Flood inundation early warning method and system for electrical equipment in flood storage and detention area

Abstract

The invention provides a flood inundation early warning method and system for electrical equipment in a flood storage and detention area, and relates to the technical field of flood disaster prevention and reduction of electrical equipment, and the method comprises the steps: obtaining historical meteorological data and historical gate water level data of the flood storage and detention area, carrying out the data cleaning and feature screening, and obtaining a training data set; constructing a water level prediction network model; real-time acquiring real-time meteorological data, forecast meteorological data and real-time sluice water level data of the flood storage and detention area, inputting the data into the water level prediction network model, and generating a future sluice water level prediction value; inputting the future gate water level predicted value and forecast rainfall data into a hydraulic model, and generating a standardized grid file containing rasterized submerging water depth information; and analyzing inundation water depth space distribution based on the standardized grid file, and generating a flood inundation early warning signal in combination with electrical equipment position information. The method is high in prediction precision and high in operation speed, and the water level prediction network model can be expanded and applied to other time series data prediction scenes.

Description

Technical Field

[0001] This invention relates to the field of flood disaster prevention and mitigation technology for electrical equipment, and in particular to a flood inundation early warning method and system for electrical equipment in flood storage and detention areas. Background Technology

[0002] In recent years, floods have become increasingly frequent. Heavy rainfall can directly submerge electrical equipment, including substations, distribution cabinets, and cable lines, causing physical damage. For example, floods can submerge infrastructure such as poles and guy wires, affecting their stability and potentially causing them to collapse. Even without submersion, water can damage the insulation materials of electrical equipment, leading to a significant decrease in insulation performance and causing electrical faults such as leakage and short circuits. This not only endangers personal safety but can also cause secondary disasters such as fires. To cope with floods exceeding standard levels—that is, when the flood flow of a river exceeds its capacity—flood storage and detention areas are activated to temporarily store excess floodwater, thereby reducing peak flow, alleviating flood control pressure on downstream important cities and regions, and minimizing flood damage. However, electrical equipment within flood storage and detention areas is highly vulnerable to flooding. Therefore, providing detailed early warning information for electrical equipment in flood storage and detention areas that may be submerged in the future is particularly important, enabling flood storage and detention area management departments to develop detailed emergency plans in advance.

[0003] In the past, flood warnings were mostly based on establishing hydrological and hydraulic models. This method has strong interpretability and low computational cost, and parameter adjustments can be completed when historical data is limited. However, it relies on a large amount of data, including vegetation, topography, land use, rainfall, and soil properties. Due to the large variety of data required, it is difficult to obtain all of them in practical applications, making engineering use quite challenging. Furthermore, hydrological models struggle to achieve accurate results when there is significant natural disturbance.

[0004] These problems have been alleviated with the emergence of deep learning-based methods. This data-driven approach requires only meteorological data such as rainfall and air pressure, and hydrological data such as water levels. By analyzing the inherent patterns in historical data, a mathematical model for a specific region is derived, leading to more accurate predictions. However, deep learning-based methods have poor interpretability and require high-quality historical data, making them difficult to train directly on infrequent events like floods.

[0005] Meanwhile, existing deep learning-based methods for flood warning can only predict future data by inputting historical meteorological and water level data. However, future forecast meteorological data is readily available and has a significant impact on the prediction results. For example, when a heavy rainstorm occurs in the future, it may cause a sudden increase in water levels. Without guidance from future rainfall information, the prediction network trained on historical data cannot predict this situation, thus giving inaccurate prediction results.

[0006] Therefore, there is an urgent need for a flood inundation early warning method and system that combines the advantages of hydraulic models and deep learning models, and can incorporate future meteorological characteristics for analyzing the impact of electrical equipment in flood storage and detention areas.

[0007] Existing technical document 1 (CN119168373A) provides a flood disaster early warning scheme in the fields of Internet of Things and artificial intelligence. It predicts flow and water level by monitoring key hydrological elements and training hydrological forecasting models, and generates early warning information based on risk assessment models. However, it only relies on artificial intelligence models to output the probability of flood occurrence, and has the shortcoming of not being able to provide specific spatial inundation depth distribution information.

[0008] Existing technical document 2 (CN119514858A) provides a solution in the field of flood impact forecasting for power grid equipment in flood storage and detention areas. It predicts water levels by constructing a neural network water level forecasting model and analyzes the impact on equipment by combining the flow formula and a two-dimensional inundation model. However, its water level prediction module only relies on historical data and has limited prediction capabilities, which is insufficient to effectively integrate forecast data to support accurate long-term prediction. Summary of the Invention

[0009] The purpose of this invention is to design a flood inundation early warning method and system for analyzing the impact of electrical equipment in flood storage and detention areas. Addressing the problems of current methods, and to alleviate the shortcomings of both hydraulic models and deep learning models in flood inundation early warning, this invention combines a deep learning-based method with a hydraulic model, integrating the advantages of both for complementary purposes. A water level prediction module is established using a deep learning-based method, predicting future gate water levels in flood storage and detention areas based on extensive historical data such as rainfall, air pressure, and water levels. Subsequently, a flood inundation module is established using a hydraulic model-based method, performing flood inundation analysis by inputting future gate water level data and forecasted rainfall data. Furthermore, this invention incorporates a data preprocessing module to standardize the input data, achieving refined flood inundation early warning for electrical facilities in flood storage and detention areas.

[0010] To address the problem that existing deep learning models can only predict future sluice gate water levels using historical data and cannot utilize forecast meteorological data, which can easily lead to inaccurate prediction results, this invention designs a future feature extraction network and a cross-attention mechanism. Without affecting operational efficiency, it introduces future features as weights to balance the shortcomings of past features in prediction, thereby achieving more accurate prediction results.

[0011] The present invention adopts the following technical solution.

[0012] The first aspect of the present invention provides a flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas, comprising the following steps:

[0013] Historical meteorological data and historical sluice gate water level data of flood storage and detention areas were obtained, and data cleaning and feature filtering were performed to obtain a training dataset.

[0014] Construct a water level prediction network model and train it offline using the training dataset;

[0015] Real-time meteorological data, forecast meteorological data and real-time sluice gate water level data of flood storage and detention areas are acquired, input into the trained water level prediction network model, and future sluice gate water level prediction values ​​are generated.

[0016] The predicted future gate water level and forecasted rainfall data are input into a pre-constructed hydraulic model to perform flood inundation simulation calculations and generate a standardized grid file containing rasterized inundation water depth information.

[0017] Based on the standardized grid file, the spatial distribution of inundation depth is analyzed, and flood inundation early warning signals are generated in combination with the location information of electrical equipment.

[0018] Optionally, the feature filtering includes:

[0019] The importance coefficients of each meteorological feature were calculated using the random forest algorithm.

[0020] Meteorological features are selected based on preset feature importance thresholds, and rainfall data that is highly correlated with sluice gate water level data is added.

[0021] Optionally, the selected meteorological feature data and the corresponding water level data can be concatenated column by column;

[0022] Paired training data are constructed using a sliding window approach and divided into two parts, past feature data and future feature data, according to a preset ratio.

[0023] Optionally, the water level prediction network model includes:

[0024] The multi-scale downsampling and feature embedding unit is used to perform multi-scale downsampling and feature embedding on past feature data to generate multi-scale high-dimensional features.

[0025] Multiple cascaded past decomposition and mixing modules are used to decompose the multi-scale high-dimensional features into seasonal terms and trend terms, perform seasonal term mixing and trend term mixing respectively, and add the mixed seasonal terms and trend terms to obtain multi-scale fused features;

[0026] A future feature encoding unit is used to encode the future feature data to obtain high-dimensional future features;

[0027] A cross-scale attention interaction unit is used to perform cross-scale attention interaction between the multi-scale fused features and the high-dimensional future features to obtain attention-weighted features.

[0028] The regression unit is used to regress the attention-weighted features to generate water level prediction values.

[0029] Optionally, the flood inundation simulation calculation includes:

[0030] Determine whether the predicted value of the future gate water level exceeds a preset threshold;

[0031] When the threshold is exceeded, the water level data is converted into flow data based on the flow conversion formula;

[0032] The flow data and forecasted rainfall data are written into a pre-built simulation project file;

[0033] Invoke the pre-configured two-dimensional hydrodynamic model to perform inundation calculations;

[0034] Generate simulation result files containing rasterized flood depth data.

[0035] Optionally, the conversion of water level data into flow rate data based on the flow rate conversion formula includes:

[0036] Obtain gate parameters, including gate opening height, number of gate openings, and clear width of gate openings;

[0037] Calculate the flow coefficient based on the water head above the weir;

[0038] The flow rate is calculated based on the flow coefficient, submersion coefficient, and gravitational acceleration.

[0039] Optionally, generating a simulation result file containing rasterized flood depth data includes:

[0040] Extract spatial coordinate data and water level data from the simulation results file;

[0041] Constructing standardized mesh files includes:

[0042] Longitude Mapping Table: A uniform sampling table generated based on the longitude range and interval of grid point coordinates.

[0043] Latitude Mapping Table: A uniform sampling table generated based on the latitude range and interval of grid point coordinates.

[0044] Water level data table: A table of inundation depth data corresponding to the spatial location of the latitude / longitude mapping table;

[0045] The standardized grid file is uploaded to the cloud storage server.

[0046] The second aspect of the present invention provides a flood inundation early warning system for analyzing the impact of electrical equipment in flood storage and detention areas. Based on the flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas described in the first aspect of the present invention, the system includes:

[0047] The data preprocessing module is used to clean and select features from the incoming data, and to divide the data into past feature data and future feature data.

[0048] The prediction triggering module is used to determine whether to trigger the warning process based on real-time water level data and rainstorm warning messages.

[0049] The water level prediction module is used to predict future gate water level data based on past and future feature data processed by the data preprocessing module and a trained water level prediction network model.

[0050] The flood inundation module is used to input forecasted rainfall data and predicted future gate water level data into a pre-built hydraulic model to simulate inundation and generate a standardized geographic data file containing rasterized inundation water depth information.

[0051] The parsing and impact analysis module is used to parse the standardized geographic data file and perform electrical equipment impact analysis.

[0052] A third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when loaded onto the processor, implements the flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas as described in the first aspect of the present invention.

[0053] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas, as described in the first aspect of the present invention.

[0054] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0055] (1) This invention combines the advantages of the latest deep learning models and hydraulic models to achieve complementary strengths. Compared with existing hydraulic model methods, the accuracy of water level prediction in this invention can reach over 90%, far exceeding existing hydraulic models, and it does not require a large amount of data such as vegetation and soil properties, making engineering applications simpler. Compared with existing deep learning model methods, this invention has stronger interpretability and higher accuracy. For example, methods based on LSTM or Transformer can only achieve about 70% to 80% accuracy in water level prediction, and they are completely black boxes. This invention utilizes mathematical methods to extract time series pattern features and deeply integrates them with deep learning methods, ensuring interpretability while enhancing accuracy.

[0056] (2) This invention designs a future feature extraction network, adds an autoencoder to extract future features, and performs cross-attention calculation with the past features extracted by the original deep learning model. The future features introduced by the forecast meteorological data can be used as weights to balance the shortcomings of past features in the prediction, making the final output result more accurate.

[0057] (3) The deep learning model used in this invention and the design of the future feature extraction network are mainly linear layers. Compared with the existing mainstream Transformer-type methods, they have lower operating costs, higher operating efficiency and accuracy, and can achieve millisecond-level calculations even on CPUs.

[0058] (4) The water level prediction network model established in this invention is scalable. When the training data is changed, in addition to water level data, it can also realize the long-term and short-term prediction of time series data such as precipitation, temperature and humidity. Attached Figure Description

[0059] Figure 1 A schematic diagram of a flood inundation early warning method based on the impact analysis of electrical equipment in flood storage and detention areas, according to an embodiment of the present invention;

[0060] Figure 2 : A schematic diagram of a flood inundation early warning system for analyzing the impact of electrical equipment in flood storage and detention areas according to an embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0062] like Figure 1 As shown in the figure, in Embodiment 1, the present invention provides a flowchart of a flood inundation early warning method for electrical equipment in flood storage and detention areas, which includes the following steps:

[0063] S1: Offline access to historical meteorological data and historical sluice gate water level data of flood storage and detention areas, and data preprocessing to obtain training data.

[0064] Preferably, in step S1, to meet the requirements of the training data, the historical meteorological data and historical sluice gate water level data of the flood storage and detention area are cleaned and the features of the historical meteorological data are selected, and the training data is output, specifically as follows:

[0065] S11: Obtain hourly meteorological data and water level data within a specified time range based on the meteorological station code and hydrological station code corresponding to the flood storage and detention area, and then concatenate the data in columns to generate a structured data file;

[0066] For example, S11 includes downloading hourly meteorological data and water level data from January 1, 2019 to August 13, 2024 according to the meteorological station code and hydrological station code corresponding to the flood storage and detention area, concatenating them in columns and saving them locally in CSV file format.

[0067] S12: Perform time series reconstruction on the structured data file, including: generating a continuous time series by sorting by timestamp, interpolating and filling missing timestamps, and performing a relocation operation when there is misaligned data.

[0068] For example, step S12 includes reading a CSV file. Due to the large amount of data during download, the time sequence is out of order. The data is sorted by time. After sorting, it is found that some timestamps are missing. The timestamps are reset hourly from January 1, 2019 to August 13, 2024, and the data is filled into the corresponding positions.

[0069] S13: Perform data cleaning processing, including: deleting abnormal data points based on anomaly marking rules, dynamically filling missing values ​​using a sliding window mean imputation algorithm, and deleting a data column when the missing rate of a single column exceeds a preset threshold;

[0070] For example, S13 includes deleting all outliers, such as -999, in the data, and filling in missing values ​​from top to bottom based on the average of the last 5 values. If more than half of the data in an entire column is missing, the column is deleted directly.

[0071] S14: Perform feature selection using embedding on the cleaned data.

[0072] Specifically, the embedding method is a approach that allows the algorithm to decide which features to use, meaning feature selection and algorithm training occur simultaneously. When using the embedding method, certain machine learning algorithms and models are first used for training to obtain the weight coefficients of each feature. Then, features are selected from largest to smallest based on these weight coefficients. Since this method yields a limited number of features, further additions are made manually based on experience, such as adding rainfall data that is highly correlated with the sluice gate water level data.

[0073] More preferably, S14 includes:

[0074] S141: The Random Forest algorithm is used to jointly train the historical meteorological data and historical sluice gate water level data of the cleaned flood storage and detention area to obtain the importance coefficients of each meteorological feature.

[0075] Specifically, the random forest algorithm is an ensemble learning algorithm based on decision trees. It combines multiple decision trees, and when a prediction is needed, the results of the multiple decision trees are integrated by an aggregator as the final output. Compared to decision tree models, random forests have stronger robustness and higher accuracy. Because random forests are based on decision tree models, they possess feature importance attributes.

[0076] S142: Filter meteorological features that meet the conditions based on the preset feature importance threshold;

[0077] Specifically, S142 includes:

[0078] By setting a feature importance threshold, meteorological features that are greater than the set threshold can be considered to be meteorological features that are highly correlated with water level data. These features are then extracted to obtain the filtered meteorological features.

[0079] S15: The selected meteorological feature data and its corresponding water level data are concatenated in columns, and paired training data are constructed by using a sliding window. The training data is divided into two parts, past feature data and future feature data, according to a preset ratio.

[0080] For example, to construct a paired dataset using a sliding window operation, suppose we want to use 96 hours of feature data to predict water levels for the next 32 hours. First, we directly extract feature data from hours 0 to 127 from the training data. The feature data from the first 96 hours (including water levels) is used as past feature data, and the feature data from the last 32 hours is used as future feature data (excluding water levels; water level data is used as the true value for backpropagation to update network parameters). Second, we slide the extracted data forward by one timestamp, extracting feature data from hours 1 to 128, and then divide it according to the above proportions. We repeat this sliding operation until the end of the dataset.

[0081] S2: Construct a water level prediction network model. Input the training data output from S1 into the network model for offline training. The network model parameters are iteratively updated using the backpropagation algorithm. When the preset training conditions are met, the network parameters after training are output.

[0082] Preferably, in step S2, offline training of the water level prediction network model is completed to meet the requirements for network parameter training, specifically as follows:

[0083] S21: Perform multi-scale downsampling and feature embedding on past feature data in the training data to generate multi-scale high-dimensional features.

[0084] More preferably, S21 includes:

[0085] Input past feature data x, and downsample it using average pooling (AvgPooling(·)) to obtain multi-scale time series data:

[0086] X = {x0,…,x} M}

[0087] Where M is the scale number, and it is projected into a high-dimensional feature X = {x0,…,x} through the embedding layer Embed(·). M The process is as follows:

[0088] X = AvgPooling(x),

[0089] X = Embed(X).

[0090] S22: Multi-scale high-dimensional features are processed through l cascaded past decomposition and mixing modules to obtain a multi-scale fused feature set X. L .

[0091] Specifically, multiple stacked Past-Decomposable-Mixing (PDM) modules are configured to perform information mixing on the multi-scale data in S21. For the l-th Past-Decomposable-Mixing module (PDM)... l (·), input is X l -1 Output X l It can be represented as:

[0092] X l =PDM l (X l-1 ).

[0093] After L stacked past decomposition and mixing modules, its output is:

[0094] More preferably, S22 specifically includes:

[0095] S221: Decompose the multi-scale high-dimensional features obtained in S21 into seasonal and trend terms;

[0096] Specifically, in the l-th past decomposition mixing module, the high-dimensional feature X in S21 is subjected to multi-scale sequence decomposition using SeriesDecomp(·). The sequence decomposition adopts the sequence decomposition unit used in the AutoTransformer paper, which is based on the moving average idea to smooth periodic terms and highlight trend terms. For each scale m∈{0,…,M}, the decomposed seasonal terms are... and trend items It can be represented as:

[0097]

[0098] More specifically:

[0099]

[0100] in, Let be the latent variables to be decomposed. and These represent the trend term and the seasonal term, respectively. For the moving average, average pooling (AvgPooling(·)) is used to perform padding (·) operations to keep the sequence length constant.

[0101] Multi-scale seasonal terms can be obtained. and trend items

[0102] S223: Perform SeasonMix(·) and TrendMix(·) respectively, and then add the seasonal and trend terms at the corresponding scales after mixing to obtain:

[0103]

[0104] Therefore, the output of the l-th past decomposition mixing module is a multi-scale fused feature set.

[0105] S23: Encode the future feature data in the training data to obtain high-dimensional features.

[0106] Specifically, the input is future feature data y, and the designed autoencoder extracts information from the future data and projects it into high-dimensional features. The autoencoder consists of a shallow extraction network ST(·) plus multiple residual general autoencoders RSAE(·).

[0107]

[0108] S24: Fuse the multi-scale feature set X L With high-dimensional features Perform cross-scale attention interaction to generate water level predictions through a regressor.

[0109] Specifically, the multi-scale fused feature set of S22 is output. Input the scale into the predictor Prediction m (·), and with the high-dimensional features in S23 CrossAttention m The (·) operation, where m∈{0,…,M}, yields the output of the attention submodule.

[0110]

[0111] The formula for calculating cross attention is:

[0112]

[0113] Where Q, K, and V are the query matrix, key matrix, and value matrix, respectively, and d k d is the dimension of the key matrix, and T denotes the transpose. First, calculate the dot product of the query and the key, representing the similarity between the two sequences at different positions; then d k As a scaling factor to prevent the value from becoming too large, the softmax function is then used to convert these similarities into a probability distribution, representing the attention weight of the query for each key. Finally, these attention weights are applied to the value V to obtain the output vector. This is equivalent to extracting the information of interest from the past feature sequence and inputting it into the regressor.

[0114] If Q, K, and V are accurate to the sequence level, they can be represented as:

[0115]

[0116] The output of the attention submodule Input the scale into the regression function. m (·) and sum them up to obtain the future dam gate water level data.

[0117]

[0118] S25: Repeat S21-S24, calculate the loss function based on the prediction results and the actual water level value, iteratively update the network parameters through backpropagation, and output the optimized network parameters as the water level prediction network model when the preset training termination condition is met.

[0119] Specifically, S25 includes:

[0120] Repeat steps S21 to S24, using MSE as the loss function for parameter optimization and weight updates.

[0121] The Adam optimizer is used, and the learning rate is dynamically adjusted using the OneCycleLR strategy. After each epoch of training, the MSE (Mean Sequence Equation) is calculated using a validation set. If the results are worse than before for 10 consecutive epochs, training is stopped early, and the network parameters corresponding to the best result are output. This approach achieves better optimization of the model parameters.

[0122] S3: Acquire real-time meteorological data, forecast meteorological data, and real-time gate water level data of the flood storage and detention area. After preprocessing the acquired data, input it into the water level prediction network model and output the predicted value of the future gate water level.

[0123] Preferably, in step S3, to meet the needs of flood inundation calculations, the prediction of future sluice gate water level data is completed, specifically as follows:

[0124] S31: Acquire real-time meteorological data, forecast meteorological data, and real-time gate water level data within the most recent preset monitoring period in the flood storage and detention area. If the water level data within the most recent preset monitoring period exceeds the preset threshold or there is a rainstorm warning signal in the forecast meteorological data, trigger the data preprocessing process; otherwise, re-collect data and repeat the judgment after a preset interval.

[0125] For example, firstly, the real-time water level data and rainstorm warning messages of the flood storage and detention area are accessed online, and the real-time water level of the last 6 hours is obtained; then, it is determined whether the water level data of the last 6 hours exceeds the threshold or whether there is a rainstorm warning. If either of the two exists, data preprocessing continues; otherwise, the real-time water level data and rainstorm warning messages are re-accessed after a 5-minute delay, and the above operation is repeated.

[0126] S32: The data preprocessing process involves accessing the actual meteorological data, forecast meteorological data, and actual water level data of the flood storage and detention area according to the meteorological characteristics selected in S12, and then performing data cleaning. The cleaned actual meteorological data and actual water level data are concatenated as past feature data, and the forecast meteorological data is used as future feature data.

[0127] S33: First, import the optimal network parameters from S25 into the water level prediction network model; then, input past feature data and future feature data into the water level prediction network model to infer the future gate water level data.

[0128] S4: Input the forecast meteorological data and future gate water level prediction values ​​of the flood storage and detention area into the pre-constructed hydraulic model to perform flood inundation simulation calculations, and generate a standardized geographic data file containing rasterized inundation water depth information, providing a basis for subsequent analysis and electrical equipment impact analysis.

[0129] Preferably, in step S4, the calculation of the inundation result NC file is completed as required for the flood inundation process simulation, specifically as follows:

[0130] S41: Extract the forecast rainfall data for the flood storage and detention area from the cleaned forecast meteorological data in S32, and input the forecast rainfall data and the future gate water level data in S33 into the flood inundation simulation process.

[0131] S42: Determine whether the predicted water level at the sluice gate exceeds a preset threshold. If it does not exceed the threshold, return to the data acquisition process. If it exceeds the threshold, convert the water level data into flow data based on the flow conversion formula, and write the flow data and rainfall data together into a pre-built simulation engineering file.

[0132] Specifically, if the input future gate water level data does not exceed the threshold, the process returns to S31 and executes again; otherwise, the flow conversion submodule is used to convert the water level data into flow data, and the flow data and forecast rainfall data are written together into the HDF file previously built using HEC-RAS. At this point, the HDF file already includes information such as the topography and land use of the flood storage and detention area.

[0133] The specific formulas for the traffic conversion submodule are as follows:

[0134]

[0135] Where Q is the flow rate, σ s is the submergence coefficient, which defaults to 1 in free outflow. e is the gate opening height, n is the number of gate openings, b is the net width of the gate opening, g is the acceleration due to gravity, and H0 is the head over the weir. U0 is the flow coefficient, which comprehensively reflects the influence of the gate opening shape and the relative opening of the gate on the discharge flow. Its calculation formula is:

[0136] u0 = 0.60 - 0.18e / H(0.1 <e / H<0.65),

[0137] In this case, since the approach velocity head is usually small and often ignored in the calculation, H is used instead of H0. In this case, H is also the head above the weir, which is equal to the water level height minus the bottom elevation.

[0138] S43: Call the pre-configured two-dimensional hydrodynamic model, perform inundation calculation based on the simulation engineering file containing terrain features, land use parameters and output data from step S42, and generate a simulation result file containing rasterized inundation depth data;

[0139] Specifically, the 2D flood inundation simulation submodule, which encapsulates HEC-RAS, is invoked. Using the HDF file in S42 as input, the inundation depth at each grid point in the flood storage area is calculated, resulting in a new HDF file. Based on the established engineering file required for the model, the 2D flood inundation simulation submodule overwrites the initial data in the engineering file by reading rainfall, flow, and other data, and then invokes the HEC-RAS model to simulate the 2D flood inundation process. Since the model's engineering file is based on DEM data, preprocessing of the DEM data is necessary to reduce errors caused by its inherent limitations. This part mainly involves improving the DEM values ​​at corresponding locations based on actual hydrological characteristics, such as the spatial distribution of dikes. After adjusting the DEM data, considering computational efficiency, it is resampled to a 100m resolution topography, and the gate locations are marked in the engineering file. Land use data and other information for the forecast area are collected to determine the parameters in the 2D flood inundation simulation submodule. Finally, hydrological data from actual flood inundation events are used to adjust the parameters in the engineering file, making it more consistent with reality.

[0140] S44: Standardize the spatial coordinate data and water level data in the simulation result file to construct a standardized grid file containing a longitude mapping table, a latitude mapping table, and a water level data table, and upload the standardized grid file to the cloud storage server.

[0141] Specifically, for the HDF file output by the 2D flood inundation simulation submodule, the coordinates of the grid center points and the water level data are extracted. The grid point coordinates are generated by Mercator projection of latitude and longitude. First, the maximum and minimum values ​​of the grid point coordinates (x and y) are found and standardized into a uniformly acquired format. Then, a table is created using the maximum and minimum values ​​and coordinate intervals, and the acquired point coordinates are preprocessed, assigning edge points to the nearest center point. Next, the water level data is filled into the table based on the one-to-one correspondence between the preprocessed coordinates and the water level data, and the position of all water level data in the table is obtained. Finally, since the longitude intervals corresponding to the coordinate intervals are different at different latitudes, two tables are created: one for longitude and one for latitude. These two tables, along with the water level table, are written into the NC file, and the output NC file is uploaded to Alibaba Cloud via OSS.

[0142] By following the steps above, a more precise early warning system can be established for the flooding process in flood storage and detention areas, enabling the management departments of these areas to develop detailed emergency plans in advance.

[0143] Furthermore, due to the unique design of the water level prediction network in this invention, which includes the extraction and mixing of seasonality and trend data from historical data, feature extraction from forecast data, and attention-weighted regression of both, these are all universal designs that can predict any time series data. This processing method enables the network to perform excellently on various time series data with forecast data, including wind speed, temperature, and humidity. Through experiments on a large amount of meteorological data, the water level prediction network proposed in this invention achieves an accuracy of over 90% on various meteorological data, far outperforming methods such as LSTM, CNN, and Transformer. At the same time, the network of this invention shows an improvement of approximately 10% compared to the case without forecast data, demonstrating the improved scalability and effectiveness of the network.

[0144] In summary, the flood inundation early warning method for electrical equipment in flood storage and detention areas provided by this invention has the following technical effects:

[0145] 1. In view of the problem that existing flood warning schemes cannot provide specific spatial distribution information of inundation depth, this invention realizes the gridded spatial distribution output of inundation depth in flood storage and detention areas by inputting the water level prediction results into a hydraulic model to simulate the inundation evolution.

[0146] 2. To address the problem that existing water level prediction models struggle to effectively integrate forecast data and have limited long-term prediction accuracy, this invention improves the long-term accuracy of water level prediction and the utilization rate of forecast data by designing a multi-scale temporal prediction network that includes an autoencoder and a cross-attention mechanism.

[0147] 3. To address the problem that existing hydraulic models rely on a large number of geographical parameters, which makes engineering applications difficult, this invention replaces the calculation of complex hydrological parameters with a deep learning model, thereby reducing the model's dependence on professional data such as vegetation and soil properties.

[0148] 4. To address the problem of low computational efficiency in time series prediction using existing deep learning models, this invention improves prediction speed by using a lightweight network architecture based on linear layers, achieving millisecond-level water level estimation in a CPU environment.

[0149] 5. To address the problem that existing early warning systems cannot adapt to multiple types of time-series data, this invention expands the high-precision prediction capability for multiple meteorological elements such as precipitation, temperature, and humidity through a general feature decomposition and fusion mechanism.

[0150] like Figure 2 As shown, in Embodiment 2, the present invention provides a flood inundation early warning system for electrical equipment in flood storage and detention areas, used to implement the flood inundation early warning method for electrical equipment in flood storage and detention areas described in Embodiment 1, comprising the following modules:

[0151] Data preprocessing module: This module performs data cleaning and feature selection on the incoming data, and divides the data into past feature data and future feature data to meet the needs of the subsequent water level prediction module.

[0152] Prediction trigger module: It exists as the startup program for the entire system. It only runs subsequent programs when certain conditions are met, thus saving computational costs.

[0153] Water level prediction module: This module is used to predict future gate water level data by inputting past and future feature data, which facilitates the calculation of subsequent flood inundation modules.

[0154] Flooding Module: This module uses the input future gate water level data and forecast rainfall data to call the two-dimensional flooding simulation submodule of HAC-RAS encapsulated in this invention for automated calculations, and finally provides an early warning of the future flooding situation in the flood storage and detention area.

[0155] Analysis and Impact Analysis Module: After parsing the future gate water level data and the NC file uploaded by the flood inundation module, it displays the data on the front end and performs an impact analysis on electrical equipment in the flood storage and detention area to determine the equipment affected by the flood inundation process.

[0156] The overall system is divided into two parts: the training process and the inference process. The training process includes a data preprocessing module and a water level prediction module that requires offline training. Preprocessed CSV data is input into the water level prediction module, and backpropagation is performed to update the network parameters. The inference process includes a prediction triggering module, a data preprocessing module, a water level prediction module using parameters provided by the training process, and a flood inundation module. First, real-time water level data and rainstorm warning messages are input into the prediction triggering module. If the triggering conditions are met, the data preprocessing module continues to run. Then, the processed data is input into the water level prediction module to obtain the predicted future water level data, which is then input into the analysis and impact analysis module. Simultaneously, the forecast rainfall data is extracted from the processed data. Next, the forecast rainfall data and future water level data are input together into the flood inundation module to obtain the NC file of the inundation results. Finally, the NC file is uploaded to Alibaba Cloud, and the analysis and impact analysis module is called to parse and visualize the file.

[0157] Embodiment 3 of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements a flood inundation early warning method for electrical equipment in flood storage and detention areas as described in Embodiment 1.

[0158] Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a flood inundation early warning method for electrical equipment in flood storage and detention areas as described in Embodiment 1.

[0159] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas, characterized in that, Includes the following steps: Historical meteorological data and historical sluice gate water level data of flood storage and detention areas were obtained, and data cleaning and feature filtering were performed to obtain a training dataset. Construct a water level prediction network model and train it offline using the training dataset; Real-time meteorological data, forecast meteorological data and real-time sluice gate water level data of flood storage and detention areas are acquired, input into the trained water level prediction network model, and future sluice gate water level prediction values ​​are generated. The predicted future gate water level and forecasted rainfall data are input into a pre-constructed hydraulic model to perform flood inundation simulation calculations and generate a standardized grid file containing rasterized inundation water depth information. Based on the standardized grid file, the spatial distribution of inundation depth is analyzed, and flood inundation early warning signals are generated in combination with the location information of electrical equipment.

2. The flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas according to claim 1, characterized in that: The feature selection includes: The importance coefficients of each meteorological feature were calculated using the random forest algorithm. Meteorological features are selected based on preset feature importance thresholds, and rainfall data that is highly correlated with sluice gate water level data is added.

3. The flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas according to claim 2, characterized in that: The selected meteorological feature data and the corresponding water level data are concatenated in columns; Paired training data are constructed using a sliding window approach and divided into two parts, past feature data and future feature data, according to a preset ratio.

4. The flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas according to claim 1, characterized in that: The water level prediction network model includes: The multi-scale downsampling and feature embedding unit is used to perform multi-scale downsampling and feature embedding on past feature data to generate multi-scale high-dimensional features. Multiple cascaded past decomposition and mixing modules are used to decompose the multi-scale high-dimensional features into seasonal terms and trend terms, perform seasonal term mixing and trend term mixing respectively, and add the mixed seasonal terms and trend terms to obtain multi-scale fused features; A future feature encoding unit is used to encode the future feature data to obtain high-dimensional future features; A cross-scale attention interaction unit is used to perform cross-scale attention interaction between the multi-scale fused features and the high-dimensional future features to obtain attention-weighted features. The regression unit is used to regress the attention-weighted features to generate water level prediction values.

5. A flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas according to claim 1, characterized in that: The flood inundation simulation calculations include: Determine whether the predicted value of the future gate water level exceeds a preset threshold; When the threshold is exceeded, the water level data is converted into flow data based on the flow conversion formula; The flow data and forecasted rainfall data are written into a pre-built simulation project file; Invoke the pre-configured two-dimensional hydrodynamic model to perform inundation calculations; Generate simulation result files containing rasterized flood depth data.

6. A flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas according to claim 5, characterized in that: The process of converting water level data into flow rate data based on the flow rate conversion formula includes: Obtain gate parameters, including gate opening height, number of gate openings, and clear width of gate openings; Calculate the flow coefficient based on the water head above the weir; The flow rate is calculated based on the flow coefficient, submersion coefficient, and gravitational acceleration.

7. A flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas according to claim 1, characterized in that: The generated simulation result file containing rasterized inundation depth data includes: Extract spatial coordinate data and water level data from the simulation results file; Constructing standardized mesh files includes: Longitude Mapping Table: A uniform sampling table generated based on the longitude range and interval of grid point coordinates. Latitude Mapping Table: A uniform sampling table generated based on the latitude range and interval of grid point coordinates. Water level data table: A table of inundation depth data corresponding to the spatial location of the latitude / longitude mapping table; The standardized grid file is uploaded to the cloud storage server.

8. A flood inundation early warning system for analyzing the impact of electrical equipment in flood storage and detention areas, based on the flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas as described in any one of claims 1-7, characterized in that, The system includes: The data preprocessing module is used to clean and select features from the incoming data, and to divide the data into past feature data and future feature data. The prediction triggering module is used to determine whether to trigger the warning process based on real-time water level data and rainstorm warning messages. The water level prediction module is used to predict future gate water level data based on past and future feature data processed by the data preprocessing module and a trained water level prediction network model. The flood inundation module is used to input forecasted rainfall data and predicted future gate water level data into a pre-built hydraulic model to simulate inundation and generate a standardized geographic data file containing rasterized inundation water depth information. The parsing and impact analysis module is used to parse the standardized geographic data file and perform electrical equipment impact analysis.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is loaded into the processor, it implements a flood inundation early warning method for analyzing the impact of electrical equipment in flood storage and detention areas, according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements a flood inundation early warning method based on the impact analysis of electrical equipment in flood storage and detention areas, according to any one of claims 1-7.

Images