Drainage basin flood process automatic deduction method based on time sequence analysis

Abstract

The invention relates to the technical field of time sequence data processing, in particular to a drainage basin flood process automatic deduction method based on time sequence analysis, which comprises the following steps: obtaining drainage basin historical hydrological data, processing by applying adaptive wavelet transform, and extracting an intermediate frequency wavelet feature sequence and a transform residual sequence as data features; establishing a first calculation model, and generating a first predicted water level sequence; obtaining an error sequence based on the first predicted water level sequence and the actual water level sequence; constructing a second calculation model by using the data features and the rainfall data, and fitting the error sequence by using the second calculation model to obtain a prediction error sequence; performing weighted fusion on the first predicted water level sequence, the predicted error sequence and the rainfall sequence to generate a second predicted water level sequence; and combining the second predicted water level sequence with a hydrodynamic numerical calculation model, and obtaining predicted data through flow field calculation and analysis. According to the method, high-precision automatic deduction is realized through automatic processing and hybrid modeling.

Description

Technical Field

[0001] The present invention relates to the technical field of time series data processing, and particularly to an automated deduction method for basin flood processes based on time series analysis. Background Art

[0002] Effective time series data calculation analysis and quantitative modeling of basin flood processes are the basis for understanding the dynamics of complex hydrological systems. However, the complexity of the actual hydrological sequence composition and its often contained noise and non-stationary characteristics pose challenges to the design of data processing algorithms (such as non-stationarity processing, data normal transformation) and the accuracy of model fitting.

[0003] In the prior art, traditional physical models are computationally complex, difficult to determine parameters, and often sensitive to parameters; common time series calculation methods use linear statistical models that are difficult to effectively process complex hydrological digital signals or machine learning calculation models whose algorithm processes are difficult to interpret and rely on a large amount of data.

[0004] Currently, there are still obvious technical challenges in automatically constructing an integrated calculation process that effectively couples different calculation modules and high-efficiency time series modeling and distribution estimation modules to achieve high-precision quantitative modeling.

[0005] Therefore, an automated deduction method for basin flood processes based on time series analysis is proposed. Summary of the Invention

[0006] The purpose of the present invention is to provide an automated deduction method for basin flood processes based on time series analysis, which realizes high-precision automated deduction through intelligent processing and hybrid modeling. First, obtain historical hydrological data of the basin, process it using adaptive wavelet transform, and extract the intermediate-frequency wavelet feature sequence and the transform residual sequence as data features; establish a first calculation model to generate a first predicted water level sequence; based on the first predicted water level sequence and the actual water level sequence, obtain an error sequence; use the data features and rainfall data to construct a second calculation model, and fit the error sequence using the second calculation model to obtain a predicted error sequence; perform weighted fusion on the first predicted water level sequence, the predicted error sequence, and the rainfall sequence to generate a second predicted water level sequence; combine the second predicted water level sequence with a hydrodynamic numerical calculation model to obtain predicted data through flow field calculation and analysis.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] An automated deduction method for basin flood processes based on time series analysis, comprising:

[0009] Obtain historical hydrological data of the basin, and select a part of the historical hydrological data of the basin as a training set;

[0010] Process the training set using adaptive wavelet transform, automatically determine the transform parameters and perform denoising operations to obtain a denoised training set, and extract the intermediate-frequency wavelet feature sequence and the transform residual sequence as data features;

[0011] Based on the denoised training set, establish a first calculation model to generate a first predicted water level sequence; based on the first predicted water level sequence and the actual water level sequence, obtain an error sequence;

[0012] Use the data features and rainfall data to construct a second calculation model, use the second calculation model to fit the error sequence to obtain a predicted error sequence; perform weighted fusion on the first predicted water level sequence, the predicted error sequence and the rainfall data to generate a second predicted water level sequence; use the second predicted water level sequence as input, combined with the hydrodynamic numerical calculation model, and obtain flood prediction data through flow field calculation and analysis.

[0013] Preferably, the transform parameters include the wavelet transform threshold, the type of wavelet basis function and the number of wavelet decomposition levels. The specific steps for automatically determining the transform parameters are as follows:

[0014] Quantify the data dispersion degree of the highest-frequency wavelet detail coefficients of the training set, automatically generate an estimated value of the signal background noise intensity based on the dispersion degree, and calculate the specific threshold value of the wavelet coefficients according to the noise intensity estimated value and the general threshold calculation formula;

[0015] Select the type of wavelet basis function by matching the skewness value and sample entropy value of the training set with predefined standards; determine the wavelet decomposition level by analyzing the time length distribution of the data and applying the rule of cumulative energy percentage.

[0016] Preferably, the process of obtaining the intermediate-frequency wavelet feature sequence includes: calculating the local energy value of the selected intermediate decomposition level wavelet detail coefficients within a sliding window to form an intermediate-frequency wavelet feature sequence that changes with time.

[0017] Preferably, the transform residual sequence includes: the remaining part after subtracting all the detail components extracted in the wavelet transform process from the original signal sequence.

[0018] Preferably, the process of obtaining the error sequence includes: subtracting the first predicted water level sequence from the actual water level observation value to obtain an error sequence.

[0019] Preferably, the second calculation model includes:

[0020] A rainfall encoder configured to receive a rainfall sequence as input and process it using a first recurrent neural network layer to generate a first hidden state sequence;

[0021] An intermediate frequency encoder, configured to receive an intermediate frequency wavelet feature sequence as input and process it using a second recurrent neural network layer to generate a second hidden state sequence;

[0022] A residual encoder, configured to receive a transform residual sequence as input and process it using a third recurrent neural network layer to generate a third hidden state sequence;

[0023] A feature fusion unit, connected to the output ends of three input coding branches, configured to combine the first, second, and third hidden state sequences at respective corresponding time steps to generate a combined feature sequence;

[0024] A temporal attention module, analyzes the combined feature sequence through a temporal attention mechanism to generate a context feature sequence, and the context feature sequence is the result of weighted aggregation of historical information;

[0025] A prediction output unit, a multi-layer perceptron, whose input end is connected to the output end of the temporal attention module, configured to calculate and output a prediction error sequence based on the context vector sequence.

[0026] Preferably, the weighted fusion of the first predicted water level sequence, the predicted value of the error sequence, and the rainfall data includes:

[0027] Based on the predicted water level sequence, the predicted error sequence, and the rainfall data at the current time step, a set of dynamic weight coefficients is calculated;

[0028] The dynamic weight coefficients are used to weight the first predicted water level sequence, the predicted error sequence, and the rainfall data to generate a second predicted water level sequence.

[0029] Preferably, the obtaining of the prediction data through flow field calculation and analysis includes: using the second predicted water level sequence as an input condition to drive the hydrodynamic numerical calculation model, and processing the spatio-temporal variation results output by the hydrodynamic numerical calculation model to determine the flood peak arrival time and the flood peak water level at a predetermined position within the basin, and the spatio-temporal variation results include spatial distribution data of water level and water depth.

[0030] Preferably, the obtaining of the prediction data through flow field calculation and analysis further includes:

[0031] Based on the spatial distribution data of the maximum water surface elevation during the flood output by the hydrodynamic numerical calculation model, in combination with the digital elevation model, a submerged water depth distribution map within the study area is determined.

[0032] Compared with the prior art, the beneficial effects of the present invention are:

[0033] 1. By introducing adaptive wavelet transform, the present invention can automatically optimize and determine the key parameters of signal processing according to the characteristics of actual hydrological time series, such as noise and non-stationarity, and extract multi-scale features that can better reflect physical processes. This automatic processing method based on data characteristics overcomes the limitations in the prior art where parameter selection depends on manual work and it is difficult to adapt to data changes, providing more reliable and targeted data input for subsequent high-precision quantitative modeling.

[0034] 2. The present invention effectively couples a basic time series model that can capture linear patterns and an error correction model that can learn complex non-linear dynamics, and is driven by the multi-scale features extracted previously. A learnable dynamic weighting mechanism is used to intelligently fuse the outputs of different models according to real-time context information. This coupling and adaptive fusion method enables the model to more comprehensively and accurately simulate complex water level changes, significantly improving the accuracy of time series modeling and achieving the goal of high-precision quantitative modeling.

[0035] 3. The present invention automatically couples and drives the high-precision data-driven time series modeling results with a hydrodynamic model based on physical processes, constructing an end-to-end integrated calculation process from data input to final state estimation. This process can not only automatically output high-precision single-point water level simulations, but also further automatically calculate and output the flood peak characteristics at multiple key locations as well as spatialized inundation ranges and water depth maps, achieving a comprehensive and high-precision estimation of flood states. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic flow chart of an automatic deduction method for basin flood processes based on time series analysis provided by an embodiment of the present invention;

[0037] Figure 2 It is a processing flow chart of adaptive wavelet transform provided by an embodiment of the present invention;

[0038] Figure 3 It is a flow chart of a second calculation model provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] Please refer to Figures 1 to 3 , the present invention provides an automatic deduction method for basin flood processes based on time series analysis, and the technical solutions are as follows:

[0041] Example 1

[0042] There is a large error between the current and the true flood peak arrival time and flood peak water level, and the prediction of the inundation range cannot be achieved. Therefore, there is an urgent need for a new solution that can automatically deduce and predict the flood arrival time, flood peak water level, and inundation range.

[0043] Figure 1 It is a schematic flow chart of an automated deduction method for the flood process in a river basin based on time series analysis provided in this embodiment.

[0044] As Figure 1 shown, an automated deduction method for the flood process in a river basin based on time series analysis includes:

[0045] S10. Obtain the historical hydrological data records of the river basin and select a part of them from the historical hydrological data of the river basin as the training set.

[0046] First, obtain the historical hydrological data records of Basin A from the regional hydrological database and meteorological database. The historical hydrological data records include the recording timestamp, the corresponding water level data, and the rainfall data that can reflect the precipitation situation in the river basin during this time period. In this embodiment, the hourly time series data of the outlet section of Basin A in the past 5 years are collected, and the data format is a CSV file, which contains three columns: time, water level, and basin rainfall data. Then, preprocess the obtained data, check the data integrity and format, convert the timestamp to a unified time format, delete the duplicate entries, detect the outliers using the IQR method, and regard them as missing values, and then use the linear interpolation method to fill all the missing values. Table 1 shows the historical hydrological data.

[0047] Table 1 Historical Hydrological Data

[0048]

[0049] Furthermore, after completing the above data collection and preprocessing, the hourly, complete, and outlier-free hydrometeorological time series data of Basin A in the past five years are obtained. The complete time series data are strictly divided into a training set, a validation set, and a test set in chronological order and according to a predetermined ratio. In this embodiment, the continuous data segment starting from the starting time point of the complete time series data and accounting for the first 90% of the total length in terms of time span is defined as the training set; the continuous data segment immediately following the end time point of the training set and accounting for the middle 6% of the total length in terms of time span is defined as the validation set; the continuous data segment immediately following the end time point of the validation set and accounting for the last 4% of the total length in terms of time span is defined as the test set.

[0050] S20. Process the training set using adaptive wavelet transform; automatically determine the transform parameters and perform denoising operations to generate a denoised training set.

[0051] Figure 2 This is the flowchart of the adaptive wavelet transform processing in this embodiment.

[0052] First, select a preliminarily selected wavelet basis function. In this embodiment, db4 is used to perform multi-level wavelet decomposition on the training set time series. The decomposition is to obtain the detail coefficient sequence of the highest frequency. Quantify the data dispersion degree of the detail coefficient sequence by calculating the absolute median difference of the detail coefficient sequence. Estimate the standard deviation of Gaussian white noise in the signal based on the absolute median difference value to estimate the noise intensity. The specific calculation formula is as follows:

[0053]

[0054] Among them, represents the estimated noise intensity, represents the absolute median difference, represents the detail coefficient sequence of the highest frequency, represents a preset constant parameter used to convert into an estimated value of the standard deviation. In this embodiment, based on the assumption that the noise approximately follows a Gaussian distribution, the value of the parameter is 0.6745.

[0055] Furthermore, adopt the general threshold calculation logic to automatically calculate the threshold according to the estimated noise intensity and the length of historical hydrological data. The specific calculation is as follows:

[0056]

[0057] Among them, represents the estimated noise intensity, represents the automatically calculated threshold, represents the length of historical hydrological data after preprocessing, represents the natural logarithm.

[0058] Furthermore, preset a candidate library containing multiple commonly used wavelet basis functions with different characteristics. The candidate library includes:

[0059] Daubechies series (for example, db4, db8), which is characterized by having compact support and good vanishing moments, and is suitable for analyzing asymmetric and signals containing transient features.

[0060] Symlets series (for example, sym4, sym8), which is an approximately symmetric version of Daubechies wavelet, reduces phase distortion while maintaining good characteristics, and is suitable for signals that require symmetry.

[0061] The Coiflets family (e.g., coif2, coif5), whose scaling function and wavelet function both have good vanishing moments and are approximately symmetric, is suitable for processing relatively smooth signals.

[0062] Furthermore, calculate the skewness value and sample entropy value for the preprocessed water level time series training set; calculate the skewness value of the numerical distribution of this time series to quantitatively evaluate its symmetry. A skewness close to zero indicates that the data distribution is more symmetric, and being far from zero indicates that the distribution has a left skew or right skew; calculate the sample entropy value of this time series. When calculating, the embedding dimension m and tolerance threshold r need to be set. The tolerance threshold is set based on the standard deviation of the sequence to quantitatively evaluate its complexity or irregularity. The smaller the sample entropy value, the more regular and self-similar the time series pattern is, and the larger the value, the more complex, random, and irregular the sequence is. Then, based on the calculated skewness value and sample entropy value.

[0063] Compare with the pre-set discrimination criteria and, according to the matching rules, automatically determine the most suitable wavelet basis function from the candidate library: preset the symmetry criterion threshold for distinguishing the level of symmetry and the complexity criterion threshold for distinguishing the level of complexity. These thresholds can be determined based on prior knowledge or experimental optimization. If the calculated symmetry index shows that the signal has good symmetry (below the symmetry criterion threshold) and the complexity index shows that the signal pattern is regular (below the complexity criterion threshold), then it is determined that the signal characteristics are suitable for selecting a wavelet basis function with good symmetry and smooth shape. At this time, preferentially select specific wavelet models from the Symlets family or Coiflets family in the candidate library. If the calculated symmetry index shows that the signal has obvious asymmetry (reaching or exceeding the symmetry criterion threshold), or the complexity index shows that the signal contains more irregular components (reaching or exceeding the complexity criterion threshold), then it is determined that the signal characteristics are more suitable for selecting a wavelet basis function that can effectively capture local details or asymmetric structures. At this time, preferentially select specific wavelet models from the Daubechies family in the candidate library, especially those with higher vanishing moments.

[0064] Furthermore, after determining the wavelet basis function, automatically determine the decomposition level used for this wavelet transform. This process adopts a rule based on the cumulative energy percentage. First, for the preprocessed water level time series training set with length N, use the previously automatically selected wavelet basis function to perform an initial multi-layer wavelet decomposition with sufficient depth (for example, the number of decomposition layers can refer to ), with the aim of obtaining the coefficients of each layer for energy analysis. Then, for each sequence of detail coefficients of each layer generated by this initial depth decomposition and the finally obtained sequence of approximation coefficients, the sum of the squares of all coefficient values within the sequence is calculated to obtain the contained energy. The cumulative energy percentage rule is applied to automatically determine the number of decomposition layers that ultimately need to be adopted, that is, the sum of the energies of all detail coefficients and the energy of the final approximation coefficients. Then, starting from the lowest decomposition layer, i.e., the first layer, the energies of the detail coefficients of each layer are accumulated layer by layer, and the percentage of this cumulative energy in the total energy of the signal is calculated. Find the decomposition layer number at which the cumulative energy percentage reaches or exceeds a preset high percentage threshold. Determine the found layer number as the number of decomposition layers finally adopted for this wavelet transform. For example, set the cumulative energy percentage to 99.5%. When decomposing to the seventh layer, the cumulative energy percentage is 99.2%, which does not reach the set energy percentage, so continue the decomposition. When decomposing to the eighth layer, the cumulative energy percentage is 99.7%, which is greater than the set energy percentage, and automatically determine that the number of decomposition layers for this decomposition is 8 layers.

[0065] By establishing a calculation process based on the characteristics of the data itself (such as noise level, symmetry, complexity, energy distribution, and time scale) and clear rules, the objectivity of the selection of key parameters (denoising threshold, wavelet basis function type, decomposition level) and the adaptability to different input data are realized. It not only avoids the subjective randomness and a large amount of repetitive labor that may be introduced by traditional manual parameter adjustment, ensures the repeatability of the analysis results and significantly improves the processing efficiency, but also improves the effectiveness of wavelet transform processing by selecting parameters that better match the internal characteristics of the data and the hydrological physical process, thereby possibly providing higher-quality data input for the subsequent training of the model and contributing to improving the simulation accuracy of the entire flood propagation system.

[0066] S30. Extract the intermediate-frequency wavelet feature sequence and the transform residual sequence as data features.

[0067] Perform a complete n-layer discrete wavelet decomposition on the water level sequence of the training set using the automatically determined parameters to obtain an n-layer sequence of detail coefficients and an n-layer approximation coefficient sequence. Use the automatically determined denoising threshold to perform threshold denoising on the detail coefficient levels that need to be used for feature extraction to reduce the impact of noise on energy calculation. Select one or more intermediate decomposition levels that can represent the predefined intermediate-frequency range. The detail coefficient sequences of these selected levels are considered to mainly contain the information on the dynamic changes of the main body of the flood.

[0068] Further, a width W (counted by the number of data points) of a sliding window and a step size S are preset in advance. The selection of the window width W should be able to capture the time scale of the local dynamic features of interest. For each selected intermediate-frequency detail coefficient sequence, the sliding window with width W and step size S is used to traverse the intermediate-frequency detail coefficient sequence. At each position where the window slides, the sum of squares of all W coefficients within the window is calculated to obtain the local energy value at the time point of this level. This process is repeated for each selected intermediate-frequency level to generate multiple independent local energy sequences. For the multiple independent local energy sequences, at each same time point, their energy values are summed to obtain the intermediate-frequency wavelet feature sequence representing the flood dynamic process.

[0069] By calculating the local energy of the signal in the intermediate-frequency band, the dynamic intensity and change rate of the flood process are effectively captured, and the occurrence, development, and recession stages of the flood event are more prominent compared with the original water level signal. At the same time, by selecting a specific frequency band and calculating the sum-of-squares energy, it plays a role in enhancing the features and suppressing noise and trend interference, providing a more effective and informative input for the machine learning model that relies on these dynamic information for error correction.

[0070] After performing the above-mentioned determined n-layer discrete wavelet decomposition on the training set water level time series, coefficient sequences representing different frequency components are obtained. Among them, the approximation coefficient sequence of the nth layer is obtained after the most number of low-pass filtering and downsampling, so it contains the component information with the lowest frequency and the slowest change in the signal. Then, n zero vectors with the same length as the corresponding level detail coefficient sequence are created, and these zero vectors will replace the actual detail coefficients and be input into the inverse transform. The approximation coefficient sequence and the zero vectors are subjected to multi-layer discrete wavelet inverse transform, ensuring that the same wavelet basis function is used as when performing the forward transform. After performing n-layer wavelet inverse transform, an output sequence is obtained, and the length of the output sequence is the same as the length of the initially input preprocessed water level sequence. The sequence obtained through selective reconstruction is the transform residual sequence of this embodiment. It is a relatively smooth sequence, mainly presenting the long-term trend, seasonal pattern, or slow change of the base flow in the original hydrological time series, and all detail components with higher frequencies and faster changes have been filtered out in this process.

[0071] Effectively separate and extract the very slow-changing base flow or long-term trend component from the original complex signal, providing a smoother input for the model that needs to distinguish or separately process trend information in the future. As the coarsest scale component of the wavelet multi-scale decomposition, together with the previously extracted intermediate-frequency feature sequence, it constitutes a decomposition representation of the original signal at different time scales, which helps to more comprehensively understand the hydrological process.

[0072] S40. Based on the denoised training set, establish a first calculation model to generate a first predicted water level sequence; based on the first predicted water level sequence and the actual water level sequence, obtain an error sequence.

[0073] First, perform the ADF test to determine the stationarity of the sequence. If the test result shows that the sequence is non-stationary, perform differencing operations on it until the differenced sequence passes the stationarity test. The number of differencing operations performed is the order d. For the differenced stationary sequence, calculate and analyze its autocorrelation function (ACF) graph and partial autocorrelation function (PACF) graph. According to the observed combination patterns of the truncation and trailing of the ACF and PACF, the candidate value ranges of the autoregressive order p and the moving average order q can be initially determined. Subsequently, combined with the principle of minimizing the information criterion AIC, through calculation and comparison within the candidate range, select the optimal (p, q) combination, and finally obtain the optimal order (p, d, q) of the ARIMA model.

[0074] Furthermore, use the training set data to fit the ARIMA model of the selected order through maximum likelihood estimation (MLE), thereby establishing the first calculation model. Input the training data into the already trained ARIMA model again for prediction to obtain the first predicted water level sequence. By subtracting the predicted value of the first predicted water level sequence from the actual water level observation value at each time point in the training set, calculate the error sequence.

[0075] The ARIMA model effectively captures and fits the linear autocorrelation law in the denoised hydrological sequence, providing a reliable and repeatable benchmark prediction for the simulation of the entire complex flood process. At the same time, accurately calculate the fitting error sequence of the ARIMA model. The error sequence clearly separates the non-linear dynamics and external factor influences in the original signal that cannot be explained by the linear model, providing a clear and necessary learning objective and training basis for the subsequent error correction model.

[0076] S50. Use the data features and rainfall data to construct a second calculation model, and use the second calculation model to fit the error sequence to obtain a predicted error sequence.

[0077] Figure 3 This is the flowchart of the second calculation model provided in this embodiment.

[0078] The second calculation model includes:

[0079] A rainfall encoder, configured to receive a rainfall sequence as input, and process it using a first recurrent neural network layer to generate a first hidden state sequence;

[0080] An intermediate frequency encoder, configured to receive an intermediate frequency wavelet feature sequence as input, and process it using a second recurrent neural network layer to generate a second hidden state sequence;

[0081] A residual encoder, configured to receive a transformed residual sequence as input, process it using a third recurrent neural network layer, and generate a third hidden state sequence;

[0082] A feature fusion unit, connected to the outputs of three input encoding branches, configured to combine the first, second, and third hidden state sequences at respective time steps to generate a combined feature sequence;

[0083] A temporal attention module, analyzes the combined feature sequence through a temporal attention mechanism to generate a context feature sequence, which is the result of weighted aggregation of historical information;

[0084] A prediction output unit, a multi-layer perceptron, whose input end is connected to the output end of the temporal attention module, configured to calculate and output a prediction error sequence based on the context vector sequence.

[0085] Specifically, the rainfall sequence collected in step S10, the intermediate-frequency wavelet feature sequence obtained in step S30, and the transformed residual sequence are used as the input of the second calculation model. The prediction error sequence output by the second calculation model and the error sequence are used as the loss to perform regression on the second calculation model. Each recurrent neural network layer of the second calculation model adopts an LSTM structure, including 2 layers, with 64 units in each layer. The feature fusion unit combines the hidden states through a concatenation operation. The temporal attention module adopts the Bahdanau attention mechanism, and the time window is 24 time steps. The prediction output unit is an MLP with 2 hidden layers, 32 nodes in each layer, uses the ReLU activation function and adds Dropout (ratio 0.2) after the hidden layer, and the output layer is linearly activated. When training, the Adam optimizer is used, the initial learning rate is 10 −3 , the batch size is 64, the mean square error is used as the loss function, and the early stopping mechanism is adopted during the training process. The loss on the validation set is monitored, and when the validation set loss does not improve within the preset number of epochs, the training is stopped to prevent overfitting of the model, and the maximum number of training epochs is set to 300 epochs.

[0086] By effectively fusing a variety of key input information, including external driving forces and internal dynamic features and background states extracted by previous adaptive wavelet transforms, it is possible to conduct a more comprehensive and in-depth learning of the error sources and patterns of the ARIMA model. It not only fully utilizes the multi-scale information decomposition results provided by wavelet analysis but also effectively captures complex hydrological response relationships, realizing high-precision and high-reliability flood deduction.

[0087] S60. Perform weighted fusion on the first predicted water level sequence, the prediction error sequence, and the rainfall sequence to generate a second predicted water level sequence.

[0088] By training a multi-layer perceptron (MLP) to calculate a weight coefficient in real time at each time point based on the current running state of the model and relevant input information. The MLP contains 2 hidden layers, with each layer configured with, for example, 32 nodes. The activation function of the hidden layer is preferably ReLU. The number of nodes in its input layer depends on the dimension of the input information. Its output layer contains 3 nodes and uses the Softmax activation function to ensure that the three output weight values are non-negative and sum to 1, corresponding to the contribution weights of the first predicted water level sequence, the prediction error sequence, and the rainfall data respectively. To improve the generalization ability, Dropout is added after the hidden layer.

[0089] Further, prepare the training data, which includes the predicted water level sequence, the predicted error sequence value sequence, and the rainfall data sequence within the corresponding time period, and perform unit unification and normalization processing. Then, input the output sequence of the first calculation model, the output sequence of the second calculation model, and the rainfall sequence into the multi-layer perceptron network. Through the forward calculation of the network, obtain a set of dynamic weight coefficients corresponding to each sample, acquire the corresponding original first predicted water level sequence, prediction error sequence, and rainfall data, and use the corresponding weights output by the MLP to perform weighted combination on these three values to calculate the predicted water level value combination of each sample to obtain the second predicted water level sequence. Compare the obtained second predicted water level sequence with the corresponding actual observed water level sequence, calculate the loss value measuring the current prediction error through the mean square error, and based on the calculated loss value, apply the backpropagation algorithm. Starting from the loss value, calculate the gradient of the loss value with respect to all trainable parameters in the multi-layer perceptron network along the reverse path of the neural network. The gradient indicates how each parameter should be adjusted to decrease the loss function value. Based on the Adam gradient descent optimization algorithm, calculate and update all trainable parameters of the multi-layer perceptron network, and complete the training iteration for a preset number of rounds on the MLP to obtain the trained MLP model.

[0090] Further, in the actual prediction stage, for each time point to be predicted, obtain the first predicted water level sequence, prediction error sequence, and rainfall data at that time point, input them into the trained MLP network, obtain a set of dynamic weights at that time point, and use the dynamic weights generated by the MLP to perform dynamic weighting on the first predicted water level sequence, prediction error sequence, and rainfall data to obtain the second predicted water level sequence.

[0091] By automatically learning and adjusting the contribution degrees of each information source to the final prediction result, a more refined and effective fusion strategy than fixed weights or simple addition is achieved. This data-driven dynamic multi-weight combination method can further improve the accuracy and robustness of the overall flood deduction.

[0092] S70. Take the second predicted water level sequence as the input and pass it to the hydrodynamic numerical calculation model. The predicted data is obtained through flow field calculation and analysis.

[0093] In this embodiment, the river channel of Basin A is relatively regular and the focus is mainly on the flood evolution in the river channel. Therefore, the one-dimensional hydrodynamic model HEC-RAS is selected for subsequent calculation and analysis. The selected HEC-RAS model needs to be pre-constructed and parameterized based on the actual geographical information of Basin A, including inputting the geometric data of the river channel (for example, the cross-sectional shape, section spacing, and river centerline coordinates obtained through field measurement or remote sensing), which can be obtained by using equipment such as total stations, GPS-RTK for cross-section measurement, LiDAR point cloud data, and aerial photogrammetry. Set the flow resistance parameters (for example, the Manning roughness coefficient n value of each river reach or section estimated or measured), and the flow resistance parameters can be obtained by referring to the typical roughness empirical value ranges given in relevant hydraulic engineering manuals and literature for different riverbed materials, vegetation cover types, and river channel morphologies. Then, combined with on-site investigation, observe the actual situation of the river channel and floodplain, and conduct empirical estimation. Finally, use remote sensing images or land use cover maps to assign corresponding initial roughness values to different surface cover types in the study area.

[0094] Furthermore, define the possible hydraulic structures such as bridges, culverts, and weirs in the model. In the constructed HEC-RAS model, pre-specify one or more calculation sections for which the calculation results need to be output and the flood characteristics analyzed (for example, it can be the section where an important city is located downstream, a specific hydrological station section, or other locations where flood risk assessment is required).

[0095] Furthermore, the training process of the hydrodynamic data calculation model includes using the selected historical flood event data as the input boundary conditions of the model to run the model, comparing the simulated water level process at the internal observation point position of the model with the measured process at this point, calculating the error index between the two using the Nash efficiency coefficient, and systematically adjusting the model parameters according to the error magnitude and the matching situation between the simulated process and the measured process. Repeat running the model and adjusting the parameters, and this process is carried out by combining genetic algorithms and particle swarm optimization until the simulated results at the internal observation points reach the preset satisfactory standard.

[0096] Furthermore, to verify the reliability of the model, measured data of one or multiple independent historical flood events at different stages are selected for model verification. Keeping the determined model parameters unchanged, the data of the verification event is used as the boundary condition to run the model. The simulated output of the model at the internal observation points is compared with the measured process of the verification event, and the Nash efficiency coefficient is used to calculate the error index. If the error index in the verification stage meets the preset accuracy requirements, it indicates that the hydrodynamic model has good prediction ability and can be used for subsequent automated deduction of flood processes.

[0097] Furthermore, the second predicted water level sequence generated in the previous step is used as the main upstream input boundary condition to drive the operation of the HEC-RAS model. The water level sequence is converted into a flow hydrograph according to the water level - discharge conversion table. The time range and calculation time step for the unsteady flow calculation of the HEC-RAS model are set, and then the model is started for unsteady flow hydrodynamic calculation. After the HEC-RAS model calculation is completed, detailed results of the water level, flow rate, and flow velocity varying with time at all predetermined positions in the model are output. Then, the complete water level time series corresponding to the predetermined cross-section is extracted from the model output result database, the maximum value in the series is found and recorded as the predicted flood peak water level of the cross-section, and the time point when the maximum value first appears is found and recorded as the predicted flood peak arrival time of the cross-section.

[0098] Compared with directly using the second predicted water level sequence to determine the flood peak water level and flood peak arrival time, using the second predicted water level sequence and the hydrodynamic model can focus on the water level changes at multiple predetermined positions, advancing from statistical law deduction to physical process simulation, thereby obtaining a more comprehensive, accurate, and physically meaningful understanding and prediction results of the flood evolution process of the entire basin.

[0099] Furthermore, the flood peak water levels reached at each calculation cross-section during the entire flood process are obtained, as well as the water surface line along the river longitudinal direction formed by the highest water levels of these cross-sections. The highest water levels of these cross-sections are interpolated into the entire potentially flooded area to obtain a continuous spatial water surface elevation distribution. Then, digital elevation model (DEM) data covering the study area is obtained. The DEM data needs to have sufficient resolution and be consistent with the coordinate system of the hydrodynamic model. The DEM data can be sourced from aerial photogrammetry and publicly available regional DEM datasets. The maximum water surface elevation data of the spatial water surface elevation distribution is subjected to raster or spatial overlay analysis with the DEM data. For each grid cell of the DEM, the maximum water surface elevation value during the flood period at that location and the ground elevation value at that location are obtained. If the maximum water surface elevation is greater than the ground elevation, it is determined that the cell is flooded, and for the flooded cell, the inundation depth is obtained. If not flooded, the water depth is recorded as 0.

[0100] Furthermore, all the raster cells determined to be inundated are visualized to generate a flood extent map reflecting the potentially maximum affected area of this flood event. The maximum inundation depth values calculated for each inundated cell are rendered using different colors or gray levels to generate an inundation depth distribution map, visually showing the severity of inundation in different areas.

[0101] By outputting the inundation depth distribution map that can be used for decision support to quantify the severity of inundation in different areas, compared with the simplified method of static mapping based only on single-point water levels, it can more realistically reflect the spatial variation of the actual water surface line, thus providing more accurate and reliable inundation results.

[0102] An automated deduction method for the flood process in a basin based on time series analysis proposed by the present invention automatically processes data and extracts key multi-scale dynamic features and trend information using adaptive wavelet transform, improving the adaptability to different data. Then, by combining the ARIMA linear model and the non-linear error correction model, the limitations of a single model are effectively overcome, and the simulation accuracy of the water level time series is improved. Dynamic weighting further intelligently fuses the model outputs according to real-time information, enhancing the robustness of the results. Finally, by driving the hydrodynamic model of the physical process with the accurate time series results, the prediction is extended from a single point to space, and the flood peak information at multiple key locations as well as intuitive flood extent and depth distribution maps are automatically obtained. This method has significant improvements in terms of automation level, prediction accuracy, comprehensiveness of results, and physical meaning, providing a more powerful and intelligent technical support for flood risk management and emergency decision-making.

[0103] Embodiment 2

[0104] For Basin B, there is a need for more accurate prediction of the flood peak arrival time, flood peak water level, and inundation method in the future. And in Basin B, there are situations such as wide and shallow river channels, broad floodplains, complex terrain, and variable water flow directions. Therefore, an automated deduction method for the flood process in a basin based on time series analysis proposed by the present invention is introduced.

[0105] First, the hourly water level data and corresponding rainfall data of Basin B in the past ten years are obtained, and the collected data is preprocessed. Then, the first 95% of the time data is divided into the training set, 95% to 98% is the validation set, and 98% to 100% is the test set. Then, adaptive wavelet transform is used to automatically determine the threshold and wavelet basis function for the training set, and the cumulative energy percentage rule (setting the threshold to 99.5%) is applied to automatically determine the final decomposition level to be adopted. In this embodiment, after calculating the energy and cumulative percentage analysis for the ten-year training set data, it is determined that when decomposed to the 9th layer, the cumulative energy percentage first exceeds 99.5%. Therefore, in this embodiment, the final decomposition level adopted for this wavelet transform is automatically determined to be 9 layers.

[0106] Further, the intermediate frequency wavelet feature sequence and the transform residual sequence are extracted as data features. Then, the training set denoised by wavelet transform is input into the ARIMA model to train and fit the parameters. The trained and fitted ARIMA model is used to predict the training set to obtain an error sequence. Finally, the intermediate frequency wavelet feature sequence, the transform residual sequence, and the rainfall sequence are input into the second calculation model for training, and the error sequence is used as the target output to perform regression on the model.

[0107] Further, a multi-layer perceptron MLP is trained to perform weighted fusion on the first calculation model and the second training model to generate a second predicted water level sequence. Then, the simulated time series is passed to the hydrodynamic numerical calculation model. Since there are situations in Basin B such as wide and shallow river channels, broad floodplains, complex terrain, and variable water flow directions, a 2D hydrodynamic model is used in this embodiment. Specifically, the two-dimensional hydrodynamic calculation module HEC-RAS2D provided by the HEC-RAS software is used. The two-dimensional model needs to be pre-constructed based on the digital elevation model (DEM) covering the study area and parameters such as the set roughness. During calculation, the final simulated water level time series obtained in the previous step is used as the upstream input boundary condition of the HEC-RAS2D model, and after setting other necessary boundary conditions, its two-dimensional (or coupled) unsteady flow simulation is run.

[0108] After the simulation is completed, the results output by HEC-RAS2D include the spatio-temporal distribution information of water level, water depth, and flow velocity within the calculation area. By extracting the time series of predetermined concern points and analyzing, the flood peak water level and arrival time at these positions are determined. By overlaying and analyzing the spatial distribution of the maximum water surface elevation output by the model with the DEM data, the maximum inundation range map and the inundation water depth distribution map are determined.

[0109] To verify the effect of the method of the present invention, an experimental comparison is made between the method of the present invention and the original prediction method of Basin B on the test set. The experimental comparison calculates the absolute error of the prediction by subtracting the measured value from the calculated value of each hourly hydrological element; the relative error is the absolute error divided by the measured value, expressed as a percentage; the degree of agreement between the calculation process and the measured process of the hydrological element can be measured by the coefficient of determination as an index. The specific calculation formula is as follows:

[0110]

[0111] represents the coefficient of determination, represents each time point, represents the predicted value at each time point, represents the actual test value at each time point, represents the mean value of the measured values within this time period, and n represents the time period length.

[0112] Table 2 shows the comparison data of flood peak prediction time and water level. It can be seen from the table that the method of the present invention significantly reduces the prediction time of the flood peak and the error of the flood peak water level prediction compared with the original prediction method of Basin B.

[0113] Table 2 Comparison of Flood Peak Prediction Time and Water Level

[0114]

[0115] In addition, to further prove the effect of the method of the present invention, the coefficient of determination is used to compare the data on the test set. The verification results show that the coefficient of determination of the method of the present invention is 0.86, and the coefficient of determination of the original method of B is 0.75. The prediction results of the method of the present invention are significantly better than those of the original method of B, verifying the effectiveness and advancement of the method of the present invention in realizing the automated deduction of the high-precision flood process.

[0116] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An automated deduction method for the basin flood process based on time series analysis, characterized in that, Including: Obtain the historical hydrological data of the basin, and select a part of the historical hydrological data of the basin as the training set; Apply adaptive wavelet transform to process the training set, automatically determine the transform parameters and perform denoising operations to obtain a denoised training set, and extract the intermediate-frequency wavelet feature sequence and the transform residual sequence as data features; Based on the denoised training set, establish a first calculation model to generate a first predicted water level sequence; based on the first predicted water level sequence and the actual water level sequence, obtain an error sequence; Use the data features and rainfall data to construct a second calculation model, use the second calculation model to fit the error sequence to obtain a predicted error sequence; perform weighted fusion on the first predicted water level sequence, the predicted error sequence and the rainfall data to generate a second predicted water level sequence; use the second predicted water level sequence as the input, combined with the hydrodynamic numerical calculation model, and obtain flood prediction data through flow field calculation and analysis.

2. The automated deduction method for the basin flood process based on time series analysis according to claim 1, characterized in that, The transform parameters include wavelet transform threshold, wavelet basis function type and wavelet decomposition level. The specific steps for automatically determining the transform parameters are as follows: Quantify the data dispersion degree of the highest-frequency wavelet detail coefficients of the training set, automatically generate an estimated value of the signal background noise intensity based on the dispersion degree, and calculate the specific threshold value of the wavelet coefficient according to the noise intensity estimated value and the general threshold calculation formula; Select the wavelet basis function type by matching the skewness value and sample entropy value of the training set with predefined standards; determine the wavelet decomposition level by analyzing the time length distribution of the data and applying the rule of cumulative energy percentage.

3. The automated deduction method for the basin flood process based on time series analysis according to claim 2, characterized in that, The process of obtaining the intermediate-frequency wavelet feature sequence includes: calculating the local energy value of the selected intermediate decomposition level wavelet detail coefficients within the sliding window to form an intermediate-frequency wavelet feature sequence that changes with time.

4. A method for automatically deducing the basin flood process based on time series analysis according to claim 1, characterized in that, The transform residual sequence includes: the remaining part after subtracting all the detail components extracted in the wavelet transform process from the original signal sequence.

5. A method for automatically deducing the basin flood process based on time series analysis according to claim 1, characterized in that, The process of obtaining the error sequence includes: subtracting the first predicted water level sequence from the actual water level observation value to obtain an error sequence.

6. The automated deduction method for the basin flood process based on time series analysis according to claim 1, characterized in that The second calculation model includes: A rainfall encoder configured to receive a rainfall sequence as input and process it using a first recurrent neural network layer to generate a first hidden state sequence; An intermediate-frequency encoder configured to receive an intermediate-frequency wavelet feature sequence as input and process it using a second recurrent neural network layer to generate a second hidden state sequence; A residual encoder configured to receive a transform residual sequence as input and process it using a third recurrent neural network layer to generate a third hidden state sequence; A feature fusion unit connected to the outputs of the three input coding branches, configured to combine the first, second and third hidden state sequences at corresponding time steps to generate a combined feature sequence; A temporal attention module analyzes the combined feature sequence through a temporal attention mechanism to generate a context feature sequence, and the context feature sequence is the result of weighted aggregation of historical information; A prediction output unit, a multi-layer perceptron, whose input end is connected to the output end of the temporal attention module, configured to calculate and output a predicted error sequence based on the context vector sequence.

7. A method for automatically deducing the basin flood process based on time series analysis according to claim 1, characterized in that The weighted fusion of the first predicted water level sequence, the predicted value of the error sequence, and the rainfall data includes: Based on the predicted water level sequence, the predicted error sequence, and the rainfall data at the current time step, a set of dynamic weight coefficients is calculated; The first predicted water level sequence, the predicted error sequence, and the rainfall data are weighted using the dynamic weight coefficients to generate a second predicted water level sequence.

8. A method for automatically deducing the basin flood process based on time series analysis according to claim 1, characterized in that The prediction data obtained through flow field calculation and analysis includes: using the second predicted water level sequence as the input condition to drive the hydrodynamic numerical calculation model, and processing the spatio-temporal variation results output by the hydrodynamic numerical calculation model to determine the flood peak arrival time and flood peak water level at a predetermined position within the basin. The spatio-temporal variation results include the spatial distribution data of water level and water depth.

9. A method for automatically deducing the basin flood process based on time series analysis according to claim 8, characterized in that, The prediction data obtained through flow field calculation and analysis further includes: Based on the spatial distribution data of the maximum water surface elevation during the flood period output by the hydrodynamic numerical calculation model, combined with the digital elevation model, a submerged water depth distribution map within the study area is determined.

Images