A hydrological forecasting method and system

Abstract

The present application discloses a hydrological forecasting method and system, which involves hydrological forecasting and artificial intelligence technology, including: obtaining remote sensing data of a target basin in advance, and establishing a rough terrain model of the target basin based on the acquired remote sensing data; collecting multiple image data of the set position of the target basin at a set time interval; constructing a global image containing the set position of the target basin based on the overlapping area; reconstructing at least part of the position of the dam contained in the global image; extracting features under a unified perspective, and inputting the extracted features into LSTM, and encoding the output of LSTM using Decoder; extracting hydrological data in the predicted image, and mapping the extracted hydrological data to the rough terrain model; performing hydrological forecasting based on the extracted hydrological data and the mapped rough terrain model. The present application can reduce the runoff lag in the upstream and downstream scales, and improve the real-time and accuracy of hydrological forecasting.

Description

Technical Field

[0001] The present application relates to the fields of hydrological forecasting and artificial intelligence technology, and in particular to a hydrological forecasting method and system. Background Art

[0002] The hydrological and meteorological data of large-scale basins are restricted by topography, regional economy and other conditions. The measured data are scarce, and the only data are difficult to support and construct the hydrological simulation of the basin, especially difficult to effectively forecast floods under future conditions. Effective runoff simulation and flood forecasting technology will provide important scientific and technological support for regional water resource utilization, ecosystem protection, agricultural development and urban planning.

[0003] Hydrological model is a classic method for runoff simulation and flood forecasting. The main methods currently include process-based physical mechanism models and data-driven models. Process-based physical mechanism models such as VIC (Variable Infiltration Capacity) and SWAT (Soil and Water Assessment Tool) are based on physical theories of water cycles and a large amount of measured data, and are highly practical. However, inductive reasoning is widely used in the creation of such models. Although it provides a reliable explanation for the hydrological process, the existing theory is not complete and the model itself is not comprehensive compared to the complex processes that actually occur. At the same time, there are a large number of complex parameters that need to be calibrated repeatedly, and it takes more time to build a regional model. Summary of the invention

[0004] The embodiment of the present application provides a hydrological forecasting method and system, which combines a coarse geographic model and a local fine model, and realizes hydrological prediction based on the local fine model through a long short-term memory network LSTM, thereby reducing the runoff lag at the upstream and downstream scales and improving the real-time and accuracy of the forecast.

[0005] The present application embodiment provides a hydrological forecasting method, including:

[0006] Pre-acquire remote sensing data of a target watershed, and establish a rough terrain model of the target watershed based on the acquired remote sensing data;

[0007] Collecting a plurality of image data of a set location of a target watershed at a set time interval, wherein the plurality of image data at least partially overlap and contain water area data;

[0008] Identify overlapping areas in the plurality of image data, and construct a global image including the set location of the target watershed based on the overlapping areas;

[0009] Reconstructing at least a portion of the location of the included dam based on the global image to obtain a local fine model;

[0010] Extract features from the local fine model at continuous moments using an Encoder under a unified perspective, input the extracted features into a long short-term memory network LSTM, and encode the output of the LSTM using a Decoder to obtain a predicted image under the unified perspective;

[0011] Extracting hydrological data from the predicted image, and mapping the extracted hydrological data to the rough terrain model;

[0012] Hydrological forecasting is performed based on the extracted hydrological data and the mapped coarse terrain model.

[0013] Optionally, identifying overlapping areas in the plurality of image data includes:

[0014] Segmenting multiple image data and calculating the similarity between image segmentations of two image data;

[0015] Segment the images with similarity greater than a preset similarity threshold and use the local area between the boundaries of the two image data as a possible overlapping area;

[0016] Edge detection is performed on possible overlapping areas of the two image data to determine the overlapping area according to the edge detection result.

[0017] Optionally, performing edge detection on possible overlapping areas of the two image data to determine the overlapping area according to the edge detection result includes:

[0018] According to the edge detection result, the two pieces of image data are superimposed based on the image segmentation with the largest calculated similarity;

[0019] Changing the transparency of a possible overlapping area of ​​one of the image data, and calculating the clarity of the possible overlapping areas of the two superimposed image data;

[0020] If the calculated clarity of the possible overlapping area is less than the preset clarity threshold, then an image segmentation is selected based on the image segmentation around the image segmentation with the maximum similarity;

[0021] Determine the pixel deviation between the same boundaries according to the edge detection result of the selected image segmentation;

[0022] The two superimposed image data are adjusted according to the direction and distance of the pixel deviation until the clarity of the calculated possible overlapping area meets the requirement.

[0023] Optionally, constructing a global image including the target watershed setting position according to the overlapping area includes:

[0024] A plurality of reference points are set according to the determined overlapping area, so as to construct a global image including the set position of the target watershed based on the reference points.

[0025] Optionally, extracting hydrological data from the predicted image and mapping the extracted hydrological data to the rough terrain model includes:

[0026] Identifying a water area boundary at a set location in the predicted image;

[0027] Extracting a representative boundary of a set pixel width based on the water area boundary, and marking the water area boundary on the representative boundary;

[0028] Under the unified perspective, searching the coarse terrain model using the representative boundary to select the position with the highest matching degree as the mapping boundary;

[0029] According to the position of the water area boundary in the representative boundary, the water area boundary is mapped to the mapping boundary, so as to map the water level data to the corresponding set position in the rough terrain model.

[0030] Optionally, extracting hydrological data from the predicted image and mapping the extracted hydrological data to the rough terrain model further includes:

[0031] Based on the water level data mapped at the set position in the rough terrain model, the water level data is filled according to the hydraulic gradient relationship of the target basin, so as to refresh part of the runoff data of the target basin based on the mapped water level data in the rough terrain model.

[0032] Optionally, performing hydrological forecasting based on the extracted hydrological data and the mapped coarse terrain model includes:

[0033] Acquiring rainfall data of the target watershed;

[0034] Based on the rough terrain model, determine the rainfall impact area, and estimate the runoff data of the remaining part according to the rainfall impact area and the rainfall data;

[0035] According to the basic runoff information, the estimated runoff data is filled into the corresponding runoff of the rough terrain model along the water flow direction;

[0036] The estimated and filled runoff data and the partial runoff data updated based on the mapped water level data are spliced ​​in the coarse terrain model to obtain the combined runoff.

[0037] Optionally, performing hydrological forecasting based on the extracted hydrological data and the mapped coarse terrain model further includes:

[0038] Performing hydrological forecasting based on the water level data, rainfall data, and combined runoff; and,

[0039] Identifying areas of water level change in the coarse terrain model during any refresh process;

[0040] The water level change area is highlighted in the terrain model.

[0041] An embodiment of the present application proposes a hydrological forecasting system, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the hydrological forecasting method as described above are implemented.

[0042] The hydrological forecasting method of the present application combines a coarse geographic model with a local fine model, and realizes hydrological prediction based on the local fine model through a long short-term memory network LSTM, thereby reducing the runoff lag at the upstream and downstream scales and improving the real-time and accuracy of the forecast.

[0043] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0045] Figure 1 The following is a schematic diagram of the basic process of the hydrological forecasting method of this embodiment. DETAILED DESCRIPTION

[0046] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0047] The present application embodiment proposes a hydrological forecasting method, such as Figure 1 As shown, the following steps are included:

[0048] In step S101, remote sensing data of the target watershed is acquired in advance, and a rough terrain model of the target watershed is established based on the acquired remote sensing data. In some examples, the accuracy of the rough terrain model can be determined according to actual needs. For example, a simplified model can be constructed for areas of the target watershed that cannot be reached during the flood season. In a specific example, the terrain model can be assigned a basic watershed water level in the initial situation, and the rough terrain model containing the basic water level is applied to the subsequent water level refresh process.

[0049] In step S102, multiple image data of a set position of a target basin are collected at a set time interval, wherein the multiple image data have at least a partially overlapping area and contain water area data. For example, the set time interval can be set according to the actual flood situation of the target basin, and the user can set a shorter interval if he wants to pay more attention to the development of the flood situation. At the same time, multiple image data of a set position are collected by, for example, monitoring equipment, for example, the multiple image data can be collected by rotating the monitoring equipment at a certain angle, and the images collected at adjacent angles have a certain area overlap.

[0050] In step S103, the overlapping areas in the multiple image data are identified, and a global image including the set position of the target watershed is constructed based on the overlapping areas. The overlapping areas are identified based on the images collected at the same time, so that after the overlapping parts are identified, a global image of the set position at that time is constructed. In some examples, the set position can be, for example, a dam area, or a monitoring station with monitoring set, etc.

[0051] In step S104, at least part of the location of the included dam is reconstructed based on the global image to obtain a local fine model. In some embodiments, for example, the fine model can be obtained by reconstructing based on the global image using Reality Capture. Specifically, the key points included in the overlapped area identified above can be input into the RealityCapture software to improve the reconstruction effect and adapt to the fine model reconstruction under various environmental lighting conditions.

[0052] In step S105, the local fine model of continuous moments is extracted by using Encoder under a unified perspective, and the extracted features are input into the long short-term memory network LSTM, and the output of LSTM is encoded by Decoder to obtain the predicted image under the unified perspective. The embodiment of the present application is based on the local fine model of continuous moments, and uses the memory capacity of LSTM to predict the hydrological data (such as water level data) at the later moment in the fine model, thereby realizing the prediction of the local water conditions at the set location.

[0053] In step S106, the hydrological data in the predicted image is extracted and mapped to the rough terrain model. In the embodiment of the present application, the predicted water data is further mapped to the rough terrain model, so that the water data of the target basin is fully covered and analyzed, and the hydrological prediction of the target basin is realized.

[0054] In step S107, hydrological forecasting is performed based on the extracted hydrological data and the mapped rough terrain model.

[0055] The hydrological forecasting method of the present application combines a coarse geographic model with a local fine model, and realizes hydrological prediction based on the local fine model through a long short-term memory network LSTM, thereby reducing the runoff lag at the upstream and downstream scales and improving the real-time and accuracy of the forecast.

[0056] In some embodiments, identifying overlapping areas in the plurality of image data includes:

[0057] The multiple image data are segmented, and the similarity between the image segmentations of two image data is calculated. The specific image data with overlapping areas can be determined according to the change rule of the calculated similarity, and the segmentation similarity of the overlapping area is greater than that of the non-overlapping area.

[0058] The image segmentation with a similarity greater than a preset similarity threshold and the local area between the boundaries of the two image data are regarded as possible overlapping areas. For example, if the preset similarity threshold is set to 80%, the image areas in the overlapping area are gradually greater than 80% in similarity from the outside to the inside.

[0059] Edge detection is performed on possible overlapping areas of the two image data to determine the overlapping area according to the edge detection result.

[0060] In some embodiments, performing edge detection on possible overlapping areas of the two image data to determine the overlapping area according to the edge detection result includes:

[0061] According to the edge detection result, the two image data are superimposed based on the image segmentation with the greatest calculated similarity. In some specific examples, according to the possible overlapping areas identified above, the image area with the greatest similarity is selected for superposition.

[0062] The transparency of a possible overlapping area of ​​one of the image data is changed, and the clarity of the possible overlapping areas of the two superimposed image data is calculated.

[0063] If the calculated clarity of the possible overlapping area is less than a preset clarity threshold, an image segmentation is selected around the image segmentation with the maximum similarity.

[0064] The pixel deviation between the same boundaries is determined based on the edge detection results of the selected image segmentation. In the specific example, the image area with the greatest similarity is superimposed based on the edge detection results and then fine-tuned. The pixel deviation (offset) of the image segmentation selected around the segment with the greatest similarity is calculated to determine the direction of fine-tuning and the amount of pixels for fine-tuning.

[0065] The two superimposed image data are adjusted according to the direction and distance of the pixel deviation until the clarity of the calculated possible overlapping area meets the requirement.

[0066] In some embodiments, constructing a global image including the target watershed setting location according to the overlapping area includes:

[0067] A plurality of reference points are set according to the determined overlapping area to construct a global image including the target watershed setting position based on the reference points. In a specific example, a plurality of reference points can be set according to the edge detection result of the overlapping area to construct a global image of the target watershed setting position at that moment.

[0068] In some embodiments, extracting the hydrological data in the predicted image and mapping the extracted hydrological data to the rough terrain model comprises:

[0069] Identify the water boundary of the set position in the predicted image. In some examples, the water boundary can also be determined by edge detection. Since the water boundary is covered by, for example, the river bank area and the dam area, there is a situation where the accuracy of boundary recognition is not high.

[0070] In the embodiment of the present application, a representative boundary with a set pixel width is further extracted based on the water boundary, and the water boundary is marked at the representative boundary. By setting a certain pixel width based on the water boundary, a representative boundary including the water boundary is extracted and marked.

[0071] Under the unified perspective, the representative boundary is used to search in the coarse terrain model to select the position with the highest matching degree as the mapping boundary. In the specific example, if the water boundary is directly used for searching, due to the influence of the river bank area and the dam area itself, and the coarse model itself is not fine enough, it is very likely to cause the problem of not being able to match the accurate position. The embodiment of the present application can greatly improve the matching accuracy of the fine model to the coarse model by setting a representative boundary of pixel width.

[0072] According to the position of the water area boundary in the representative boundary, the water area boundary is mapped to the mapping boundary to map the water level data to the corresponding set position in the coarse terrain model. For example, if the set position is a dam area, the water area boundary is mapped to the mapping boundary of the coarse terrain model based on the positional relationship between the water area boundary identifier and the pixel width.

[0073] In some embodiments, extracting the hydrological data in the predicted image and mapping the extracted hydrological data to the coarse terrain model further includes: filling the water level data based on the water level data mapped at the set position in the coarse terrain model according to the hydraulic gradient relationship of the target basin, so as to refresh part of the runoff data of the target basin in the coarse terrain model based on the mapped water level data. The solution of the present application can reflect the overall hydrological situation of the target basin through the coarse model by associating and mapping the refined local model with the coarse terrain model of the target basin, thereby achieving the effect of confirming the whole by the point.

[0074] In some embodiments, performing hydrological forecasting based on the extracted hydrological data and the mapped coarse terrain model includes:

[0075] Acquiring rainfall data of the target watershed;

[0076] Based on the rough terrain model, a rainfall impact area is determined, and runoff data of the remaining part is estimated according to the rainfall impact area and the rainfall data. In some examples, the flow of introduced runoff can be estimated according to empirical data or a relationship between fitting rainfall and rainfall impact area.

[0077] According to the basic runoff information, the estimated runoff data is filled into the corresponding runoff of the rough terrain model along the water flow direction.

[0078] The estimated filled runoff data and the partial runoff data updated based on the mapped water level data are spliced ​​in the coarse terrain model to obtain the combined runoff. Based on the above example, the coarse terrain model is filled according to the estimated runoff data and the basic runoff information, so that the partial runoff not covered by the water level data filling according to the hydraulic gradient relationship of the target watershed in the above example is filled and connected, thereby forming the combined runoff of the target watershed.

[0079] In some embodiments, performing hydrological forecasting based on the extracted hydrological data and the mapped coarse terrain model further comprises:

[0080] Performing hydrological forecasting based on the water level data, rainfall data, and combined runoff; and,

[0081] Identifying areas of water level change in the coarse terrain model during any refresh process;

[0082] The water level change area is highlighted in the terrain model. In the embodiment of the present application, the water level change area is highlighted in the terrain model so that users can pay more attention to the change area. In particular, for the case of rising water levels, the predicted flood inundation area can be determined after refreshing, thereby reducing the runoff lag in the upstream and downstream, improving the real-time and accuracy of the forecast, and improving the auxiliary effect on flood control scheduling.

[0083] An embodiment of the present application proposes a hydrological forecasting system, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the hydrological forecasting method as described above are implemented.

[0084] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present disclosure with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. It is not limited to the examples described in this specification or during the implementation of this application, and its examples will be interpreted as non-exclusive.

[0085] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art may use other embodiments when reading the above description.

[0086] The above embodiments are merely exemplary embodiments of the present disclosure. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present disclosure, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present invention.

Claims

1. A hydrological forecasting method, characterized in that: include: Pre-acquire remote sensing data of a target watershed, and establish a rough terrain model of the target watershed based on the acquired remote sensing data; Collecting multiple image data of a set location in a target watershed at a set time interval, wherein the multiple image data have at least a partial overlapping area and contain water area data; Identify overlapping areas in the plurality of image data, and construct a global image including the set location of the target watershed based on the overlapping areas; Reconstructing at least a portion of the location of the included dam based on the global image to obtain a local fine model; Extract features from the local fine model at continuous moments using an Encoder under a unified perspective, input the extracted features into a long short-term memory network LSTM, and encode the output of the LSTM using a Decoder to obtain a predicted image under the unified perspective; Extracting hydrological data from the predicted image, and mapping the extracted hydrological data to the rough terrain model; Hydrological forecasting is performed based on the extracted hydrological data and the mapped coarse terrain model.

2. The hydrological forecasting method according to claim 1, characterized in that: Identifying overlapping areas in multiple image data includes: Segmenting multiple image data and calculating the similarity between image segmentations of two image data; Segment the images according to the similarity greater than a preset similarity threshold and use the local area between the boundaries of the two image data as the possible overlapping area; Edge detection is performed on possible overlapping areas of the two image data to determine the overlapping area according to the edge detection result.

3. The hydrological forecasting method according to claim 2, characterized in that: Performing edge detection on possible overlapping areas of the two image data to determine the overlapping areas according to the edge detection result includes: According to the edge detection result, superimposing the two pieces of image data based on the image segmentation with the largest calculated similarity; Changing the transparency of a possible overlapping area of ​​one of the image data, and calculating the clarity of the possible overlapping areas of the two superimposed image data; If the calculated clarity of the possible overlapping area is less than the preset clarity threshold, then an image segmentation is selected based on the image segmentation around the image segmentation with the maximum similarity; Determine the pixel deviation between the same boundaries according to the edge detection result of the selected image segmentation; The two superimposed image data are adjusted according to the direction and distance of the pixel deviation until the clarity of the calculated possible overlapping area meets the requirement.

4. The hydrological forecasting method according to claim 3, characterized in that: According to the overlapping area, constructing a global image including the target watershed setting position includes: A plurality of reference points are set according to the determined overlapping area, so as to construct a global image including the set position of the target watershed based on the reference points.

5. The hydrological forecasting method according to claim 3, characterized in that: Extracting hydrological data from the predicted image and mapping the extracted hydrological data to the rough terrain model includes: Identifying a water area boundary at a set location in the predicted image; Extracting a representative boundary of a set pixel width based on the water area boundary, and marking the water area boundary on the representative boundary; Under the unified perspective, searching the coarse terrain model using the representative boundary to select the position with the highest matching degree as the mapping boundary; According to the position of the water area boundary in the representative boundary, the water area boundary is mapped to the mapping boundary, so as to map the water level data to the corresponding set position in the rough terrain model.

6. The hydrological forecasting method according to claim 5, characterized in that: Extracting hydrological data from the predicted image and mapping the extracted hydrological data to the rough terrain model further includes: Based on the water level data mapped at the set position in the rough terrain model, the water level data is filled according to the hydraulic gradient relationship of the target basin, so as to refresh part of the runoff data of the target basin based on the mapped water level data in the rough terrain model.

7. The hydrological forecasting method according to claim 6, characterized in that: Based on the extracted hydrological data and the mapped coarse terrain model, the hydrological forecast is performed including: Acquiring rainfall data of the target watershed; Based on the rough terrain model, determine the rainfall impact area, and estimate the runoff data of the remaining part according to the rainfall impact area and the rainfall data; According to the basic runoff information, the estimated runoff data is filled into the corresponding runoff of the rough terrain model along the water flow direction; The estimated and filled runoff data and the partial runoff data updated based on the mapped water level data are spliced ​​in the coarse terrain model to obtain the combined runoff.

8. The hydrological forecasting method according to claim 6, characterized in that: Based on the extracted hydrological data and the mapped coarse terrain model, the hydrological forecasting also includes: Performing hydrological forecasting based on the water level data, rainfall data, and combined runoff; and, Identifying areas of water level change in the coarse terrain model during any refresh process; The water level change area is highlighted in the terrain model.

9. A hydrological forecasting system, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the hydrological forecasting method according to any one of claims 1 to 8 are implemented.

Images