and the flood forecast results of the hydrologic departments on the next three years of the New year's river basin, the inventor has tried the invention on He Zhou stations, and detailed embodiments are described below.
He Jiang river basin water system (from a tortoise reservoir dam to a Hezhou city section) is shown in fig. 2 (the figure is a landform diagram), event sample data which obviously affects the water level of a Hezhou station after a He Jiang river basin barrage is dismantled (the construction of 2016 is 4 months) is studied based on a neural network in 2017, a flood forecasting scheme with forecasting accuracy of level C is obtained, and the scheme is disclosed in a literature 1 published by the inventor (Guangxi university Studies of Shu: hezhou station flood forecasting study and application of Hezhou river basin based on the neural network, and authors Su Cheng, 2017, 12 months).
The embodiment is based on the paper, makes technical improvement, improves forecasting precision, breaks through the limitation that the neural network in document 1 can only analyze one flood process each time on the basis of fully considering objective rules of flood development, continuously perfects, increases assessment of forecasting timeliness, and proposes and completes the technical scheme and application of the embodiment. The method specifically comprises the following steps:
1. construction of Arcgis-based Thiessen polygons
At present, a plurality of methods for calculating the average rainfall of the area are available, and an arithmetic average method, a numerical method, an isopoint line method, a Thiessen polygon method and the like are commonly used. Among these methods, the Thiessen polygon method is most suitable for the situation that rainfall stations or rainfall distribution in a river basin or an area is uneven, and can greatly improve the accuracy of average rainfall calculation. The rain collecting area at the upstream of the Hezhou station is formed by combining mountain land, small basin land and hills, the terrain is complex, and the average rainfall of the area is calculated by using a Thiessen polygon method.
Thiessen polygons, proposed by the Holland pneumologist A.H. Thiessen, are a method of calculating the average rainfall in an area by the rainfall of discrete rainfall points. The adjacent discrete rainfall points are connected into a triangle, and vertical lines on each side of the triangle are formed, so that a polygon (namely a Thiessen polygon) is surrounded by the vertical lines around each discrete point, and the rainfall in the range of the polygon is represented by the rainfall observed by the unique rainfall in the Thiessen polygon. According to the method, a Thiessen polygon net is formed in a full-river basin or a certain area, each Thiessen polygon corresponds to one rainfall observation point in the net, and the rainfall of the rainfall observation point represents the rainfall in the corresponding Thiessen polygon. Thus, the average rainfall in the area can be calculated by the rainfall at all rainfall observation points in the network.
Let Thiessen polygon number be n, rainfall of each rainfall observation point be x i The ratio of each corresponding Thiessen polygon to the area of the area, namely the rainfall weight coefficient, is A i Average rainfall in the area
The method can be obtained by the following formula:
firstly, drawing a rain collecting area (figure 2 a) of a greetings state station in a greetings river basin required for research on a topographic map, and marking the position of a rainfall station; then the region map is exported (fig. 2 b); finally, according to the principle and the characteristics of the Thiessen polygon, the Arcgis10.5 is used as a platform, and the Thiessen polygon net of the Hezhou station rain collecting area (from the turtle reservoir dam to the Hezhou station section) of the Hejiang river basin required for research is obtained by constructing the Thiessen polygon of the area map, as shown in fig. 2c.
Meanwhile, 9 rainfall weight coefficients A are calculated for 9 rainfall stations in the figure 2c i (ratio of each corresponding Thiessen polygon to area of the region) as in Table 1. The below mentioned rainfall stations will be indicated with corresponding sequence numbers in table 1.
TABLE 1 regional rainfall weight coefficient table calculated by Thiessen polygon method
2. Adjusting and expanding original data warehouse
Original data warehouse creation mainly includes dimension modeling and data preparation, and adopts a star-shaped mode according to main factors (see fig. 3) affecting the water level of the greetings station (the mode is the prior art, and is specifically seen in Korean, data mining: concept and technology (3 rd edition of original book) [ M ]. Beijing: mechanical industry Press 2012.)
According to the dimension modeling method in the document 1, dimension modeling is carried out, a data warehouse is built by taking Microsoft Excel as a platform according to a simple and easy principle, a data structure is unified, and the data warehouse is built by acquiring, cleaning, checking and integrating data. With respect to document 1, the present application provides an adjustment and expansion of a data warehouse. The original data warehouse condition and the adjustment expansion are described as follows:
1. the river dike park construction in He Jiangcheng section of 2015 and the demolition of the upstream barrage (eight-step barrage) of the urban area smart peak bridge at the end of the year 2016 are affected, while the Hezhou station is located at the upstream of the barrage by about 600 meters, and the under-pad surface (factors such as a riverbed) of He Jiang urban area is obviously changed, so that the data warehouse only records the water and rain conditions and hydraulic engineering scheduling data affecting the water level of the Hezhou station since 2016 years in the original data warehouse.
2. Increasing the regional average rainfall is a major factor affecting the water level at the greetings station, as shown in fig. 3.
3. And the flood discharge flow of the tortoise-stone reservoir is adjusted to be the delivery flow, the delivery flow=the flood discharge flow+the power generation flow+the irrigation flow, the data detail is further optimized, and the reliability and the accuracy of analysis are improved.
4. Since 2018, flood water does not occur in the Hezhou station by the beginning of 10 months, the highest peak value is only 101.7m, the difference between the peak value and the warning water level is 103.5m and is 1.8m, the difference between the peak value and the warning water level is 100.8 and is 0.9 m, the research significance is not great, and therefore 2018 related data are not added into a data warehouse.
5. The unit meter of all water level data in the original data warehouse is adjusted to be cm, so that the magnitude relation between the unit meter and the unit millimeter of rainfall data is reduced, and the precision is improved.
3. Establishment of flood forecast scheme based on neural network
3.1 overview
According to the established flood forecasting scheme of the Hezhou station in the Hejiang river basin, according to the basic condition of the upstream of the Hezhou station in the He Jiang river basin and the construction condition of hydraulic engineering facilities, relevant data such as rainfall, water level and the outlet flow of a tortoise stone reservoir required by flood forecasting of the Hezhou station are collected by referring to data sources of flood forecasting of hydrology departments, wherein the rainfall and the water level sources are the urban hydrology bureau, the work is completed in the literature 1, the outlet flow source of the tortoise stone reservoir is the tortoise stone reservoir, the work is completed in the application, and the flood peak flow data of the Hezhou station cannot be collected in time due to time relations, so that the embodiment only forecasts the flood peak water level. However, as long as there is enough data, the method proposed in the present application is equally applicable to peak traffic prediction.
The embodiment is designed to design a plurality of flood forecasting schemes, and the most accurate one of the flood forecasting schemes is selected as a river basin greeting state station flood forecasting scheme through multiple improvements of neural network analysis. And on the premise of ensuring the flood forecasting precision and the reliability, evaluating and checking the He Jiang river basin Hezhou station flood forecasting scheme according to the related requirements of flood forecasting precision evaluation.
3.2 multilayer feedforward neural network
The multi-layer feed forward neural network iteratively learns a set of weights for sample data prediction, which is a class of neural network, including an input layer, one or more hidden layers, and an output layer, as in fig. 4, in general, the number of neurons of the hidden layers is proportional to the factor of the input. The multi-layer feedforward neural network is hereinafter referred to as a neural network.
Two important processes and functions of neural networks are learning and execution. The neural network learns a certain information mode for a plurality of times, and records the information learned for a plurality of times on the link weights of all nodes, so that the neural network is particularly sensitive to the information mode, and when similar information which is learned by the neural network is executed again, a more accurate reasoning conclusion or predicted value can be made.
The working principle of the neural network is as follows: the independent variable (factor) is continuously transmitted from the input layer to the hidden layer and then to the output layer, the hidden layer neuron is equivalent to the adopted statistical analysis model, the processing mode of the received corresponding signal is continuously adjusted, and finally the independent variable output by the output layer becomes a more correct reasoning conclusion or predicted value.
In order to prevent data from being excessively fitted and reasonably split sample data in neural network analysis, a training set and a verification set are formed according to a certain proportion, and the common proportion is 7:3 or 5:5.
3.3 design flood forecast scheme
3.3.1 design specifications
The data warehouse is observed, and the total 3 fields of super-alert floods are observed in the Hezhou station in 2016 years, in the 3 flood events, the outlet flow of the upstream tortoise and stone reservoir reaches 450 cubic meters per second, the flood discharging time is long, the Hezhou station upstream of Hejiang river basin has strong rainfall influence to the tortoise dam section, the sample data is most abundant, and the data are extremely high in analysis value and representativeness and cannot be compared with other events. The adoption of the 3 event sample compiling and inspection forecasting scheme can improve the reliability of the scheme to the greatest extent, so that the following flood forecasting scheme mainly surrounds the 3 events, and the event pairs are shown in table 2.
Table 2 event comparison table
In combination with practical situations, the embodiment adopts 2016-year hyper-warning flood event (first two events in table 2, namely 2016.5.20 and 2016.6.13) sample data to compile a forecasting scheme and evaluate, and adopts 2017-year hyper-warning flood event (third event in table 2, namely 2017.7.2) samples which do not participate in the compiling of the flood forecasting scheme to carry out accuracy inspection on the design scheme.
3.3.2 forecasting methods
The scheme is as follows: on the basis of the data warehouse which is built in the literature 1 and takes the research object as the subject, adding the regional average rainfall calculated by the Thiessen polygon method, adjusting and expanding the original data warehouse according to the thought of the flood forecast synthetic flow method and combining innovation points, innovatively adopting the water rain condition and hydraulic engineering scheduling data of a plurality of continuous earlier periods in the data warehouse as a data analysis unit to carry out neural network analysis modeling, and recursively obtaining the flood peak flow (water level) of the research object in the foreseeing period in real time.
The implementation process comprises the following steps:
1. the data units in the extended data warehouse (i.e., the data units for which neural network analysis is to be performed) are adjusted.
According to the thought of the synthetic flow method, the reservoir outlet flow of the tortoise stone, the rainfall of each rainfall station and the flow (water level) of Zhong Shan stations reach the greetings station with a certain propagation time (as shown in table 3), the flows reaching the greetings station at the same time of each factor are overlapped according to the propagation time to perform flow synthesis, namely, column displacement conversion is performed on unit row data according to the propagation time in a data warehouse, and the corresponding regional average rainfall is calculated on the converted row rainfall data (the calculation result is used as a new factor to participate in neural network analysis), so that each row of data shows the actual influence on the water level of the greetings station. As shown in fig. 5A.
In the embodiment, all time-by-time data of main factors affecting the water level of the greeting station in the first 6 hours after the flow is synthesized are innovatively adopted to forecast the water level of the greeting station in the next hour, namely 15 data units studied in the literature 1 are adjusted to 79 data units (15 data units are used for solving the problem that the independent variable sequence number cannot completely show the influence on the water level of the greeting station, then the influence is deleted), a neural network analysis modeling is carried out by adopting water rain conditions and hydraulic engineering scheduling data of a plurality of continuous earlier periods as data analysis units, the flood development process is represented more comprehensively, so that the neural network analysis can fully excavate the hidden rule of the flood process, the probability that the index 1 represents the flood development process to analyze and unreasonably fit is avoided, the flood process time of each time can be analyzed is not limited, and the data of 13 factor periods (see the synthetic flow diagram in fig. 5) are 6+1 forecast object time period data=79.
Table 3 propagation time table
| Site(s) | Propagation time (hours) |
| Turtle stone reservoir delivery flow | 8 |
| Zhong Shan station flow (Water level) | 6 |
| Rain station for water well | 1 |
| White Sha Yuliang station | 8 |
| He Zhou rainfall station | 0 |
| Zhong Shanyu measuring station | 6 |
| Iron-feces apron rainfall station | 8 |
| Fishpond rainfall station | 7 |
| Gate |
4 |
| High rainfall station | 7 |
| Dam head rainfall station | 6 |
2. The number of columns in the extended data warehouse (i.e., the number of data units to be analyzed by the neural network) is adjusted.
The water rain condition and hydraulic engineering scheduling data of a plurality of continuous earlier time periods are used as data analysis units to carry out neural network analysis modeling, so that the flood process which can be analyzed each time is not limited, and then the data units are integrated into the same table, and 381 columns are added, namely 381 time period data.
Meanwhile, the method is adopted to adjust and expand 2017 hyper-warning flood event sample data, so as to prepare for precision inspection.
3. Planning a foresight period
Under the condition that the data unit is complete, reasonable and effective, the neural network model can make more accurate predictions. As can be seen from fig. 5, the forecast water level at n+1 is made for the greetings station at n, and the greetings station data at n+1 is unknown, but can be obtained through weather forecast, so that the forecast with the forecast period of 1 hour is completed. As can be seen from the observation of Table 3, the propagation time of 8 stations is more than 6 hours, the other 3 stations are all rainfall data, and the rainfall can be obtained through weather forecast, so that the data units of the next 6 hours can be complete, reasonable and effective, and further the water level of the Hezhou station in n+6 hours can be gradually and reasonably recursively estimated, namely, the forecast period is planned to be 6 hours.
4. Neural network analysis
Based on a JMP10 platform of a data mining tool, selecting 2016-year total hyperwarning flood (total 2 times) sample data which are adjusted and expanded through the steps from a data warehouse, loading the sample data into JMP for modeling analysis of a neural network, and outputting dependent variables (response): n+1, the greetings station water level, the input argument (factor): the method comprises the steps of (1) setting the training set and the verification set of a neural network to be 2:1, wherein the training set and the verification set of the neural network are set to be 15 in proportion, namely the number of hidden nodes is set to be 15, according to the n-time Hezhou station water level, (n+1-propagation time) clock mountain station water level, (n+1-propagation time) tortoise-stone ex-warehouse flow, (n+1-propagation time) station data of each rainfall, average rainfall of corresponding areas and the like. Modeling. And directly carrying out preliminary precision inspection on the model obtained through neural network analysis by using 2017 super-alert flood event sample data which does not participate in flood forecasting scheme compiling, and if the inspection result is accurate, carrying out complete standard precision assessment and inspection on the model, repeatedly modeling, inspecting and correcting errors to gradually obtain a set of relatively satisfactory flood forecasting scheme.
5. Determining a forecast scheme
Through inspection, the prediction period is reduced from 6 hours to 5 hours in order to ensure the prediction accuracy. The final forecasting scheme is shown in fig. 5B.
The forecasting formula:
wherein, h is the forecast water level of the next hour's New year station, and the water level of the next 5 hours' New year station can be recursively deduced according to the designed forecast scheme; i represents hidden layer neurons of the neural network, 15 in total; k represents the weight coefficient of the neuron, R is a constant value, and is obtained by utilizing an excavating tool jmp through a neural network; h represents the fitting value of the neuron, which is the hyperbolic tangent tanh function value, is calculated by 78 factors through a neural network by using a mining tool jmp, and has the following formula,
wherein T is a judgment ratio before carrying out tanh function calculation, L is a limiting coefficient, P is a weight coefficient of each factor corresponding to the neuron, and the weight coefficient is obtained by utilizing an excavating tool jmp through a neural network; z is a factor, and m is a factor number. The correspondence is shown in Table 3A below.
TABLE 3A
3.4 precision assessment
According to the standard of hydrologic information forecast Specification (GB/T22482-2008), the accuracy assessment and inspection of the forecast scheme of the embodiment are carried out by adopting a deterministic coefficient, and the flood peak forecast aging is assessed.
3.4.1 assessment and inspection of accuracy of forecast protocol
The certainty factor (DC) represents the coincidence degree between the flood forecasting process and the actual measurement process, and the accuracy of the flood forecasting scheme is classified into three grades A, B and C according to DC >0.9, DC 0.9 is more than or equal to 0.7 and DC 0.7 is more than or equal to 0.5. In the embodiment, the accuracy evaluation and inspection are carried out on the flood forecasting process of the scheme by adopting a deterministic coefficient, and the deterministic coefficient calculation formula is as follows:
wherein: DC-deterministic coefficient (taking 2-bit decimal);
y c (i) -a forecast value;
y 0 (i) -an actual measurement value;
n-the data sequence length.
According to the requirements of the hydrologic information forecast regulations, all data which are involved in flood forecast establishment are selected for assessment, all data which are not involved in flood forecast establishment are selected for inspection, a forecast process is adopted every 5 hours, forecast results are calculated, the assessment results are shown in fig. 6 and 7, wherein the X axis is time sequence, the Y axis is water level, the unit is cm, DC=0.98 in fig. 6, DC=0.98 in fig. 7, and the measured water level is smoother in a curve.
The test result of the forecasting scheme in this embodiment is shown in fig. 8 (all the data not involved in flood forecasting are selected for evaluation), wherein dc=0.82, and the measured water level is smoother in the curve; the prediction scheme of reference 1 is shown in fig. 9, where dc=0.77, and the measured water level is more rounded in the curve. This example is 0.05 higher than document 1, and the predicted peak level and peak occurrence time are more accurate.
In summary, according to the deterministic coefficient formula, the DC values of the two flood processes participating in flood forecast establishment in the embodiment are all 0.98, and the assessment precision reaches the first level; the DC value of the flood process which does not participate in the flood forecast is 0.82, and the inspection precision is level B.
Evaluation of 3.4.2 flood peak forecast aging
The utility timeliness coefficient at the time of flood peak prediction is expressed and calculated according to the following formula: cet=epf/TPF
Wherein:
cet—timeliness coefficient (taking 2-bit decimal);
EPF-effective forecast period [ means the time interval from the time of issuing forecast time to the occurrence of flood peak (or forecast object) of the station, 1 decimal place is taken, and the unit is hours (h);
TPF-theoretical prediction period [ refer to the time interval from the occurrence of the principal rainfall stopping or predicting basis element to the occurrence of the flood peak (or predicting object) of the own station, taking 1 decimal place ], and the unit is hours (h).
The flood peak forecast aging grade of a single river reach (river basin) is determined according to the table 4, when CET >1.00 is advanced forecast, the flood peak forecast is issued when the flood peak forecast basis element does not appear, and the forecast aging is unqualified when the forecast aging does not reach the third grade. The forecast time of the water level flow process can also be determined by the forecast value with the longest forecast period according to the flood peak forecast time grade regulation.
Table 4 table of age level of flood peak forecast for single river reach (river basin)
| Ageing grade | Nail (Rapid) | Second (in time) | Polypropylene (qualification) |
| Coefficient of timeliness | CET≥0.95 | 0.95>CET≥0.85 | 0.85>CET≥0.70 |
Meanwhile, the CET of each aging grade is calculated by taking the following values of a job consumption value dh (comprising water regime information receiving and processing time, dh=TPF-EPF) as an upper limit, namely, grade A is less than or equal to 0.6h, grade B is less than or equal to 0.8h, and grade C is less than or equal to 1.0h.
By contrast with the above regulations, the main content of the forecasting operation of the forecasting scheme is that forecasting basis factors (water level, rainfall and turtle reservoir delivery flow data) at n are substituted into a forecasting formula so as to recursively obtain the water level of the Hezhou station at n+5, and the forecasting value can be obtained by calculating through Excel, writing small programs and a data mining tool and inputting data required by the forecasting formula; the main time consumption of the operation comprises water regime information receiving processing and forecast data analysis processing of three rainfall stations, the time consumption dh of the whole operation process can be controlled below 0.2h (namely 12 minutes), if a small program can be written to be embedded into a mountain torrent disaster monitoring and early warning platform and the forecast data analysis processing work of the three rainfall stations is finished in advance, and the main time consumption of the operation is only the time consumption of the water regime information receiving processing.
To sum up, the TPF of the present forecast scheme is 5h, the operation time dh is 0.2h, the cet=0.96 is calculated, and the aging grade is grade a.
3.4.3 Total analysis
1. Compared with the method disclosed in the document 1, the accuracy test value of the forecasting scheme of the embodiment is improved, the flood peak level and the peak time are more accurate, and particularly, the continuous forecasting of the first flood peak is very accurate, and the flood peak level and the peak time are as shown in the table 5.
Table 5 continuous real-time forecasting of flood peak process by the forecasting scheme of this embodiment
2. According to the embodiment, the water rain condition and hydraulic engineering scheduling data of a plurality of continuous earlier time periods after transformation according to the idea of a flood forecast synthetic flow method are adopted as a data analysis unit to carry out neural network analysis modeling, so that the flood development process is represented more comprehensively, and the implicit rule of the flood process can be fully excavated by the neural network analysis; the probability of unreasonable fitting of analysis performed by using serial numbers to represent flood development process in document 1 is avoided; the number of flood process sessions that can be analyzed each time is not limited.
3. In the embodiment, the forecasting ageing is evaluated on the forecasting scheme, and the forecasting ageing of the first level fully illustrates the high efficiency of the forecasting scheme.
4. The flood forecasting scheme has a forecasting period of 5 hours, the forecasting precision is a grade B to be determined (only the 2017 flood data is adopted for precision inspection, the accuracy inspection is not less than 2 years according to regulations, but the flood is not generated by the 10 th month of 2018 first year, and the highest peak process is not representative), the forecasting timeliness is grade A, and the flood forecasting method can be used for forecasting in real time and for reference forecasting.
According to the embodiment, the study is based on the study of the document 1, the neural network analysis modeling is carried out by taking the water rain condition and hydraulic engineering scheduling data of a single period as the data analysis unit, the neural network analysis modeling is carried out by using the serial number to represent the flood development process on a unilateral basis, the flood process data of only one occasion at a time is analyzed, all the time-by-time data of the main factors affecting the water level of the greetings in the first 6 hours are innovatively adopted to forecast the water level of the greetings in the next hour, the water level of the greetings in the next 5 hours is further recursively obtained (the data in the first 6 hours are input, the forecast water level after 5 hours is output and is the data recorded once per hour), the limitation that the neural network of the document 1 can only analyze one flood process at a time is broken through, the forecast scheme is compiled by adopting the existing flood peak process data of all floods, and the forecast accuracy is improved.
The present embodiment is described by taking a water level forecast as an example, but the method can also be used for flow forecast.
The flood forecast of the embodiment is to cut out the object to be forecasted in a measuring mode, business is used as a drive, data warehouse is used as a basis, deep rules implicit in the flood process of the flood to be forecasted can be extracted from a large amount of water conditions, rain conditions and work condition data affecting the flood peak (water level) of the object to be forecasted, meanwhile, the average rainfall of the area calculated by the Thiessen polygon method is added, the original data warehouse is adjusted and expanded according to the flood forecast synthetic flow method, and the water rain conditions and hydraulic engineering scheduling data of a plurality of flood fields transformed according to the idea of the flood forecast synthetic flow method are innovatively adopted as a data analysis unit to carry out neural network analysis modeling, so that a flood forecast model with higher precision is obtained; by adopting a recurrence algorithm, the flood forecast period is prolonged to a certain extent by combining the actual situation of the forecast object; according to the obtained flood forecast model, a forecast formula is extracted, and the real-time forecast value can be obtained by inputting the data required by the forecast formula, so that the forecast operation time efficiency is greatly improved. The method of the embodiment has higher forecasting precision, longer forecasting period, high forecasting operation time efficiency, real-time forecasting, relatively low application implementation difficulty and shorter period, and can provide more powerful support for flood prevention decision and emergency rescue scheduling. In addition, the method can also be applied to reservoir hydropower station dispatching, thus generating certain economic benefit
The method of the embodiment can be applied to the forecasting of the hydropower stations of sites and reservoirs with comprehensive hydrologic data (water level stations, rainfall stations and hydraulic engineering scheduling data which have main influence on the forecasting objects), can be written into software modules to be embedded into a mountain torrent disaster monitoring and early warning platform, and provides more powerful support for flood prevention decision and emergency rescue scheduling.
Claims (7)
1. The flood forecasting method based on the neural network is characterized by comprising the following steps:
s1, adding the regional average rainfall calculated by a Thiessen polygon method on the basis of a data warehouse taking a research object as a theme;
s2, adjusting and expanding the data warehouse, wherein the adjusting and expanding of row data and column numbers in the data warehouse are included, the row data are data units to be analyzed by the neural network, and the column numbers are the number of the data units to be analyzed by the neural network; the adjusting and expanding the data warehouse comprises the following steps: increasing the average rainfall of the area as a main factor affecting the water level of the rainfall station to be forecasted; adjusting the flood discharge flow of the upstream water reservoir to be a delivery flow, wherein the delivery flow is equal to the flood discharge flow, the power generation flow and the irrigation flow;
s3, carrying out neural network analysis modeling by adopting water rain conditions and hydraulic engineering scheduling data of a plurality of flood fields subjected to transformation according to a flood forecast synthetic flow method in a data warehouse, and recursively obtaining at least one of flood peak flow and water level of a research object in a forecast period in real time;
in the step S3, the number of flood process occasions capable of carrying out neural network analysis modeling each time is not limited;
the forecasting water level of the forecasting object in the n time is made n+1, and the rainfall station data of the forecasting object in the n+1 time is obtained through weather forecasting according to a flood forecasting synthetic flow method, so that forecasting with the forecasting period of 1 hour is completed; the minimum value of the propagation time of the upstream site is m, and the data units after adjustment and expansion in the future m hours, namely the data of continuous multi-early period correspondingly transformed according to the synthetic flow method affecting the factor of the forecast object, can be complete, reasonable and effective, and further can gradually and reasonably recursively estimate the water level and flow of the forecast object when n+m, namely the forecast period is the future m hours, wherein m and n are positive integers;
the water level forecast formula is:
wherein h is the forecast water level of the forecast object in the next hour, and the forecast water level of the forecast object in the forecast period can be recursively deduced according to the designed forecast scheme; i represents the ith hidden layer neuron of the neural network, and q is the number of hidden layer neurons of the neural network; k represents the weight coefficient of the neuron, R is a constant value, and is obtained by utilizing an excavating tool jmp through a neural network; h represents the fitting value of the neuron, which is the hyperbolic tangent tanh function value, and the formula is as follows:
wherein T is a judgment ratio before carrying out tanh function calculation, L is a limiting coefficient, P is a weight coefficient of each factor corresponding to the neuron, and the weight coefficient is obtained by utilizing an excavating tool jmp through a neural network; z is a factor, and m is a factor number.
2. The neural network-based flood forecast method of claim 1, wherein in step S1, a Thiessen polygon network is formed in the full-drainage basin or in a certain area according to the Thiessen polygon method, and each Thiessen polygon corresponds to a rainfall observation point in the network, and the rainfall of the rainfall observation point represents the rainfall in the corresponding Thiessen polygon; therefore, the average rainfall in the area is calculated through the rainfall of all rainfall observation points in the network.
3. The neural network-based flood forecast method of claim 2, wherein the method of constructing a Thiessen polygonal network is: firstly, drawing a rain collecting area of a certain rainfall station of a drainage basin required for research on a topographic map, and marking the position of the rainfall station; then, a region map is exported; and finally, constructing a Thiessen polygon of the region map according to the principle and the characteristics of the Thiessen polygon, and obtaining the Thiessen polygon network of the rain-collecting region required by research.
4. A neural network-based flood forecast method as claimed in claim 3, wherein the regional average rainfall
The:. Can be determined by the following formula>
Wherein n is the number of Thiessen polygons, and x i And A is the rainfall weight coefficient which is the ratio of each corresponding Thiessen polygon to the area of the area for the rainfall of each rainfall observation point.
5. The neural network-based flood forecast method of claim 1, wherein in step S3, flow synthesis is performed on each factor according to the propagation time according to the synthetic flow method, that is, column displacement transformation is performed on the event sample data corresponding to the propagation time, and corresponding regional average rainfall is calculated on the rainfall data in the transformed data units, and the calculation result is used as a new factor to participate in the neural network analysis, so that the data units to be subjected to the neural network analysis show the actual influence of the water level of the rainfall station to be forecasted, and the data of the data units to be subjected to the neural network analysis are added with the data of the continuous multiple earlier time periods correspondingly transformed according to the synthetic flow method.
6. The neural network-based flood forecasting method of claim 1, wherein the neural network analysis modeling is performed on at least one of the flood peak flow and the water level of the forecasting object by using the water rain condition and hydraulic engineering dispatching data of a plurality of flood orders which are transformed according to the flood forecasting synthetic flow method as a data analysis unit, the model obtained through the neural network analysis directly uses the hyper-warning flood event sample data which does not participate in the flood forecasting scheme to perform preliminary precision inspection, if the inspection result is accurate, the complete standard precision assessment and inspection are performed on the model, the modeling, the inspection and the error correction are repeated, the forecasting period can be properly adjusted according to the precision inspection result, and finally a set of relatively satisfactory flood forecasting scheme is obtained.
7. A neural network based flood forecasting system employing the method of any one of claims 1-6 for flood forecasting.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811375542 | 2018-11-19 | ||
| CN2018113755421 | 2018-11-19 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111199298A CN111199298A (en) | 2020-05-26 |
| CN111199298B true CN111199298B (en) | 2023-06-27 |
Family
ID=70746981
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811454308.8A Active CN111199298B (en) | 2018-11-19 | 2018-11-30 | Flood forecasting method and system based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111199298B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738600A (en) * | 2020-06-23 | 2020-10-02 | 南通大学 | Urban road air quality evaluation method based on high-precision PM2.5 inversion result |
| CN112417554B (en) * | 2020-11-17 | 2021-11-09 | 中国水利水电科学研究院 | Method for identifying local flood control influence risk factors by adopting fishbone map to carry out cross-basin water transfer engineering |
| CN112967476A (en) * | 2021-04-09 | 2021-06-15 | 黑龙江文图测绘地理信息有限责任公司 | Flood forecasting system based on hyperbolic flow distribution model |
| CN113642699A (en) * | 2021-06-30 | 2021-11-12 | 天津大学 | Intelligent river flood forecasting system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1438386A (en) * | 2002-01-31 | 2003-08-27 | 株式会社东芝 | River-level prediction device |
| CN102360468A (en) * | 2011-10-13 | 2012-02-22 | 贵州东方世纪科技有限责任公司 | Flood forecasting system |
| CN104615845A (en) * | 2014-12-19 | 2015-05-13 | 兰州大学 | Supercritical flood risk analysis method |
| CN105115573A (en) * | 2015-07-18 | 2015-12-02 | 厦门理工学院 | Correction method and device for flood flow forecasting |
| CN106326656A (en) * | 2016-08-24 | 2017-01-11 | 东南大学 | Simulation and prediction method for extreme rainstorm flood level of engineering facility |
| CN106650767A (en) * | 2016-09-20 | 2017-05-10 | 河海大学 | Flood forecasting method based on cluster analysis and real time correction |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140304007A1 (en) * | 2013-04-05 | 2014-10-09 | The Travelers Indemnity Company | Systems and methods for account-level flood risk assessment |
| GB2540331A (en) * | 2015-03-20 | 2017-01-18 | Hannah Smith Alexis | Flood prevention system and method |
| CN108446436A (en) * | 2018-02-08 | 2018-08-24 | 广州地理研究所 | The spatial distribution method for early warning of storm flood nonlinear model rainwater loss parameter |
-
2018
- 2018-11-30 CN CN201811454308.8A patent/CN111199298B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1438386A (en) * | 2002-01-31 | 2003-08-27 | 株式会社东芝 | River-level prediction device |
| JP2003293344A (en) * | 2002-01-31 | 2003-10-15 | Toshiba Corp | River water level predicting apparatus |
| CN102360468A (en) * | 2011-10-13 | 2012-02-22 | 贵州东方世纪科技有限责任公司 | Flood forecasting system |
| CN104615845A (en) * | 2014-12-19 | 2015-05-13 | 兰州大学 | Supercritical flood risk analysis method |
| CN105115573A (en) * | 2015-07-18 | 2015-12-02 | 厦门理工学院 | Correction method and device for flood flow forecasting |
| CN106326656A (en) * | 2016-08-24 | 2017-01-11 | 东南大学 | Simulation and prediction method for extreme rainstorm flood level of engineering facility |
| CN106650767A (en) * | 2016-09-20 | 2017-05-10 | 河海大学 | Flood forecasting method based on cluster analysis and real time correction |
Non-Patent Citations (2)
| Title |
|---|
| 基于人工神经网络的倒水河洪水预报模型;王一峰;王心亚;;水资源研究(第02期);42-50 * |
| 天然河道水位流量预测模型分析;高鹏;刘正东;;陕西水利(第S1期);221-230 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111199298A (en) | 2020-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111199298B (en) | Flood forecasting method and system based on neural network | |
| CN103729550B (en) | Multiple-model integration Flood Forecasting Method based on propagation time cluster analysis | |
| Wahl et al. | The stream-gaging program of the US Geological Survey | |
| Mehta et al. | Potential impacts on hydrology and hydropower production under climate warming of the Sierra Nevada | |
| Ahmed | Numerical modeling of the Rideau valley watershed | |
| Cui et al. | Improved Thomas–Fiering and wavelet neural network models for cumulative errors reduction in reservoir inflow forecast | |
| Fang et al. | Application of long short-term memory (LSTM) on the prediction of rainfall-runoff in karst area | |
| CN108269199B (en) | Object-oriented small reservoir group time-space distribution type outflow calculation method | |
| CN104636834A (en) | Improved optimization method for joint probability programming model system | |
| Zhao et al. | Risk assessment of flood disaster and forewarning model at different spatial-temporal scales | |
| Dong et al. | A hybrid hydrologic modelling framework with data-driven and conceptual reservoir operation schemes for reservoir impact assessment and predictions | |
| Quiroga et al. | Modelling high speed railroad geometry ageing as a discrete-continuous process | |
| An et al. | Identify and assess hydropower project’s multidimensional social impacts with rough set and projection pursuit model | |
| CN102880753B (en) | Based on the land utilization space characteristic dimension conversion method of fractal dimension | |
| Velasco et al. | Days-ahead water level forecasting using artificial neural networks for watersheds | |
| Yang et al. | Evaluation and Interpretation of Runoff Forecasting Models Based on Hybrid Deep Neural Networks | |
| CN116796799A (en) | Method for creating small-river basin flood rainfall threshold model in area without hydrologic data | |
| Ross et al. | A SPATIALLY RESPONSIVE HYDROLOGIC MODEL TO PREDICT EROSION AND SEDIMENT TRANSPORT 1 | |
| Chermahini et al. | Investigation of Climate Change Occurrence and its Impact on Wheat Growth (Case Study: Fars Province) | |
| CN117236199B (en) | Method and system for improving water quality and guaranteeing water safety of river and lake in urban water network area | |
| Hadipour et al. | Groundwater Level Forecasting by Application of Artificial Neural Network Approach: A Case Study in Qom Plain, Iran | |
| Fry et al. | The frequency of channel-forming discharges in a tributary of Upper Big Walnut Creek, Ohio | |
| Mandal | Uncertainty modeling in the assessment of climate change impacts on water resources management | |
| CN116341365A (en) | Flood forecasting method and system based on neural network | |
| Kokkinos et al. | Development and evaluation of an adaptive neuro fuzzy inference system for the calculation of soil water recharge in a watershed |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |