Flood session division method and device based on deep learning

Abstract

The invention provides a flood session division method and device based on deep learning, and belongs to the technical field of flood data analysis, and the method comprises the steps: dividing historical flood data into flood sequence data and non-flood sequence data; and performing data expansion on the flood sequence data to obtain target flood sequence data. And performing data splicing on the target flood sequence data and the non-flood sequence data according to a time sequence to obtain a flow-rainfall time sequence sample. And based on the flow-rainfall time sequence sample, training the multi-layer long and short-term memory network by using the target loss function to obtain a flood prediction model. And performing flood session division on the flow-rainfall time sequence data to be divided by using the flood prediction model to obtain a flood session division result. According to the flood session division method and device based on deep learning provided by the invention, the accuracy of flood session identification and division in a complex flood scene can be improved.

Description

Technical Field

[0001] The present disclosure belongs to the technical field of flood data analysis, and more specifically, relates to a method and device for flood event division based on deep learning. Background Art

[0002] Hydrological observation data is usually continuously recorded annually for analysis and statistics. During a year, when a basin encounters continuous rainfall or the superposition of floods from different tributaries, a compound flood process with multiple peaks often forms. In practical applications, it is necessary to extract individual flood events. The extracted flood data can be used for compiling hydrological yearbooks, studying flood characteristics, and providing samples for parameter calibration of flood forecasting models.

[0003] However, traditional threshold-based peak screening methods are difficult to accurately identify the effective recession process of the valleys between peaks. Traditional methods are prone to missing the judgment of "low peak, long duration" flood events, and such floods are of great significance in water resources scheduling and ecological impact assessment.

[0004] In summary, traditional flood event identification and division methods cannot adapt to the non-linear dynamics of flood processes and the time-varying nature of environmental changes, and there is a problem of low accuracy in flood event identification when dealing with complex flood scenarios. Summary of the Invention

[0005] The purpose of the present disclosure is to provide a method and device for flood event division based on deep learning to improve the accuracy of flood event identification and division in complex flood scenarios.

[0006] In the first aspect of the embodiments of the present disclosure, a method for flood event division based on deep learning is provided, including: Dividing historical flood data into flood sequence data and non-flood sequence data; performing data augmentation on the flood sequence data to obtain target flood sequence data; splicing the target flood sequence data and the non-flood sequence data according to the time series to obtain a flow-rainfall time series sample; Based on the flow-rainfall time series sample, training a multi-layer long short-term memory network using a target loss function to obtain a flood prediction model; Using the flood prediction model to perform flood event division on the flow-rainfall time series data to be divided to obtain a flood event division result.

[0007] In the second aspect of the embodiments of the present disclosure, a device for flood event division based on deep learning is provided, including: A data processing module, configured to divide historical flood data into flood sequence data and non-flood sequence data; perform data augmentation on the flood sequence data to obtain target flood sequence data; and splice the target flood sequence data and the non-flood sequence data according to a time series to obtain a flow-rainfall time series sample. A model training module, configured to train a multi-layer long short-term memory network based on the flow-rainfall time series sample by using a target loss function to obtain a flood prediction model. A model flood event division module, configured to use the flood prediction model to perform flood event division on the flow-rainfall time series data to be divided to obtain a flood event division result.

[0008] In a third aspect of the embodiments of the present disclosure, there is provided an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the above-mentioned flood event division method based on deep learning are implemented.

[0009] In a fourth aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned flood event division method based on deep learning are implemented.

[0010] The beneficial effects of the flood event division method and device based on deep learning provided by the embodiments of the present disclosure are as follows: On the one hand, for complex flood scenarios, the embodiments of the present disclosure use a multi-layer LSTM network to model the long-term dependence relationship of the flow-rainfall sequence, avoiding the mechanical segmentation defect of a single threshold, and can improve the accuracy of flood boundary division.

[0011] On the other hand, the embodiments of the present disclosure alleviate the model bias caused by scarcity by augmenting flood samples, enabling the model to fully learn the transition characteristics between flood peaks and base flows, and enhancing the generalization ability of the model. The hierarchical feature extraction ability of the multi-layer LSTM can adapt to hydrological differences in different basins, achieve accurate division of the complete flood boundary, and combining with the target loss function can improve the accuracy and robustness of model training.

[0012] In summary, the method provided by the embodiments of the present disclosure can improve the accuracy of flood event recognition and division in complex flood scenarios, and provide standardized high-precision results for hydrological yearbook compilation and forecast model calibration. Description of the Drawings

[0013] To more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present disclosure. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0014] Figure 1 It is a schematic flowchart of a flood event division method based on deep learning provided by an embodiment of the present disclosure; Figure 2 It is a schematic flowchart of a flood event division method based on deep learning provided by another embodiment of the present disclosure; Figure 3 It is a single-peak and multi-peak single-event data graph of a typical flood event interval in a specific basin provided by an embodiment of the present disclosure; Figure 4 It is a multi-peak multi-event data graph of a specific basin-type flood event interval provided by an embodiment of the present disclosure; Figure 5 It is a flood event division result graph of a deep learning model for a specific basin provided by an embodiment of the present disclosure; Figure 6 It is a structural block diagram of a flood event division device based on deep learning provided by an embodiment of the present disclosure; Figure 7 It is a schematic block diagram of an electronic device provided by an embodiment of the present disclosure. Specific Embodiments

[0015] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present disclosure. However, those skilled in the art should clearly understand that the present disclosure can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present disclosure.

[0016] To make the purpose, technical solutions, and advantages of the present disclosure clearer, the following will be described through specific embodiments with reference to the drawings.

[0017] Please refer to Figure 1 , Figure 1 It is a schematic flowchart of a flood event division method based on deep learning provided by an embodiment of the present disclosure, and this method may include S101 to S103.

[0018] S101: Divide the historical flood data into flood sequence data and non-flood sequence data. Perform data augmentation on the flood sequence data to obtain the target flood sequence data. Concatenate the target flood sequence data and the non-flood sequence data in time series to obtain the flow-rainfall time series samples.

[0019] In this embodiment, dividing the historical flood data into flood sequence data and non-flood sequence data includes: Divide the time interval corresponding to the historical flood data into flood event intervals and non-flood periods. The flood event interval is the time interval during which a flood event occurs. The non-flood period is the time interval during which no flood event occurs.

[0020] Use the flood data corresponding to the flood event interval as the flood sequence data.

[0021] Use the base flow data corresponding to the non-flood period as the non-flood sequence data.

[0022] In this embodiment, before dividing the historical flood data into flood sequence data and non-flood sequence data, it further includes: performing data preprocessing on the historical flow data and historical rainfall data of the target basin to obtain the historical flood data.

[0023] In this embodiment, the historical flood data refers to the time series data of flow and rainfall monitored for a long time within the target basin, including the original records of flood events and non-flood periods. The target basin refers to a specific basin to be analyzed and referenced. The flood sequence data refers to the flow and rainfall data corresponding to the flood event interval extracted from the historical data, including the complete flood process such as rising water, peak value, and falling water. The non-flood sequence data refers to the flow and rainfall data corresponding to the period (non-flood period) when no flood event occurs, mainly including base flow or low-fluctuation data.

[0024] Exemplarily, as Figures 2 to 5 shown, the flood event division method based on deep learning may include: 1. Data collection and data preprocessing: For each monitoring station within a specific basin, collect the time series data of its flow and rainfall. Specifically, the rain gauge monitoring station provides historical rainfall data, and the flow monitoring station provides the corresponding flow data and records the corresponding time points. The flow data refers to the river channel flow at each time point, reflecting the flood intensity. The rainfall data refers to the rainfall amount at each time point, serving as the flood triggering condition.

[0025] Perform data preprocessing operations such as outlier detection on the collected data, including clearing errors and outliers, such as negative values or unreasonably high values. Screen the consistency of all data in time, and unify the timestamp format and data frequency. The timestamp refers to the time coordinate accurate to minutes to ensure the time consistency of the data.

[0026] Specifically, since the rainfall and flow rate in natural processes should exhibit continuous and stable changes, sudden numerical changes are regarded as anomalies. In this embodiment, a moving window method can be used for outlier detection, including: (1) Set the window size: Select a window size W. For example, W can be the previous 5 time points of the current time point.

[0027] (2) Calculate the sliding window statistic: For each time point t, calculate the mean value of multiple data points within the preset window W before this time point and the standard deviation :

[0028]

[0029] Wherein, is the rainfall or flow rate data corresponding to time point i.

[0030] (3) Outlier detection: Perform outlier detection on the current data point based on the following constraints: Condition 1: The current data point suddenly drops to 0, and the previous data point or the next data .

[0031] Condition 2:

[0032] For outliers and missing values in each data sequence, linear interpolation is used for filling. At time point the missing data and data , respectively use the values of their adjacent known data points before and after and for linear interpolation. , , , The rainfall interpolation formula is:

[0033] The flow interpolation formula is:

[0034] The flow interpolation formula is:

[0035] Wherein, is the missing rainfall data at the current time point t, is The previous adjacent rainfall data, is the next adjacent rainfall data, is the missing flow data at the current time point t, is the previous adjacent flow data of, is the next adjacent flow data of, is the time point adjacent to the previous of the missing data, is the time point adjacent to the next of the missing data.

[0036] (4)Time consistency processing: Convert the timestamps of all monitoring stations into a unified format, unified as "YYYY-MM-DD HH:MM".

[0037] 2. Flood event interval division: Based on specific research purposes and requirements, divide the typical single-peak and multi-peak flood events into flood event intervals. Each flood event can be represented as a flood event interval, that is, the start time and end time of the flood event. Determine the peak flow and peak shape characteristics corresponding to the flood event in each flood event interval. The peak flow is used to distinguish the flood level, and the peak shape characteristics such as single-peak / multi-peak type and rising / falling water slope can reflect the flood process form. Mark the flood event intervals, peak flows, and peak shape characteristics in the processed flood data to obtain flood sequence data. Divide the sequence data outside the flood event intervals into non-flood periods to obtain non-flood sequence data.

[0038] 3. Data augmentation and splicing: After obtaining the labeled flood sequence data, in order to enhance the generalization ability of the model, the data needs to be augmented. The methods of data augmentation can include operations such as stretching and compressing time series, and lengthening and shortening time lengths. Splice the augmented data and non-flood period data to obtain a complete time series flood division sample.

[0039] For the generation of non-flood period data, non-flood period data or base flow data with small flow and small fluctuations can be extracted from historical flood data. Intersperse these non-flood period data in the final long-time sample to enhance the model's ability to identify non-flood periods. To construct longer and more diverse time series samples, randomly splice the processed flood event data of various types and non-flood period data. The randomized splicing method can prevent the model from learning fixed flood and non-flood sequence patterns and enhance the generalization ability of the model. The spliced data sequence is expressed as:

[0040] Among them, Represents the complete flow-rainfall time series data, including flood style information, Represents the random flood event data, Represents the non-flood period data.

[0041] 4. Data standardization and preparation: Standardize the flow and rainfall data to prepare the data for subsequent training of the multi-layer long short-term memory network.

[0042] 5. Construction and training of the LSTM network: Construct a multi-layer LSTM network and train it using the standardized data to capture the time dependence and complex dynamics of flood events. The multi-layer LSTM network model architecture includes multiple LSTM and Dropout layers to prevent overfitting. Use binary cross-entropy as the loss function and the Adam optimizer for training. The evaluation metric is binary classification accuracy.

[0043] 6. Model prediction and threshold setting: Use the trained multi-layer LSTM model to predict new data to obtain the probability of flood events occurring. By setting an appropriate threshold, convert the predicted probability into a binary classification result to determine whether a flood event has occurred. Finally, obtain a set of multiple non-overlapping time intervals. For example, it can be expressed as E={[t1,t2],[t3,t4],...,[tᵢ,tⱼ]}. Each interval [tᵢ,tⱼ] corresponds to a specific flood event, where tᵢ and tⱼ represent the start time and end time of the flood event respectively.

[0044] S102: Based on the flow-rainfall time series samples, train the multi-layer long short-term memory network using the target loss function to obtain a flood prediction model.

[0045] In this embodiment, the multi-layer long short-term memory network includes: The first layer of LSTM, the first layer of Dropout, the second layer of LSTM, the second layer of Dropout, the third layer of LSTM, the third layer of Dropout, and the fully connected layer.

[0046] The first layer of LSTM is used to receive the input time series data and obtain the first sequence output. The first layer of Dropout is used to randomly discard neurons from the first sequence output to obtain the first input sequence.

[0047] The second layer of LSTM is used to obtain the second sequence output based on the first input sequence. The second layer of Dropout is used to randomly discard neurons from the second sequence output to obtain the second input sequence.

[0048] The third - layer LSTM is used to obtain and output the hidden state at the last time step based on the second input sequence. The third - layer Dropout is used to randomly discard neurons from the hidden state to obtain the target output.

[0049] The fully - connected layer is used to calculate and output the flood event prediction probability based on the target output.

[0050] In this embodiment, the target loss function is:

[0051] where, is the target loss function, M is the number of samples included in a single training batch, is the true label, is the flood event prediction probability.

[0052] In this embodiment, the multi - layer long short - term memory network is a deep - learning model based on LSTM cells. The multi - layer long short - term memory network enhances the ability to capture long - distance dependencies and complex dynamic features in time series by stacking multiple LSTM layers, and is suitable for time - series modeling of flood processes. The target loss function is a function used to measure the difference between the model's predicted flood event probability and the true label. The target loss function can guide the optimization of model parameters and belongs to the loss function for binary - classification problems. The input time - series data can include flow - rainfall time - series samples, flow - rainfall time - series data to be partitioned, etc.

[0053] Exemplarily, the model architecture of the multi - layer long short - term memory network can include: The first - layer LSTM: The input sequence has a length of L, a feature dimension of 2, 100 hidden units, and returns the output of the entire sequence.

[0054] For each time step t from 1 to L, calculate the state of the LSTM cell:

[0055] where, represents the input feature vector at the k - th time step, represents the hidden state at the previous time step, represents the sigmoid activation function, represents the hyperbolic tangent activation function, represents the element - wise product. W, U, and b represent the input weight matrix, recurrent weight matrix, and bias vector of the network, respectively.

[0056] The first - layer Dropout layer: Add a Dropout layer with a dropout rate of 0.2, which can randomly discard 20% of the neurons to prevent overfitting.

[0057] Second - layer LSTM: Similar to the first layer, it has 100 hidden units and returns the output of the entire sequence.

[0058] Second - layer Dropout layer: Add a Dropout layer again, with a dropout rate of 0.2.

[0059] Third - layer LSTM: It has 100 hidden units, does not return the sequence output, and only outputs the hidden state of the last time step. 。

[0060] Third - layer Dropout layer: The dropout rate is 0.2.

[0061] Output layer: That is, a fully - connected layer, using the sigmoid activation function to output the final flood event prediction probability. The flood event prediction probability function is:

[0062] Where, represents the weight matrix of the output layer, represents the bias vector of the output layer.

[0063] Exemplarily, the model compilation process can use the Adam optimizer with a learning rate set to 0.01. Using the training data X and the target result binary label y, train for 200 epochs, with a batch size of 64 and a validation set ratio of 20%. In each epoch, update the model parameters to minimize the loss.

[0064] S103: Use the flood prediction model to perform flood event division on the flow - rainfall time - series data to be divided, and obtain the flood event division result.

[0065] In this embodiment, using the flood prediction model to perform flood event division on the flow - rainfall time - series data to be divided, and obtaining the flood event division result includes: Use the flood prediction model to calculate the flood event prediction probability of the flow - rainfall time - series data to be divided.

[0066] Based on the comparison result between the flood event prediction probability and the event division threshold, determine the target flood event and the target flood event interval corresponding to the target flood event.

[0067] Take all the target flood events and the target flood event intervals corresponding to the target flood events as the flood event division result.

[0068] In this embodiment, the flood prediction model is a multi-layer LSTM network trained with historical flood data, which can capture the temporal dependence and complex dynamic characteristics of the flow-rainfall time series and output the probability of flood events occurring. The flow-rainfall time series refers to the target data for which flood event sessions need to be divided, including the flow and rainfall sequences with time alignment, and the format is the same as the training data. The flood event prediction probability refers to the probability output of the flood prediction model for whether each time step belongs to a flood event, reflecting the possibility of the time step being in a flood state. The session division threshold refers to the critical value used to convert the prediction probability into a binary classification result.

[0069] The target flood event refers to the continuous data with a flood event prediction probability ≥ the session division threshold, representing a complete flood event. The target flood session interval refers to the time range corresponding to each target flood event.

[0070] In this embodiment, the target flood event is determined based on the comparison result between the flood event prediction probability and the session division threshold, including: If the flood event prediction probability is greater than or equal to the session division threshold, it is determined as the target flood event.

[0071] If the flood event prediction probability is less than the session division threshold, it is determined as a non-target flood event.

[0072] Exemplarily, the trained flood prediction model is used to predict new flow and rainfall data to generate the flood event prediction probability. The session division threshold is set to 0.5, and the flood event prediction probability is converted into a binary classification result. If it is higher than the session division threshold, it is determined as the target flood event (label 1), and if it is lower than the session division threshold, it is determined as the non-target flood event (label 0), and finally the occurrence of the flood event is determined. The flood event discrimination function is:

[0073] where, represents the label value corresponding to the flood data.

[0074] By obtaining the start and end times of each interval, a set of time intervals that do not overlap with each other can be obtained, and each interval corresponds to a specific flood event.

[0075] It can be concluded from the above that, on the one hand, for complex flood scenarios, this embodiment uses a multi-layer LSTM network to model the long-term dependence relationship of the flow-rainfall sequence, avoiding the mechanical segmentation defect of a single threshold, and can improve the accuracy of flood boundary division.

[0076] On the other hand, in this embodiment, by expanding flood samples, the model bias caused by scarcity is alleviated, enabling the model to fully learn the transition characteristics between flood peaks and base flows, and enhancing the generalization ability of the model. The hierarchical feature extraction ability of the multi-layer LSTM can adapt to the hydrological differences in different basins, accurately divide the complete boundary of floods, and combined with the target loss function, can improve the accuracy and robustness of model training.

[0077] In summary, through the organic combination of data quality optimization, deep model innovation, and hydrological logic constraints, this embodiment realizes the precision, automation, and scenario generalization of flood event division, providing efficient and reliable technical support for flood feature analysis, forecast model calibration, etc.

[0078] In an embodiment of the present disclosure, data augmentation is performed on flood sequence data to obtain target flood sequence data, including: Determine the first flood peak and the second flood peak based on all flood peak data of the flood sequence data.

[0079] Obtain unimodal flood sequence data based on the flood sequence data.

[0080] Determine the unimodal adjustment constraint based on the size relationship between the peak of the unimodal flood sequence data, the first flood peak, and the second flood peak.

[0081] Generate the first amplitude scaling factor based on the unimodal adjustment constraint.

[0082] Adjust the unimodal flood sequence data based on the first time scaling factor and the first amplitude scaling factor, and add the adjusted unimodal flood sequence data to the target flood sequence data. The first time scaling factor is a preset time adjustment parameter.

[0083] In this embodiment, the first flood peak and the second flood peak are flood peaks of different recurrence periods determined by frequency analysis based on historical flood data, and are used to constrain the rationality of data augmentation. For example, the first flood peak can correspond to the flood peak of once in a hundred years (recurrence period of 100 years, denoted as p = 1%), representing the upper limit of extreme flood intensity. The second flood peak can correspond to the flood peak of once in two years (recurrence period of 2 years, p = 50%), representing the lower limit of conventional flood intensity.

[0084] The unimodal adjustment constraint refers to setting the adjustment range of the first amplitude scaling factor according to the size relationship between the peak of the unimodal flood, the first flood peak, and the second flood peak, ensuring that the generated data conforms to hydrological laws. The first time scaling factor is a preset time axis scaling factor, used to simulate small changes in the flood occurrence speed. The first amplitude scaling factor is a parameter used to adjust the flow intensity of the unimodal flood sequence data, and can generate flood samples of different intensities by slightly scaling the original flood flow.

[0085] Exemplarily, this embodiment can collect the flow observation data of the target basin over the years and screen out the peak flows of all flood events in history. This embodiment adopts the Pearson-III frequency analysis method to statistically fit the historical peak flows, establish the relationship curve between the flow and the recurrence period, and extract the upper limit of the extreme intensity of the flood peak and the lower limit of the conventional intensity of the flood peak as the first flood peak and the second flood peak respectively, which are used to dynamically limit the amplitude scaling range to ensure that the flood intensity generated during data augmentation conforms to the natural laws of the basin and avoid the appearance of unrealistic extreme values.

[0086] This embodiment extracts the events containing only a single flood peak from the historical flood data and excludes the multi-peak flood events. Calculate the peak flow of each single-peak flood, and divide the single-peak floods into three categories according to their size relationship with the first flood peak and the second flood peak: Extreme single-peak flood: The peak exceeds the first flood peak.

[0087] Conventional single-peak flood: The peak is between the first flood peak and the second flood peak.

[0088] Low-intensity single-peak flood: The peak is lower than the first flood peak.

[0089] This embodiment sets different single-peak adjustment constraints for different single-peak flood types. For example, for extreme single-peak floods, it is determined that the single-peak adjustment constraint is: the interval of the first amplitude scaling factor a is [0.95, 1.05]. For low-intensity single-peak floods, it is determined that the single-peak adjustment constraint is: the interval of the first amplitude scaling factor is [0.90, 1.10].

[0090] This embodiment presets the first time scaling factor s (the range of s is [0.95, 1.05]), for example, randomly selects 0.98 and 1.02, and calculates the adjusted time axis . Interpolate the original flow data to obtain a new data sequence :

[0091] The start time and end time of the flood event interval change with the first time scaling factor.

[0092] Randomly select the first amplitude scaling factor within the single-peak adjustment constraint range. Calculate the flow data after amplitude adjustment :

[0093] The start time and end time of the flood event interval remain unchanged because the amplitude scaling does not affect the division of the event interval.

[0094] In this embodiment, the flood peak values with different recurrence periods are determined through frequency analysis, and reasonable amplitude and time scaling factors are generated accordingly to adjust the single-peak flood sequence data. This embodiment can generate diverse flood samples that conform to hydrological laws, alleviate data scarcity, enhance the adaptability of the model to floods with different intensities and durations, and improve the accuracy of flood event classification in complex scenarios.

[0095] In one embodiment of the present disclosure, when performing data augmentation on flood sequence data to obtain target flood sequence data, it further includes: Obtaining multi-peak flood sequence data based on the flood sequence data.

[0096] Extracting the valley points in the multi-peak flood sequence data, and determining the first adjustment constraint for the multi-peak based on the valley points and the base flow threshold.

[0097] Determining the second adjustment constraint for the multi-peak based on the magnitude relationship between the maximum peak value of the multi-peak flood sequence data and the first flood peak value and the second flood peak value.

[0098] Generating a second amplitude scaling factor based on the first adjustment constraint for the multi-peak and the second adjustment constraint for the multi-peak.

[0099] Adjusting the multi-peak flood sequence data based on the second time scaling factor and the second amplitude scaling factor, and adding the adjusted multi-peak flood sequence data to the target flood sequence data. The second time scaling factor is a preset time adjustment parameter.

[0100] In this embodiment, the multi-peak flood sequence data refers to the flood event data containing two or more flood peaks, which is manifested as multiple peak values existing in the flow hydrograph, and the valley value between the peaks does not fall back to the base flow or there are obvious flow fluctuations. The valley point refers to the lowest flow point between adjacent flood peaks in the multi-peak flood event, that is, the lowest point of the recession between two rising water processes. The base flow threshold refers to the stable minimum flow value in the basin during the non-flood period, which is used to judge whether the valley value belongs to an effective recession process. When the valley point is greater than or equal to the base flow threshold, it can be regarded as the fluctuation of the same flood.

[0101] The first adjustment constraint for the multi-peak is based on the relationship between the valley point and the base flow threshold, and sets the lower limit constraint of the valley value after amplitude scaling to ensure that the valley value after adjustment is not lower than the base flow threshold and maintain the physical rationality of the flood event. The second adjustment constraint for the multi-peak is a dynamically limited amplitude scaling range according to the magnitude relationship between the maximum peak value of the multi-peak flood and the first flood peak value and the second flood peak value, to avoid generating flood samples with supernatural intensities. The second time scaling factor is used to globally adjust the time axis of the multi-peak flood, and the second amplitude scaling factor is generated according to the two adjustment constraints, which is used to scale the flow intensity of the multi-peak flood proportionally to keep the relative amplitude of each peak unchanged.

[0102] Exemplarily, in this embodiment, the flood sequence data can be traversed to extract the valley points (local minimum of flow) between adjacent flood peaks, and the flow values and time points corresponding to the valley points are recorded. Check whether the flow at the valley point is higher than the base flow threshold. If it is less than the base flow threshold, it is determined as an internal fluctuation of the same flood and the flood events are not segmented.

[0103] For the obtained multi-peak flood sequence data, determine the multi-peak first adjustment constraint and the multi-peak second adjustment constraint. Multi-peak first adjustment constraint: Ensure that the valley point value after data adjustment needs to be greater than or equal to the base flow threshold, so as to obtain the lower limit value of the second amplitude scaling factor. Multi-peak second adjustment constraint: According to the size relationship between the maximum peak and the first flood peak and the second flood peak, set different ranges of the second amplitude scaling factor. Finally, determine the range of the second amplitude scaling factor that simultaneously satisfies the multi-peak first adjustment constraint and the multi-peak second adjustment constraint. Generate the second amplitude scaling factor based on this range.

[0104] In this embodiment, a second time scaling factor is preset to adjust the time axis of the multi-peak flood sequence data, and interpolation is performed on the adjusted data sequence to obtain a new multi-peak flood sequence. Randomly select a second amplitude scaling factor within the range of the multi-peak first adjustment constraint and the multi-peak second adjustment constraint, calculate the flow data after amplitude adjustment of the multi-peak flood sequence data according to the second amplitude scaling factor, and obtain the extended multi-peak flood sequence data.

[0105] In this embodiment, for multi-peak flood data, the rationality of the adjusted valley value is ensured by the constraint between the valley point and the base flow threshold, and the amplitude scaling range is limited by combining the relationship between the maximum peak and the design flood peak, so as to generate multi-peak samples that conform to the hydrological law. This embodiment can effectively expand diverse and reasonable multi-peak data, alleviate the scarcity problem, and enhance the adaptability and division accuracy of the model to multi-peak scenarios.

[0106] Corresponding to the above-mentioned embodiment of the flood event division method based on deep learning, Figure 6 is the structural block diagram of a flood event division device based on deep learning provided by an embodiment of the present disclosure. For the sake of convenience of description, only the parts related to the embodiments of the present disclosure are shown. Refer to Figure 6 The flood event division device 20 based on deep learning includes: a data processing module 21, a model training module 22, and a model event division module 23.

[0107] Among them, the data processing module 21 is used to divide the historical flood data into flood sequence data and non-flood sequence data. Perform data augmentation on the flood sequence data to obtain target flood sequence data. Concatenate the target flood sequence data and the non-flood sequence data to obtain a flow-rainfall time series sample.

[0108] The model training module 22 is used to train a multi-layer long short-term memory network based on the flow-rainfall time series samples by using the target loss function to obtain a flood prediction model.

[0109] The model flood event division module 23 is used to divide the flow-rainfall time series data to be divided by using the flood prediction model to obtain a flood event division result.

[0110] In an embodiment of the present disclosure, the data processing module 21 is specifically used to divide the time interval corresponding to the historical flood data into a flood event interval and a non-flood period. The flood event interval is the time interval when a flood event occurs. The non-flood period is the time interval when no flood event occurs.

[0111] The flood data corresponding to the flood event interval is used as flood sequence data.

[0112] The base flow data corresponding to the non-flood period is used as non-flood sequence data.

[0113] In an embodiment of the present disclosure, the data processing module 21 is specifically further used to determine a first flood peak and a second flood peak based on all the flood peak data of the flood sequence data.

[0114] Unimodal flood sequence data is obtained based on the flood sequence data.

[0115] A unimodal adjustment constraint is determined based on the magnitude relationship between the peak value of the unimodal flood sequence data, the first flood peak, and the second flood peak.

[0116] A first amplitude scaling factor is generated based on the unimodal adjustment constraint.

[0117] The unimodal flood sequence data is adjusted based on the first time scaling factor and the first amplitude scaling factor, and the adjusted unimodal flood sequence data is added to the target flood sequence data. The first time scaling factor is a preset time adjustment parameter.

[0118] In an embodiment of the present disclosure, the data processing module 21 is specifically further used to obtain multi-peak flood sequence data based on the flood sequence data.

[0119] The valley points in the multi-peak flood sequence data are extracted, and a multi-peak first adjustment constraint is determined based on the valley points and the base flow threshold.

[0120] A multi-peak second adjustment constraint is determined based on the magnitude relationship between the maximum peak value of the multi-peak flood sequence data, the first flood peak, and the second flood peak.

[0121] A second amplitude scaling factor is generated based on the multi-peak first adjustment constraint and the multi-peak second adjustment constraint.

[0122] Adjust the multi-peak flood sequence data based on the second time scaling factor and the second amplitude scaling factor, and add the adjusted multi-peak flood sequence data to the target flood sequence data. The second time scaling factor is a preset time adjustment parameter.

[0123] In an embodiment of the present disclosure, the model training module 22 is specifically configured that the multi-layer long short-term memory network includes: The first layer of LSTM, the first layer of Dropout, the second layer of LSTM, the second layer of Dropout, the third layer of LSTM, the third layer of Dropout, and the fully connected layer.

[0124] The first layer of LSTM is used to receive the input time series data and obtain the first sequence output. The first layer of Dropout is used to randomly discard neurons from the first sequence output to obtain the first input sequence.

[0125] The second layer of LSTM is used to obtain the second sequence output based on the first input sequence. The second layer of Dropout is used to randomly discard neurons from the second sequence output to obtain the second input sequence.

[0126] The third layer of LSTM is used to obtain and output the hidden state of the last time step based on the second input sequence. The third layer of Dropout is used to randomly discard neurons from the hidden state to obtain the target output.

[0127] The fully connected layer is used to calculate and output the flood event prediction probability based on the target output.

[0128] In an embodiment of the present disclosure, the model training module 22 is further specifically configured that the target loss function is:

[0129] Where, is the target loss function, M is the number of samples included in a single training batch, is the true label, is the flood event prediction probability.

[0130] In an embodiment of the present disclosure, the model flood event division module 23 is specifically configured to calculate the flood event prediction probability of the flow-rainfall time series data to be divided by using the flood prediction model.

[0131] Determine the target flood event and the target flood event corresponding target flood event interval based on the comparison result between the flood event prediction probability and the flood event division threshold.

[0132] Take all the target flood events and the target flood event corresponding target flood event intervals as the flood event division result.

[0133] SeeFigure 7 , Figure 7 is a schematic block diagram of an electronic device provided by an embodiment of the present disclosure. As Figure 7 shown, the electronic device 300 in this embodiment may include: one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The above-mentioned processors 301, input devices 302, output devices 303, and memories 304 communicate with each other through a communication bus 305. The memory 304 is used to store computer programs, and the computer programs include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. Among them, the processor 301 is configured to call the program instructions to execute the functions of each module in the above-mentioned device embodiments, for example Figure 6 shown, the functions of the data processing module 21, the model training module 22, and the model session division module 23.

[0134] It should be understood that in the embodiments of the present disclosure, the so-called processor 301 may be a central processing unit (CPU), and this processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or this processor may also be any conventional processor, etc.

[0135] The input device 302 may include a touchpad, a fingerprint acquisition sensor (for acquiring the fingerprint information and the fingerprint direction information of the user), a microphone, etc., and the output device 303 may include a display (such as an LCD), a speaker, etc.

[0136] The memory 304 may include a read-only memory and a random access memory, and provide instructions and data to the processor 301. A part of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store information about the device type.

[0137] In specific implementation, the processors 301, input devices 302, and output devices 303 described in the embodiments of the present disclosure may execute the implementation manners described in the first and second embodiments of the flood session division method based on deep learning provided by the embodiments of the present disclosure, and may also execute the implementation manner of the electronic device 300 described in the embodiments of the present disclosure, which will not be elaborated here.

[0138] In another embodiment of the present disclosure, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, all or part of the processes in the method of the above embodiments are implemented. It can also be completed by instructing relevant hardware through the computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0139] The computer-readable storage medium can be the internal storage unit of the electronic device in any of the foregoing embodiments, such as the hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device. Further, the computer-readable storage medium can also include both the internal storage unit and the external storage device of the electronic device. The computer-readable storage medium is used to store the computer program and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store the data that has been output or will be output.

[0140] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present disclosure.

[0141] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described electronic devices and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0142] In several embodiments provided by this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces or units, and can also be electrical, mechanical or other forms of connection.

[0143] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiments of the present disclosure.

[0144] In addition, each functional unit in various embodiments of the present disclosure can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0145] The above is only the specific implementation manner of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present disclosure can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered by the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A method for flood event division based on deep learning, characterized in that Including: Dividing historical flood data into flood sequence data and non-flood sequence data; Performing data augmentation on the flood sequence data to obtain target flood sequence data; Concatenating the target flood sequence data and the non-flood sequence data according to the time series to obtain a flow-rainfall time series sample; Based on the flow-rainfall time series sample, training a multi-layer long short-term memory network using a target loss function to obtain a flood prediction model; Using the flood prediction model to perform flood event division on the flow-rainfall time series data to be divided, and obtaining a flood event division result.

2. The method for flood event division based on deep learning according to claim 1, wherein The dividing the historical flood data into flood sequence data and non-flood sequence data includes: Dividing the time interval corresponding to the historical flood data into flood event intervals and non-flood periods; the flood event interval is the time interval when flood events occur; the non-flood period is the time interval when no flood events occur; Taking the flood data corresponding to the flood event interval as flood sequence data; Taking the base flow data corresponding to the non-flood period as non-flood sequence data.

3. The method for flood event division based on deep learning according to claim 1, wherein The performing data augmentation on the flood sequence data to obtain target flood sequence data includes: Determining a first flood peak and a second flood peak based on all flood peak data of the flood sequence data; Obtaining single-peak flood sequence data based on the flood sequence data; Determining a single-peak adjustment constraint based on the magnitude relationship between the peak value of the single-peak flood sequence data, the first flood peak, and the second flood peak; Generating a first amplitude scaling factor based on the single-peak adjustment constraint; Adjusting the single-peak flood sequence data based on a first time scaling factor and the first amplitude scaling factor, and adding the adjusted single-peak flood sequence data to the target flood sequence data; the first time scaling factor is a preset time adjustment parameter.

4. The method for flood event division based on deep learning according to claim 3, wherein, The performing data augmentation on the flood sequence data to obtain target flood sequence data further includes: Obtaining multi-peak flood sequence data based on the flood sequence data; Extracting valley points from the multi-peak flood sequence data, and determining a multi-peak first adjustment constraint based on the valley points and a base flow threshold; Determining a multi-peak second adjustment constraint based on the magnitude relationship between the maximum peak value of the multi-peak flood sequence data, the first flood peak, and the second flood peak; Generating a second amplitude scaling factor based on the multi-peak first adjustment constraint and the multi-peak second adjustment constraint; Adjusting the multi-peak flood sequence data based on a second time scaling factor and the second amplitude scaling factor, and adding the adjusted multi-peak flood sequence data to the target flood sequence data; the second time scaling factor is a preset time adjustment parameter.

5. The method for flood event division based on deep learning according to claim 1, characterized in that, The multi-layer long short-term memory network includes: The first layer of LSTM, the first layer of Dropout, the second layer of LSTM, the second layer of Dropout, the third layer of LSTM, the third layer of Dropout, and a fully connected layer; The first layer of LSTM is used to receive input time series data and obtain a first sequence output, and the first layer of Dropout is used to randomly discard neurons from the first sequence output to obtain a first input sequence; The second-layer LSTM is used to obtain a second sequence output based on the first input sequence, and the second-layer Dropout is used to randomly discard neurons in the second sequence output to obtain a second input sequence; The third-layer LSTM is used to obtain and output the hidden state at the last time step based on the second input sequence; the third-layer Dropout is used to randomly discard neurons in the hidden state to obtain a target output; The fully connected layer is used to calculate and output the flood event prediction probability based on the target output.

6. The method for flood event division based on deep learning according to claim 1, wherein The target loss function is: Among them, is the target loss function, M is the number of samples in a single training batch, is the true label, is the predicted probability of flood events.

7. The method for flood event division based on deep learning according to claim 1, characterized in that, Using the flood prediction model to perform flood event division on the flow-rainfall time series data to be divided, and obtaining the flood event division result, including: Calculating the flood event prediction probability of the flow-rainfall time series data to be divided by using the flood prediction model; Determining the target flood event and the target flood event interval corresponding to the target flood event based on the comparison result between the flood event prediction probability and the event division threshold; Taking all the target flood events and the target flood event intervals corresponding to the target flood events as the flood event division result.

8. A device for flood event division based on deep learning, characterized in that, Including: A data processing module, configured to divide historical flood data into flood sequence data and non-flood sequence data; Performing data augmentation on the flood sequence data to obtain target flood sequence data; splicing the target flood sequence data and the non-flood sequence data in a time series to obtain a flow-rainfall time series sample; A model training module, configured to train a multi-layer long short-term memory network by using a target loss function based on the flow-rainfall time series sample to obtain a flood prediction model; A model event division module, configured to perform flood event division on the flow-rainfall time series data to be divided by using the flood prediction model to obtain a flood event division result.

9. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images