Reservoir landform data acquisition method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of hydraulic engineering and geographic information, and discloses a reservoir landform data acquisition method and system based on multi-source data fusion, and the method comprises the steps: obtaining multi-source remote sensing data and sensor data, carrying out the preprocessing and time sequence alignment adjustment, and obtaining a corrected time sequence consistent feature sequence; performing fusion processing according to the feature sequence to obtain optimized landform model input; performing landform elevation calculation according to the optimized landform model input, and determining a high-precision reservoir landform DEM; and carrying out geomorphic space analysis according to the DEM to obtain a high-precision reservoir geomorphic acquisition result. The method can solve the technical problems of low landform data acquisition precision and model distortion.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of water conservancy and geographic information technology, and in particular to a reservoir geomorphology data acquisition method and system based on multi-source data fusion. BACKGROUND

[0002] At present, reservoir geomorphology data acquisition is a key link in the field of water conservancy and geographic information, and its accuracy is directly related to the decision-making effect of water resources regulation, flood control and disaster reduction, and ecological protection.

[0003] In one prior art, the geomorphology acquisition method often relies on isolated or static data processing strategies. Such strategies are difficult to handle the format heterogeneity and noise interference of multi-source data when facing the dynamic changes of the reservoir environment; at the same time, they ignore the matching deviation between historical water level and real-time geomorphology, leading to fragmentation of the feature extraction process, and finally causing distortion of the geomorphology model in different water level stages.

[0004] Therefore, there is a technical problem of low accuracy and model distortion in the prior art. SUMMARY

[0005] The present application provides a reservoir geomorphology data acquisition method and system based on multi-source data fusion to solve the technical problem of low accuracy and model distortion in the prior art.

[0006] In a first aspect, to solve the above technical problem, the present application provides a reservoir geomorphology data acquisition method based on multi-source data fusion, comprising:

[0007] Obtain multi-source remote sensing data and sensor data and preprocess to obtain a denoised remote sensing sensor unified feature set;

[0008] According to the denoised remote sensing sensor unified feature set, perform historical water level feature correlation analysis to obtain a matching deviation index;

[0009] Perform time sequence alignment adjustment for the matching deviation index to determine a corrected time sequence consistent feature sequence;

[0010] According to the corrected time sequence consistent feature sequence, perform remote sensing resolution fusion processing to obtain an optimized geomorphology model input;

[0011] According to the optimized geomorphology model input, perform geomorphology elevation calculation to determine a high-precision reservoir geomorphology DEM;

[0012] According to the high-precision reservoir geomorphology DEM, perform geomorphology spatial analysis and classification to obtain a high-precision reservoir geomorphology acquisition result.

[0013] Preferably, the multi-source remote sensing data and sensor data are acquired and preprocessed to obtain a denoised remote sensing sensor unified feature set, including:

[0014] The multi-source remote sensing data and sensor data are acquired and noise filtering and format heterogeneous conversion are performed to obtain multi-source fusion initial data.

[0015] According to the multi-source fusion initial data, remote sensing image band selection and sensor reading timestamp are extracted, and timestamp alignment and clustering grouping denoising processing are performed to determine the denoised remote sensing sensor unified feature set.

[0016] Preferably, the denoised remote sensing sensor unified feature set is used for historical water level feature correlation analysis to obtain a matching deviation index, including:

[0017] A dynamic feature description sequence is obtained from the denoised remote sensing sensor unified feature set.

[0018] According to the dynamic feature description sequence, feature correlation analysis is performed to obtain a first matching preparation data set.

[0019] For the first matching preparation data set, historical water level sequence comparison is performed in combination with the dynamic feature description sequence to determine the matching deviation index.

[0020] Preferably, the matching deviation index is subjected to timing alignment adjustment to determine a corrected timing consistent feature sequence, including:

[0021] For the matching deviation index, bias cumulative value timing alignment adjustment is performed to obtain a bias adjustment feature set.

[0022] According to the bias adjustment feature set, a sliding window translation correction is used to obtain a timing consistent sequence.

[0023] The timing consistent sequence is subjected to timing continuity statistical analysis to determine the corrected timing consistent feature sequence.

[0024] Preferably, the corrected timing consistent feature sequence is subjected to remote sensing resolution fusion processing to obtain an optimized topographic model input, including:

[0025] For the corrected timing consistent feature sequence, remote sensing image resolution multi-dimensional description integration is performed to obtain a fusion resolution data set.

[0026] According to the fusion resolution data set, a resolution consistent description set is obtained through spatial grid alignment.

[0027] The resolution consistent description set is subjected to spatial smoothness analysis to determine the optimized topographic model input.

[0028] Preferably, the step of calculating the geomorphic elevation based on the optimized geomorphic model input to determine the high-precision reservoir geomorphic DEM includes:

[0029] A coherent landform description is extracted from the input of the optimized landform model and fused with the deviation impact of the water level change scenario simulation to obtain the deviation impact dataset.

[0030] Based on the deviation impact dataset, deviation correction processing is performed to obtain a deviation optimization description set;

[0031] Based on the deviation optimization description set, the geomorphic elevation is calculated to obtain the high-precision reservoir geomorphic DEM.

[0032] Preferably, the step of performing geomorphological spatial analysis and classification based on the high-precision reservoir geomorphological DEM to obtain high-precision reservoir geomorphological acquisition results includes:

[0033] The slope gradient is calculated based on the high-precision reservoir topography DEM to obtain the slope distribution matrix;

[0034] The slope distribution matrix is ​​used to simulate the water flow path, and path interpolation is performed to determine the continuous path chain.

[0035] Based on the continuous path chain and the slope distribution matrix, the watershed algorithm is used to divide the region into blocks;

[0036] The aforementioned regional blocks are classified to obtain landform category groups;

[0037] The high-precision reservoir landform acquisition results are obtained by integrating the collected data according to the landform category group.

[0038] Secondly, the present invention provides a reservoir geomorphological data acquisition system based on multi-source data fusion, comprising:

[0039] The data preprocessing module is used to acquire multi-source remote sensing data and sensor data, and perform preprocessing to obtain a unified feature set of remote sensing sensors after denoising.

[0040] The correlation analysis module is used to perform historical water level feature correlation analysis based on the unified feature set of the denoised remote sensing sensor to obtain the matching deviation index.

[0041] The timing adjustment module is used to perform timing alignment adjustment on the matching deviation index and determine the corrected timing consistency feature sequence.

[0042] The fusion processing module is used to perform remote sensing resolution fusion processing based on the corrected temporally consistent feature sequence to obtain an optimized landform model input.

[0043] The geomorphological modeling module is used to calculate the geomorphological elevation based on the input of the optimized geomorphological model and determine the high-precision reservoir geomorphological DEM.

[0044] The spatial analysis module is used to perform geomorphological spatial analysis and classification based on the high-precision reservoir geomorphology DEM to obtain high-precision reservoir geomorphology acquisition results.

[0045] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the reservoir geomorphological data acquisition method based on multi-source data fusion as described above.

[0046] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the reservoir geomorphological data acquisition method based on multi-source data fusion described above.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] (1) This invention acquires multi-source data such as remote sensing images and sensor readings, and performs noise filtering, format unification and timestamp alignment to form a unified feature set; then extracts dynamic features and performs correlation analysis with historical water level sequences to quantify matching deviation indicators; this method solves the problem of fragmented feature extraction and ignoring historical matching deviation caused by heterogeneous data formats and noise interference in the prior art, and provides a reliable data foundation for subsequent high-precision modeling.

[0049] (2) This invention corrects the cumulative error by performing temporal alignment on the matching deviation index and integrates the resolution multidimensional description of remote sensing images to generate a temporally and spatially consistent optimized geomorphic model input. This method solves the problem of geomorphic model distortion at different water level stages caused by inconsistent data temporal sequence and inability to accurately correspond with historical water levels in the prior art, and improves the coherence and reliability of geomorphic description.

[0050] (3) This invention utilizes a high-precision reservoir landform DEM to further extract landform boundaries, calculate slopes, and simulate water flow paths, thereby achieving a complete closed loop from data fusion and deviation correction to specific landform spatial analysis. This method solves the problem of low accuracy of the final acquisition results caused by model distortion and lack of coherent description of landform dynamic features in the prior art, and finally obtains high-precision reservoir landform acquisition results. Attached Figure Description

[0051] Figure 1This is a schematic diagram of the process for a reservoir geomorphological data acquisition method based on multi-source data fusion provided in the first embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of the reservoir geomorphological data acquisition system based on multi-source data fusion provided in the second embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Reference Figure 1 The first embodiment of the present invention provides a method for acquiring reservoir geomorphological data based on multi-source data fusion, including the following steps:

[0055] S11: Acquire multi-source remote sensing data and sensor data, and perform preprocessing to obtain a denoised unified feature set of remote sensing sensors;

[0056] S12, based on the unified feature set of the denoised remote sensing sensor, perform a correlation analysis of historical water level features to obtain a matching deviation index;

[0057] S13, perform time-series alignment adjustment on the matching deviation index to determine the corrected time-series consistent feature sequence;

[0058] S14. Based on the corrected temporally consistent feature sequence, perform remote sensing resolution fusion processing to obtain the optimized landform model input.

[0059] S15, Based on the input of the optimized geomorphological model, perform geomorphological elevation calculation to determine the high-precision reservoir geomorphological DEM;

[0060] S16. Based on the high-precision reservoir landform DEM, perform landform spatial analysis and classification to obtain high-precision reservoir landform acquisition results.

[0061] In step S11, multi-source remote sensing data and sensor data are acquired and preprocessed to obtain a denoised unified feature set of remote sensing sensors, including:

[0062] The multi-source remote sensing data and sensor data are acquired, and noise filtering and format heterogeneous conversion are performed to obtain the initial multi-source fusion data;

[0063] Based on the initial data from the multi-source fusion, the remote sensing image band selection and sensor reading timestamps are extracted, and timestamp alignment and clustering grouping denoising are performed to determine the unified feature set of the denoised remote sensing sensor.

[0064] In one implementation, this embodiment acquires the multi-source remote sensing data (e.g., satellite images containing remote sensing image resolution values) and sensor data (e.g., ground monitoring station data containing sensor readings). This embodiment filters the sensor readings for noise based on a preset measurement threshold.

[0065] It should be noted that the preset measurement threshold is determined through statistical analysis of historical data. In this embodiment, sensor readings (e.g., water level data) are collected for at least 24 hours under stable operating conditions. The mean and standard deviation of this dataset are calculated, and based on the 3-sigma principle, the value corresponding to "mean + 3 times the standard deviation" is determined as the preset measurement threshold. Any sampling point whose instantaneous reading exceeds the preset measurement threshold is identified as a noise point and filtered out.

[0066] For example, suppose the preset measurement threshold is determined to be 1000 mm through historical data analysis. In the acquired sensor data, a continuous water level sequence is [800, 850, 1200, 750] (unit: mm). The reading of 1200 mm is identified as a noise point and filtered out, and the sequence becomes [800, 850, 750].

[0067] It is worth noting that after noise filtering, this embodiment performs heterogeneous format conversion on the multi-source remote sensing data and sensor data. This embodiment unifies all data in different formats (e.g., TIFF format remote sensing images, GeoJSON format vector data, CSV format sensor time-series data) into a standardized, self-describing geospatial data format (e.g., NetCDF format) and unifies their coordinate reference system (e.g., WGS 84) to obtain the initial data for multi-source fusion.

[0068] In another implementation, this embodiment extracts remote sensing image band selection and sensor reading timestamps based on the initial data from the multi-source fusion, and performs timestamp alignment and clustering grouping noise reduction processing.

[0069] It should be noted that the specific implementation of the timestamp alignment is as follows: In this embodiment, the timestamp sequences of each data source are extracted from the initial multi-source fusion data. This embodiment employs time series resampling processing to unify all data to a fixed time interval. This time interval is determined based on the data source with the slowest update frequency in the initial multi-source fusion data (e.g., minute-level updates of remote sensing images). For example, if the update interval of the slowest data source (remote sensing image) is 1 minute, then the unified, fixed time interval is set to 1 minute.

[0070] It should be noted that for data with an update frequency faster than the unified time interval (e.g., sensor readings at the second level), this embodiment performs aggregation processing within that time interval (i.e., within 1 minute) (e.g., calculates the average or median of all second-level readings within the window), and uses the aggregation result as the representative value of that 1-minute time point to ensure the synchronization of all data in the time dimension.

[0071] It is worth noting that the specific implementation of the clustering and grouping denoising process in this embodiment is as follows: For data features aligned with timestamps (e.g., specific remote sensing image band selection values, sensor readings), the k-means clustering algorithm is used. This algorithm divides data points within a time window into k clusters. The value of k can be determined through offline analysis, using the Elbow Method to cluster historical datasets. This historical feature dataset is constructed by collecting and storing representative historical timestamp-aligned data features (including remote sensing image band selection values ​​and sensor readings). In this embodiment, the number of clusters corresponding to the inflection point where the total variance (Total Sum of Squares) gradually flattens out is analyzed using the Elbow Method and selected as the value of k.

[0072] This embodiment identifies clusters that meet either of the following conditions as noise clusters: first, the total number of data points within the cluster is less than a preset cluster size threshold; second, the Euclidean distance between the cluster center (centroid) and the total centroid of all data points within the time window exceeds a preset outlier threshold. The preset cluster size threshold can be determined by clustering historical normal datasets, analyzing their cluster size distribution, and selecting the value corresponding to the 5th percentile of that distribution. This method is a robust statistical method for nonparametric distributions and can effectively define the boundaries of statistically significant "small clusters." The preset outlier threshold can be determined by clustering historical normal datasets, analyzing their cluster center distance distribution, and using the 3-sigma principle to determine the value corresponding to "mean distance + 3 times the standard deviation." In this embodiment, all data points within these identified noise clusters are marked as noise and removed. The final output of this processing is the denoised unified feature set of the remote sensing sensor.

[0073] In step S12, based on the unified feature set of the denoised remote sensing sensors, a correlation analysis of historical water level features is performed to obtain a matching deviation index, including:

[0074] A dynamic feature description sequence is obtained from the denoised remote sensing sensor unified feature set;

[0075] Based on the dynamic feature description sequence, feature correlation analysis is performed to obtain the first matching preparation dataset;

[0076] Based on the first matching preparation dataset and in conjunction with the dynamic feature description sequence, historical water level sequences are compared to determine the matching deviation index.

[0077] In one implementation, this embodiment extracts specific feature columns (e.g., reflectivity of a specific band, reading values ​​of a specific sensor) related to reservoir landform changes (such as water level and inundation area) from the unified feature set of the denoised remote sensing sensors obtained in S11, and sorts them according to the timestamps aligned in S11 to form a feature vector that can reflect the changes in reservoir landform over time. This feature vector is the dynamic feature description sequence.

[0078] It should be noted that the specific implementation of the feature correlation analysis in this embodiment is to use a pre-trained Convolutional Neural Network (CNN) model. The structure of the CNN model may include a one-dimensional convolutional layer (e.g., containing 32 convolutional kernels of size 3) for receiving the dynamic feature description sequence, a ReLU activation layer, a one-dimensional max pooling layer (e.g., pooling size of 2), a flattening layer, and a fully connected layer for outputting correlation weights.

[0079] It should be noted that the CNN model was determined through offline training. The training process, in this embodiment, uses a historical dataset containing historical dynamic feature description sequences and their corresponding real historical water level sequences as labels. This embodiment uses Mean Squared Error (MSE) as the loss function and iteratively trains the CNN model using the Adam optimizer until the model's loss value converges on the independent validation set. The hyperparameters in the model (e.g., the number of convolutional kernels is 32, and the kernel size is 3) are determined by using a grid search strategy during the offline training phase to optimize within a preset parameter space (e.g., the number of convolutional kernels ranges from [16, 32, 64]) and select the parameter combination that minimizes the loss value on the validation set.

[0080] In another implementation, this embodiment inputs the dynamic feature description sequence into the trained convolutional neural network model. The model captures local temporal patterns in the dynamic feature description sequence through its one-dimensional convolutional layer and analyzes the nonlinear dependence between these patterns and water level change indicators. Finally, it outputs a correlation weight matrix that quantifies the strength of this dependence. This correlation weight matrix is ​​the first matching preparation dataset.

[0081] It is worth noting that this embodiment compares the first matching preparation dataset with the dynamic feature description sequence and a preset historical water level sequence. This embodiment compares the association weight values ​​in the first matching preparation dataset (i.e., the association weight matrix) with a preset weight threshold. If, at a certain point in time, the association weight value exceeds the preset weight threshold, this embodiment determines that the current dynamic feature deviates significantly from historical patterns. At that same point in time, this embodiment extracts the current feature value (e.g., the current average water level of 0.80 meters) from the dynamic feature description sequence and subtracts it from the baseline value in the historical water level sequence (e.g., the historical average for the same period of 0.75 meters). The calculated difference is determined as the matching deviation index.

[0082] It should be noted that the process of constructing the preset historical water level sequence is as follows: In this embodiment, historical water level measurement data of the reservoir (e.g., the past five years) are collected and aggregated according to the time dimension (e.g., by calendar day or by week), and the statistical benchmark value (e.g., the historical average value for the same period) for each time unit is calculated. The set of benchmark values ​​constitutes the preset historical water level sequence.

[0083] It should be noted that the preset weight threshold is determined through statistical analysis of historical data. The determination process involves collecting a large number of related weight values ​​under normal historical operating conditions (e.g., non-flood season, non-extreme dry season) to construct a benchmark dataset. This embodiment calculates the probability density distribution of this benchmark dataset and selects the value corresponding to the 95th percentile of this distribution as the preset weight threshold. Selecting the 95th percentile is a robust statistical method commonly used to define the statistical upper limit of normal operating conditions in non-normally distributed data and to identify high-confidence anomalies.

[0084] For example, suppose the preset weight threshold is determined to be 0.85 through historical data analysis. A certain dynamic feature obtained in S11 has an association weight value (i.e., the first matching preparation dataset) of 0.92 calculated by CNN. Since 0.92 exceeds 0.85, this embodiment determines that a deviation has occurred. At this time, this embodiment obtains the dynamic feature value (current water level 800 mm) and the historical water level sequence value (750 mm) corresponding to that time point, and records the difference (50 mm) as the matching deviation index.

[0085] In step S13, time-series alignment adjustment is performed on the matching deviation index to determine the corrected time-series consistent feature sequence, including:

[0086] For the aforementioned matching deviation index, a time-series alignment adjustment of the cumulative deviation value is performed to obtain a deviation adjustment feature set;

[0087] Based on the aforementioned deviation adjustment feature set, a sliding window translation correction is used to obtain a time-consistent sequence;

[0088] Perform time-series coherence statistical analysis on the time-consistent sequence to determine the corrected time-consistent feature sequence.

[0089] In one implementation, this embodiment performs time-series alignment adjustment on the cumulative deviation value of the matching deviation index obtained in S12. Specifically, this adjustment involves first calculating the cumulative sum of the matching deviation index within a time window to obtain a cumulative deviation value. This embodiment compares the cumulative deviation value with a preset cumulative threshold. If the cumulative deviation value exceeds the preset cumulative threshold, this embodiment applies Min-Max normalization to the deviation index sequence within the time window. This is achieved by multiplying the difference between the original value and the minimum value of the sequence by the preset cumulative threshold, and then dividing the product by the difference between the maximum and minimum values ​​of the sequence. This linearly maps the sequence values ​​to a new interval [0, the preset cumulative threshold], thereby obtaining the deviation adjustment feature set.

[0090] It should be noted that the preset cumulative threshold is determined through statistical analysis of historical data. The determination process involves collecting the matching deviation index under historical normal operating conditions and calculating its cumulative deviation value. This embodiment constructs a statistical distribution for all historical cumulative deviation values ​​and selects the value corresponding to the 95th percentile of this distribution as the preset cumulative threshold. This percentile is a robust statistical method widely used in system performance engineering. Its basis lies in the fact that this value can effectively cover the vast majority (95%) of normal operating fluctuations while excluding the interference of the top 5% of extreme outliers, thereby achieving a balance between alarm sensitivity and system stability.

[0091] In another implementation, this embodiment uses a sliding window translation correction based on the deviation adjustment feature set. Specifically, this correction employs a sliding average filter with a preset window size (e.g., 3 time units) to iterate through the deviation adjustment feature set point by point. The arithmetic mean of all data points within the window is replaced with the new value at the center point of the window, thereby smoothing local noise spikes in the sequence and obtaining the time-consistent sequence.

[0092] It is worth noting that this embodiment performs a time-series coherence statistical analysis on the time-consistent sequence. Specifically, this analysis involves calculating the first-order difference (i.e., the rate of change between data points at adjacent time points) of the time-consistent sequence and obtaining the variance of this first-order difference sequence. This variance value quantifies the degree of local jitter in the sequence.

[0093] It should be noted that in this embodiment, the calculated variance value is compared with the preset coherence threshold. The preset coherence threshold is determined through offline analysis. The determination process is as follows: In this embodiment, a large number of known historical geomorphic sequences labeled as "highly coherent" are collected, and their first-order difference variance is calculated. The 95th percentile of this variance distribution is selected as the preset coherence threshold. If the currently calculated variance value is lower than the preset coherence threshold (indicating that the sequence is smooth and coherent), this embodiment determines the temporally consistent sequence as the corrected temporally consistent feature sequence.

[0094] In step S14, remote sensing resolution fusion processing is performed based on the corrected temporally consistent feature sequence to obtain the optimized landform model input, including:

[0095] For the corrected temporally consistent feature sequence, a multi-dimensional description and integration of remote sensing image resolution is performed to obtain a fused resolution dataset;

[0096] Based on the fused resolution dataset, a resolution-consistent description set is obtained through spatial grid alignment;

[0097] Spatial smoothness analysis is performed on the resolution-consistent description set to determine the input of the optimized geomorphological model.

[0098] In one implementation, this embodiment performs multi-dimensional description and integration of remote sensing image resolution for the corrected temporally consistent feature sequence obtained in S13.

[0099] It should be noted that the specific implementation of this integration includes spatial interpolation processing and raster fusion processing. Since the corrected temporally consistent feature sequence (e.g., water level change vector) is one-dimensional data existing at discrete sensor locations, while the resolution multidimensional description (e.g., pixel density vector) is two-dimensional spatial data distributed throughout the entire remote sensing image, this embodiment first employs the Inverse Distance Weighting (IDW) method to estimate the feature values ​​of each raster unit within the entire study area based on the known values ​​and spatial coordinates of each sensor location, generating a "temporal feature raster map." Subsequently, this embodiment performs weighted summation processing, performing a raster-by-raster weighted summation between the "temporal feature raster map" and the resolution multidimensional description (rasterized) to obtain the fused resolution dataset.

[0100] It should be noted that the weighting coefficients in the weighted summation process are determined through offline regression analysis. The determination process involves using a historical dataset, with the final prediction accuracy as the dependent variable and time-series consistency features and resolution multidimensional descriptions as independent variables. Regression analysis is used to calculate the contribution of each to the prediction accuracy, and these contributions are then normalized and used as the weighting coefficients.

[0101] It should be noted that this embodiment uses spatial grid alignment based on the fused resolution dataset. Specifically, this alignment is implemented by first defining a unified reference spatial grid (e.g., a 10m x 10m geographic grid); then, bilinear interpolation is used. Specifically, for each new node in the reference spatial grid, this embodiment finds the four closest original data points to that node in the fused resolution dataset, calculates the value of the new node by weighted averaging the values ​​of these four points and their distances to the new node, and maps this value to the corresponding node in the reference spatial grid, thereby obtaining the resolution-consistent description set with consistent spatial coordinates.

[0102] It is worth noting that this embodiment performs spatial smoothness analysis on the resolution-consistent description set (as a raster image). Specifically, this analysis involves using the Laplacian operator to convolve the resolution-consistent description set, calculating its second derivative, and then calculating the variance of this second derivative result. This variance value quantifies the noise level or "roughness" of the spatial data.

[0103] It should be noted that in this embodiment, the calculated variance value is compared with the preset accuracy threshold. The preset accuracy threshold is determined through offline analysis. The determination process is as follows: In this embodiment, a large amount of known historical geomorphic raster data, confirmed as "high-precision," is collected, and its Laplace variance is calculated. The 95th percentile of this variance distribution is selected as the preset accuracy threshold (representing the acceptable upper limit of noise). If the currently calculated variance value is lower than the preset accuracy threshold (indicating that the data is smooth and has low noise), this embodiment determines the resolution-consistent description set as the input to the optimized geomorphic model.

[0104] In step S15, based on the optimized geomorphological model input, geomorphological elevation calculation is performed to determine a high-precision reservoir geomorphological DEM, including:

[0105] A coherent landform description is extracted from the input of the optimized landform model and fused with the deviation impact of the water level change scenario simulation to obtain the deviation impact dataset.

[0106] Based on the deviation impact dataset, deviation correction processing is performed to obtain a deviation optimization description set;

[0107] Based on the deviation optimization description set, the geomorphic elevation is calculated to obtain the high-precision reservoir geomorphic DEM.

[0108] In one implementation, this embodiment extracts elevation values ​​of a series of raster cells from the optimized geomorphic model input (as two-dimensional raster data) obtained in S14, along the same preset path representing the key profile of the reservoir (e.g., the centerline of the main channel), to form a coherent geomorphic description representing continuous terrain features (e.g., elevation sequence [120 m, 125 m, 122 m]), and a series of deviation values ​​spatially interpolated in S14 to form the deviation impact of the water level change scenario simulation (e.g., deviation sequence [0.05 m, 0.07 m, 0.06 m]). This embodiment performs weighted summation processing, fusing the coherent geomorphic description (elevation sequence) and the deviation impact of the water level change scenario simulation (deviation sequence) point by point according to spatial location to obtain the deviation impact dataset.

[0109] It should be noted that the weighting coefficients in the weighted summation process (e.g., 0.6 for geomorphic description and 0.4 for deviation influence) are determined through offline multivariate regression analysis. The determination process is as follows: In this embodiment, a historical dataset is constructed, containing historical geomorphic descriptions, deviation influence data (as independent variables), and the corresponding, known final inundation prediction accuracy (as dependent variables). This embodiment employs a multiple linear regression model and fits the model using the ordinary least squares (OLS) method to calculate the standard regression coefficients for each independent variable. In this embodiment, the absolute values ​​of the standard regression coefficients are normalized (e.g., the absolute value of each coefficient is divided by the sum of the absolute values ​​of all coefficients) to obtain the final weighting coefficients, which sum to 1.

[0110] In another implementation, this embodiment determines the deviation correction index set based on the deviation impact dataset. Specifically, this determination involves first calculating the deviation fluctuation amplitude of the deviation impact dataset (e.g., calculating the standard deviation of the sequence). This embodiment then compares the deviation fluctuation amplitude with a preset fluctuation threshold.

[0111] It should be noted that the preset fluctuation threshold is determined through statistical analysis of the deviation fluctuation amplitude of historical stable geomorphological datasets. The determination process involves calculating the probability density distribution of the fluctuation amplitude of the historical dataset in this embodiment, and selecting the value corresponding to the 95th percentile of this distribution as the preset fluctuation threshold. This percentile is a robust statistical method widely used in system performance engineering. Its basis is that this value can effectively cover the vast majority (95%) of normal operation fluctuations while excluding the interference of the top 5% of extreme outliers, thereby achieving a balance between alarm sensitivity and system stability.

[0112] It is worth noting that if the deviation fluctuation amplitude is greater than the preset fluctuation threshold, this embodiment initiates deviation correction based on a deviation correction index set. This deviation correction index set includes a window size parameter (e.g., a window size of 5) for median filtering. The deviation impact dataset is smoothed according to this parameter to reduce noise interference, resulting in the deviation optimization description set. If the deviation fluctuation amplitude does not exceed the preset fluctuation threshold, the deviation impact dataset is directly determined as the deviation optimization description set.

[0113] In one implementation, this embodiment calculates the geomorphic elevation based on the deviation optimization description set. Specifically, this calculation employs a pre-trained geomorphic elevation regression model.

[0114] It should be noted that the geomorphic elevation regression model (e.g., a Gradient Boosting Decision Tree (GBDT) model) is determined through offline training. The training process, in this embodiment, uses a historical dataset containing historical bias optimization descriptions (as input features) and corresponding known high-precision geomorphic elevation values ​​(e.g., raster cell elevations obtained through satellite radar measurements or LiDAR scans) as labels. This embodiment uses this dataset for supervised learning of the GBDT model, with the training objective being to minimize the root mean square error (RMSE).

[0115] It should be noted that a key hyperparameter of the GBDT model, such as the learning rate or the number of trees (n_estimators), is determined by using a grid search strategy during the offline training phase to optimize within a preset parameter space (e.g., the learning rate range [0.1, 0.01, 0.001]) and selecting the parameter combination with the smallest RMSE value on the independent validation set.

[0116] It is worth noting that after receiving the bias optimization description set (which is itself a feature set corresponding to the two-dimensional raster data of S14), the trained GBDT model in this embodiment traverses each raster cell in the input of the optimized geomorphological model and inputs the feature vector in the bias optimization description set corresponding to that cell into the GBDT model; the model outputs a predicted elevation value for that raster cell (for example, predicting the elevation of that cell to be 121.5 meters). The set of predicted elevation values ​​of all raster cells constitutes an elevation raster map containing (X, Y, Z) spatial location information (i.e., the DEM generated by this invention), and this elevation raster map is the high-precision reservoir geomorphological DEM.

[0117] In step S16, based on the high-precision reservoir topography DEM, geomorphological spatial analysis and classification are performed to obtain high-precision reservoir topography acquisition results, including:

[0118] The slope gradient is calculated based on the high-precision reservoir topography DEM to obtain the slope distribution matrix;

[0119] The slope distribution matrix is ​​used to simulate the water flow path, and path interpolation is performed to determine the continuous path chain.

[0120] Based on the continuous path chain and the slope distribution matrix, the watershed algorithm is used to divide the region into blocks;

[0121] The aforementioned regional blocks are classified to obtain landform category groups;

[0122] The high-precision reservoir landform acquisition results are obtained by integrating the collected data according to the landform category group.

[0123] In one implementation, this embodiment calculates the slope gradient based on the high-precision reservoir topography DEM obtained in S15. Specifically, this calculation involves analyzing the raster data in the elevation raster map, quantifying the terrain slope of each raster cell by calculating the ratio of the elevation difference to the horizontal distance between each raster cell and its adjacent raster cells, and finally obtaining the slope distribution matrix.

[0124] It is worth noting that this embodiment simulates the water flow path using the slope distribution matrix. Specifically, this simulation employs the D8 (Deterministic 8) flow direction algorithm. This algorithm traverses each grid cell in the slope distribution matrix and determines that the water flow (or potential flow) moves in the direction of the largest slope gradient among its eight adjacent grid cells, thereby generating a water flow path sequence. If there are path breaks in the water flow path sequence (e.g., encountering local depressions or flat areas), this embodiment uses a linear interpolation algorithm to fill the break points. Specifically, this interpolation identifies the two endpoints (P1 and P2) of the broken path and calculates a series of new intermediate coordinate points between these two points with a fixed step size. These new coordinate points are then inserted into the water flow path sequence to ensure the acquisition of the continuous path chain.

[0125] In one implementation, this embodiment uses a watershed algorithm to divide the region into blocks based on the continuous path chain and the slope distribution matrix. Specifically, this division uses the continuous path chain as a "river network" or "pour points" in hydrological analysis, and the slope distribution matrix as the flow direction basis. The watershed algorithm starts from the position of the continuous path chain, traces all grid cells flowing towards that path chain in reverse, and aggregates all grid cells merging into the same path chain segment into an independent polygonal region, which is one of the region blocks (i.e., a sub-catchment basin).

[0126] In one implementation, this embodiment performs classification processing on the regional blocks and uses a pre-trained Random Forest classifier.

[0127] It should be noted that the random forest classifier is determined through offline training. The training process, in this embodiment, uses a historical dataset containing input features extracted from the regional blocks and labeled terrain categories (e.g., "valley" or "slope").

[0128] It should be noted that the input feature extraction process is as follows: In this embodiment, firstly, based on the high-precision reservoir topographic DEM (elevation raster map) obtained in S15, the corresponding slope value, aspect value, and curvature value are calculated for each raster cell in the DEM. The slope value and aspect value are obtained by calculating the elevation change rate (first derivative) of each raster cell relative to its neighboring cells. The curvature value is calculated by fitting a 3×3 neighborhood local polynomial surface on each raster cell and calculating the second derivative of the surface. Subsequently, in this embodiment, for each region block, the average values ​​of the slope value, aspect value, and curvature value of all raster cells within the block are calculated, and these average values ​​are used as the input features.

[0129] In this embodiment, the historical dataset is used to supervise the learning of the random forest classifier until the model's classification accuracy on the independent validation set converges. The convergence criterion is that if the improvement in validation set accuracy is less than a preset convergence threshold (e.g., 1e-4) over multiple consecutive training epochs (e.g., 10 epochs), the model is considered converged. A key hyperparameter of the classifier, such as the number of decision trees in the forest (n_estimators), is determined by using a grid search strategy during offline training to optimize within a preset parameter space (e.g., [50, 100, 200] trees) and selecting the parameter value with the highest F1 score or classification accuracy on the independent validation set. The output of the random forest classifier represents the terrain category group.

[0130] For example, in this embodiment, the collected data is integrated according to the landform category group. For instance, for a landform category group classified as "valley", this embodiment integrates and correlates data such as the sensor readings of the area collected in S11 and the predicted elevation value of the area (e.g., 121.5 meters) obtained in S15, thereby obtaining the high-precision reservoir landform collection results containing rich spatial semantics and physical quantitative indicators.

[0131] In summary, this invention solves the technical problems of low accuracy and model distortion in existing technologies caused by heterogeneous data formats, inconsistent time series, and model distortion. It filters noise, unifies formats, and aligns time series of multi-source data, and introduces historical water level sequences for deviation correction to generate an optimized geomorphic model input with consistent time and space. Then, it generates a high-precision DEM by solving geomorphic elevation, and performs spatial analysis and classification on it.

[0132] Reference Figure 2The second embodiment of the present invention provides a reservoir geomorphological data acquisition system based on multi-source data fusion, comprising:

[0133] The data preprocessing module is used to acquire multi-source remote sensing data and sensor data, and perform preprocessing to obtain a unified feature set of remote sensing sensors after denoising.

[0134] The correlation analysis module is used to perform historical water level feature correlation analysis based on the unified feature set of the denoised remote sensing sensor to obtain the matching deviation index.

[0135] The timing adjustment module is used to perform timing alignment adjustment on the matching deviation index and determine the corrected timing consistency feature sequence.

[0136] The fusion processing module is used to perform remote sensing resolution fusion processing based on the corrected temporally consistent feature sequence to obtain an optimized landform model input.

[0137] The geomorphological modeling module is used to calculate the geomorphological elevation based on the input of the optimized geomorphological model and determine the high-precision reservoir geomorphological DEM.

[0138] The spatial analysis module is used to perform geomorphological spatial analysis and classification based on the high-precision reservoir geomorphology DEM to obtain high-precision reservoir geomorphology acquisition results.

[0139] It should be noted that the reservoir geomorphological data acquisition system based on multi-source data fusion provided in this embodiment of the invention is used to execute all the process steps of the reservoir geomorphological data acquisition method based on multi-source data fusion in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0140] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a reservoir geomorphology data acquisition program based on multi-source data fusion. When the processor executes the computer program, it implements the steps described in the various embodiments of the reservoir geomorphology data acquisition method based on multi-source data fusion, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments, such as the data preprocessing module.

[0141] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0142] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0143] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0144] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0145] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0146] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0147] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A reservoir geomorphology data acquisition method based on multi-source data fusion, characterized in that, The method comprises the following steps: acquiring multi-source remote sensing data and sensor data, and preprocessing to obtain a denoised remote sensing sensor unified feature set; performing historical water level feature correlation analysis according to the denoised remote sensing sensor unified feature set to obtain a matching deviation index; performing time sequence alignment adjustment on the matching deviation index to determine a corrected time sequence consistent feature sequence; performing remote sensing resolution fusion processing according to the corrected time sequence consistent feature sequence to obtain an optimized geomorphic model input; performing geomorphic elevation calculation according to the optimized geomorphic model input to determine a high-precision reservoir geomorphic DEM; performing geomorphic spatial analysis and classification according to the high-precision reservoir geomorphic DEM to obtain a high-precision reservoir geomorphic collection result.

2. The reservoir geomorphic data acquisition method based on multi-source data fusion according to claim 1, characterized in that, The method comprises the following steps: acquiring multi-source remote sensing data and sensor data, and preprocessing to obtain a denoised remote sensing sensor unified feature set; acquiring the multi-source remote sensing data and sensor data, and performing noise filtering and format heterogeneous conversion to obtain multi-source fusion initial data; 3. The reservoir geomorphology data acquisition method based on multi-source data fusion according to claim 1, characterized in that, extracting remote sensing image band selection and sensor reading time stamp from the multi-source fusion initial data, and performing time stamp alignment and clustering grouping denoising processing to determine the denoised remote sensing sensor unified feature set. The method comprises the following steps: acquiring multi-source remote sensing data and sensor data, and preprocessing to obtain a denoised remote sensing sensor unified feature set; acquiring the multi-source remote sensing data and sensor data, and performing noise filtering and format heterogeneous conversion to obtain multi-source fusion initial data; 4. The reservoir geomorphology data acquisition method based on multi-source data fusion according to claim 1, characterized in that, extracting remote sensing image band selection and sensor reading time stamp from the multi-source fusion initial data, and performing time stamp alignment and clustering grouping denoising processing to determine the denoised remote sensing sensor unified feature set. The method comprises the following steps: acquiring multi-source remote sensing data and sensor data, and preprocessing to obtain a denoised remote sensing sensor unified feature set; acquiring the multi-source remote sensing data and sensor data, and performing noise filtering and format heterogeneous conversion to obtain multi-source fusion initial data; 5. The reservoir geomorphology data acquisition method based on multi-source data fusion according to claim 1, characterized in that, extracting remote sensing image band selection and sensor reading time stamp from the multi-source fusion initial data, and performing time stamp alignment and clustering grouping denoising processing to determine the denoised remote sensing sensor unified feature set. The method comprises the following steps: acquiring multi-source remote sensing data and sensor data, and preprocessing to obtain a denoised remote sensing sensor unified feature set; acquiring the multi-source remote sensing data and sensor data, and performing noise filtering and format heterogeneous conversion to obtain multi-source fusion initial data; 6. The reservoir geomorphic data acquisition method based on multi-source data fusion according to claim 1, characterized in that, extracting remote sensing image band selection and sensor reading time stamp from the multi-source fusion initial data, and performing time stamp alignment and clustering grouping denoising processing to determine the denoised remote sensing sensor unified feature set. The method comprises the following steps: acquiring multi-source remote sensing data and sensor data, and preprocessing to obtain a denoised remote sensing sensor unified feature set; acquiring the multi-source remote sensing data and sensor data, and performing noise filtering and format heterogeneous conversion to obtain multi-source fusion initial data; extracting remote sensing image band selection and sensor reading time stamp from the multi-source fusion initial data, and performing time stamp alignment and clustering grouping denoising processing to determine the denoised remote sensing sensor unified feature set. The method comprises the following steps: acquiring multi-source remote sensing data and sensor data, and preprocessing to obtain a denoised remote sensing sensor unified feature set; acquiring the multi-source remote sensing data and sensor data, and performing noise filtering and format heterogeneous conversion to obtain multi-source fusion initial data; extracting remote sensing image band selection and sensor reading time stamp from the multi-source fusion initial data, and performing time stamp alignment and clustering grouping denoising processing to determine the denoised remote sensing sensor unified feature set. The method comprises the following steps: acquiring multi-source remote sensing data and sensor data, and preprocessing to obtain a denoised remote sensing sensor unified feature set; acquiring the multi-source remote sensing data and sensor data, and performing noise filtering and format heterogeneous conversion to obtain multi-source fusion initial data; extracting remote sensing image band selection and sensor reading time stamp from the multi-source fusion initial data, and performing time stamp alignment and clustering grouping denoising processing to determine the denoised remote sensing sensor unified feature set.

7. The reservoir geomorphology data acquisition method based on multi-source data fusion according to claim 1, characterized in that, The high-precision reservoir geomorphology DEM is used for geomorphology spatial analysis and classification to obtain high-precision reservoir geomorphology collection results, including: The slope gradient is calculated according to the high-precision reservoir geomorphology DEM to obtain a slope distribution matrix; The water flow path is simulated through the slope distribution matrix, and path interpolation processing is performed to determine a continuous path chain; The region block is divided by using a watershed algorithm according to the continuous path chain and the slope distribution matrix; The region block is classified to obtain a geomorphology category group; The geomorphology category group is integrated to obtain the high-precision reservoir geomorphology collection results.

8. A reservoir geomorphology data acquisition system based on multi-source data fusion, characterized in that, The data preprocessing module is configured to acquire multi-source remote sensing data and sensor data, and perform preprocessing to obtain a denoised remote sensing sensor unified feature set. The correlation analysis module is configured to perform historical water level feature correlation analysis according to the denoised remote sensing sensor unified feature set to obtain a matching deviation index. The time sequence adjustment module is configured to perform time sequence alignment adjustment on the matching deviation index to determine a corrected time sequence consistent feature sequence. The fusion processing module is configured to perform remote sensing resolution fusion processing according to the corrected time sequence consistent feature sequence to obtain an optimized geomorphology model input. The geomorphology modeling module is configured to perform geomorphology elevation calculation according to the optimized geomorphology model input to determine a high-precision reservoir geomorphology DEM. The spatial analysis module is configured to perform geomorphology spatial analysis and classification according to the high-precision reservoir geomorphology DEM to obtain high-precision reservoir geomorphology collection results. ​

Images