Groundwater level change data classification method and system based on recurrent neural network

Abstract

The application discloses a groundwater level change data classification method and system based on a recurrent neural network. The method comprises the following steps: collecting groundwater level change data and atmospheric pressure change data and performing data preprocessing to obtain two-dimensional time series data; training a recurrent neural network model based on the two-dimensional time series data to obtain an optimal recurrent neural network model; and classifying the groundwater level change data based on the optimal recurrent neural network model to obtain a classification result. The embodiment of the application can classify and process the groundwater level change data in a short time period and improve the classification accuracy of the groundwater level change data. The application can be widely applied to the technical field of groundwater level change classification.

Description

Technical Field

[0001] This application relates to the field of groundwater level change classification technology, and in particular to a groundwater level change data classification method and system based on recurrent neural networks. Background Technology

[0002] Changes in groundwater levels are related to factors such as precipitation, seasonal snowmelt, and seismic waves. By selecting earthquake-related groundwater level fluctuation data from among numerous relevant factors, and further analyzing seismic wave information, it is of great significance for understanding the response mechanism between earthquakes and the Earth's surface.

[0003] There are two main methods for identifying earthquake-related groundwater level changes from high-sampling-rate groundwater level data. The first is mathematical modeling. This method involves mathematically modeling the factors affecting groundwater levels, calculating the theoretical changes caused by each factor, and then subtracting these factors from the observed data one by one to obtain unexplained water level changes. These unexplained changes are then compared with earthquake records over time to determine the water level changes at the time of the earthquake. However, this method requires extensive surveys and experiments in the observation area to obtain the parameters needed for mathematical modeling, resulting in a long modeling cycle and high manpower consumption. For unknown and unmeasurable parameters, the mathematical model can only be adjusted through assumptions or guesses, which can introduce significant errors. The second method is filtering. This method utilizes the unique periodicity of water level changes caused by different phenomena. By filtering groundwater level observation data from low to high, high-frequency, short-period groundwater level fluctuations are retained. These are then compared with the earthquake time to identify earthquake-related groundwater level changes. However, this method requires multiple filtering calculations of the measured water level, and each calculation will inevitably produce errors. The accumulation of errors after multiple calculations will be superimposed on the observed data, ultimately leading to a large prediction error.

[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0005] The main objective of this application is to propose a groundwater level change data classification method and system based on recurrent neural networks, which can classify groundwater level change data in a shorter time period and improve the classification accuracy of groundwater level change data.

[0006] To achieve the above objectives, one aspect of this application proposes a groundwater level change data classification method based on a recurrent neural network, the method comprising:

[0007] Data on groundwater level changes and atmospheric pressure changes were collected and preprocessed to obtain two-dimensional time series data.

[0008] The recurrent neural network model is trained based on the two-dimensional time series data to obtain the optimal recurrent neural network model;

[0009] The groundwater level change data were classified based on the optimal recurrent neural network model, and the classification results were obtained.

[0010] In some embodiments, the process of collecting groundwater level change data and atmospheric pressure change data and performing data preprocessing to obtain two-dimensional time series data includes:

[0011] Collect data on groundwater level changes and atmospheric pressure changes;

[0012] Set the data periodic sequence;

[0013] The groundwater level change data and the atmospheric pressure change data are processed according to the data period sequence to obtain groundwater level time change data and atmospheric pressure time change data.

[0014] The groundwater level time variation data and the atmospheric pressure time variation data are integrated to obtain the two-dimensional time series data.

[0015] In some embodiments, training the recurrent neural network model based on the two-dimensional time series data to obtain the optimal recurrent neural network model includes:

[0016] Construct a recurrent neural network model;

[0017] The two-dimensional time series data is input into the recurrent neural network model for training, and the weights between the two-dimensional time series data and each network layer of the recurrent neural network model are obtained through feedforward calculation.

[0018] The loss value of the recurrent neural network model is obtained by performing loss calculation on the recurrent neural network model during the training process using the binary cross-entropy loss function.

[0019] Based on the loss value and the weights, the hyperparameters of the recurrent neural network model are iteratively optimized using a Bayesian adaptive method to obtain the optimal recurrent neural network model.

[0020] In some embodiments, the recurrent neural network model includes an input layer, a hidden layer, and an output layer, wherein the output of the input layer is connected to the input of the hidden layer, and the output of the hidden layer is connected to the input of the output layer, wherein:

[0021] The input layer comprises two neurons;

[0022] The hidden layer comprises several neurons and uses the sigmoid function as the activation function.

[0023] The output layer consists of one neuron.

[0024] In some embodiments, the step of inputting the two-dimensional time series data into the recurrent neural network model for training, and obtaining the weights between the preceding and following two-dimensional time series data and between each network layer of the recurrent neural network model through feedforward calculation, includes:

[0025] The two-dimensional time series data is input into the recurrent neural network model;

[0026] Based on the input layer of the recurrent neural network model, the two-dimensional time series data is input into the hidden layer;

[0027] Based on the hidden layer of the recurrent neural network model, feedforward calculation is performed on the two-dimensional time series data to obtain the weights between the two-dimensional time series data before and after the label to be assigned and between each network layer of the recurrent neural network model.

[0028] Based on the output layer of the recurrent neural network model, the weights between the two-dimensional time series data before and after the data to be labeled and between each network layer of the recurrent neural network model are labeled to obtain the preliminary weights between the two-dimensional time series data before and after the data and between each network layer of the recurrent neural network model.

[0029] Repeat the above steps of inputting the two-dimensional time series data into the recurrent neural network model for training until the loss function of the recurrent neural network model reaches its minimum value, and output the weights between the two-dimensional time series data and between each network layer of the recurrent neural network model.

[0030] In some embodiments, the weight calculation expression for the hidden layer of the recurrent neural network model is as follows:

[0031] h t =sig(W ih X t +b ih +W hh h t-1 +b hh )

[0032] In the above formula, h t h represents the weights of the two-dimensional time series vector data output at the current time. t-1 W represents the weights of the two-dimensional time series vector data output at the previous time step, sig(·) represents the sigmoid function, and W ih W represents the weight matrix between the input layer and the hidden layer.hh b represents the weight matrix connecting the states of two sequences in the hidden layer at different time steps. ih b represents the bias vector of the input unit. hh X represents the bias vector of the hidden unit. t This indicates the input data.

[0033] In some embodiments, the expression for the output layer of the recurrent neural network model is:

[0034] y t =sig(W ho h t +b o )

[0035] In the above formula, y t W represents the output of the recurrent neural network model. ho b represents the weight matrix between the hidden layer and the output layer. o h represents the bias vector in the output layer. t This represents the output of the hidden layer in a recurrent neural network model, and sig(·) represents the sigmoid function.

[0036] In some embodiments, the expression for the binary cross-entropy loss function is:

[0037]

[0038] In the above formula, BCELoss(·) represents the binary cross-entropy loss function, n represents the number of samples, and y i Let p(y) represent the true label of the i-th sample. i =1) represents the predicted probability of the i-th sample.

[0039] In some embodiments, the step of iteratively optimizing the hyperparameters of the recurrent neural network model using a Bayesian adaptive method based on the loss value and the weights to obtain an optimal recurrent neural network model includes:

[0040] Set the hyperparameter adjustment range of the recurrent neural network model;

[0041] The hyperparameters of each group of the recurrent neural network models within the given hyperparameter adjustment range are tested using the Bayesian adaptive method.

[0042] Based on the loss value and the weights, compare the loss value of the recurrent neural network model obtained by training with the hyperparameters of each group of the recurrent neural network models, select the hyperparameters of the recurrent neural network model with the smallest loss value, and output the optimal recurrent neural network model.

[0043] To achieve the above objectives, another aspect of this application proposes a groundwater level change data classification system based on a recurrent neural network, the system comprising:

[0044] The first module is used to collect groundwater level change data and atmospheric pressure change data and perform data preprocessing to obtain two-dimensional time series data.

[0045] The second module is used to train the recurrent neural network model based on the two-dimensional time series data to obtain the optimal recurrent neural network model.

[0046] The third module is used to classify groundwater level change data based on the optimal recurrent neural network model and obtain the classification results.

[0047] The embodiments of this application include at least the following beneficial effects: This application provides a method and system for classifying groundwater level change data based on recurrent neural networks. This scheme collects groundwater level change data and atmospheric pressure change data and performs data preprocessing to obtain two-dimensional time series data. Atmospheric pressure change data is used as a linkage reference for groundwater level to determine whether the water level change is related to earthquakes. A recurrent neural network model is constructed and trained using two-dimensional time series data. It can flexibly handle input sequences of different lengths and capture the dependencies in the sequences. The trained RNN model can automatically distinguish groundwater level changes caused by earthquakes without performing multiple filtering calculations. It can classify groundwater level change data within a short time period and improve the classification accuracy of groundwater level change data. Attached Figure Description

[0048] Figure 1 This is a flowchart of the groundwater level change data classification method based on recurrent neural networks provided in the embodiments of this application;

[0049] Figure 2 This is a schematic diagram of the structure of the recurrent neural network model provided in the embodiments of this application;

[0050] Figure 3 This is a schematic diagram of the structure of the groundwater level change data classification system based on recurrent neural networks provided in the embodiments of this application. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of systems and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0052] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0053] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0055] Before providing a detailed description of the embodiments of this application, the factors contributing to the changes in groundwater level involved in the embodiments of this application will be explained as follows.

[0056] Groundwater level changes are related to factors such as precipitation, seasonal snowmelt, and seismic waves. Screening earthquake-related groundwater level fluctuation data from numerous relevant factors for further analysis of seismic wave information is crucial for understanding the response mechanism between earthquakes and the Earth's surface. Earthquake-induced specific water level changes identified from groundwater level observation data can be categorized into long-term and short-term changes. Long-term changes generally refer to changes in water level that persist for more than a week after an earthquake causes a rise or fall. Short-term changes can be further divided into short-term sustained changes and seismic changes. Short-term sustained changes generally refer to water levels returning to the pre-earthquake average level within one day after an earthquake-induced rise or fall. Seismic changes refer to short-term fluctuations in water level caused by an earthquake. Seismic changes are a key phenomenon in studying how external vibrations such as earthquakes alter groundwater levels. Existing research suggests that the mechanism by which vibration causes changes in groundwater level is usually due to static deformation caused by earthquakes, such as surface uplift, which leads to the expansion of underground space and a drop in groundwater level. Vibration can also cause the expansion of fissures in the underground aquifer where groundwater is located, resulting in a drop in groundwater level. Vibration may also weaken or break the skeleton composed of rock particles in the underground aquifer, causing local compaction, which increases the ground pressure on the groundwater and leads to a rise in groundwater level.

[0057] Research on earthquake-induced groundwater level vibrations is still in its early stages. There are two main reasons for this: First, the hardware requirements for groundwater level observation are high. For a long time, the data sampling rate for groundwater level observation has been low, mostly around once per hour, with even higher precision methods only sampling once per minute. Second, the data information is too complex. As high-sampling-rate groundwater level data is collected, it becomes increasingly difficult to distinguish earthquake-related groundwater level vibrations from the groundwater level itself, especially in deep groundwater exceeding 100 meters, which is highly sensitive to changes in the underground environment. Groundwater level fluctuations include information from precipitation, surface vibrations, and air pressure changes.

[0058] However, existing research often uses calculations (mathematical models or filtering) on ​​observational data to enhance the representation of groundwater levels during earthquakes, and then artificially determines whether the short-term groundwater level changes were caused by the earthquake by comparing the data with the earthquake's time frame. This existing research method has the following drawbacks:

[0059] First, water level data is measured after multiple calculations, and each calculation inevitably introduces errors. These errors may stem from the mathematical model or the calculation itself (such as the number of bits used in floating-point calculations of the source water level signal). These errors accumulate after multiple calculations, making it impossible to distinguish water level signals from noise caused by small-scale vibrations, thus making it difficult to differentiate water level changes caused by more small-scale earthquakes. Furthermore, the accumulated errors easily distort the water level data, making it impossible to accurately determine the start time of water level changes when comparing with earthquake times.

[0060] Secondly, manual screening is needed to confirm whether changes in groundwater levels are caused by earthquakes. This is primarily due to the randomness of earthquake occurrence times. Experienced personnel are required to compare the arrival time of seismic waves with the time of water level changes to determine if the changes are caused by earthquakes. This requirement cannot be avoided using traditional methods.

[0061] In view of this, this invention, based on deep learning technology, utilizes a recurrent neural network (RNN) to establish a model for analyzing highly sampled groundwater level data, automatically distinguishing between earthquake-related and non-earthquake-related groundwater level changes. A dataset is established using sampled groundwater level change data caused by various factors, and the RNN model is trained on this dataset. The trained RNN model can automatically identify groundwater level changes caused by earthquakes. The groundwater level change dataset used in this invention is source data from measurements taken from wells in the Oda area of ​​Kanagawa Prefecture, central Japan, since 2011, thus avoiding the possibility of data distortion. Experimental observations show that the RNN model trained using this dataset automatically identifies the type of groundwater level, overcoming the shortcomings of previous methods that relied heavily on manual input, and can identify more groundwater level changes caused by small-scale earthquakes.

[0062] Reference Figure 1 , Figure 1 A flowchart illustrating a groundwater level change data classification method based on a recurrent neural network, provided as an embodiment of the present invention, is shown below. Figure 1 The method includes the following steps:

[0063] S100: Collect groundwater level change data and atmospheric pressure change data and perform data preprocessing to obtain two-dimensional time series data;

[0064] It should be noted that in some embodiments, step S100 may include: S110, collecting groundwater level change data and atmospheric pressure change data; S120, setting a data periodic sequence; S130, dividing and processing the groundwater level change data and atmospheric pressure change data according to the data periodic sequence to obtain groundwater level time change data and atmospheric pressure time change data; S140, integrating the groundwater level time change data and atmospheric pressure time change data to obtain two-dimensional time series data.

[0065] In some specific embodiments, the present invention uses groundwater level data and atmospheric pressure data sampled during the same period in Oda area, Kanagawa Prefecture, central Japan, from 2011 to 2012 to establish a dataset. The data is sampled once per second, and a total of 10,000 data points are selected, of which 6,000 are at earthquake times and 4,000 are at non-earthquake times.

[0066] It should be noted that the dataset was constructed using groundwater level and atmospheric pressure data sampled concurrently in the Oda area of ​​Kanagawa Prefecture, central Japan, during 2011-2012. This data was sampled every second. Atmospheric pressure data is typically correlated with diurnal variations in groundwater levels, serving as a reference for determining whether groundwater level changes are earthquake-related. A total of 10,000 data points were selected, with 6,000 at earthquake times and 4,000 at non-earthquake times. A 20-second interval was used as an input sequence, meaning the dataset contains sequences of 300 (6000 / 20) earthquake times and 200 (4000 / 20) non-earthquake times.

[0067] S200. The recurrent neural network model is trained based on two-dimensional time series data to obtain the optimal recurrent neural network model;

[0068] It should be noted that in some embodiments, step S200 may include: S210, constructing a recurrent neural network model; S220, inputting the two-dimensional time series data into the recurrent neural network model for training, and obtaining the weights between the preceding and following two-dimensional time series data and between each network layer of the recurrent neural network model through feedforward calculation; S230, performing loss calculation processing on the recurrent neural network model during the training process using the binary cross-entropy loss function to obtain the loss value of the recurrent neural network model; S240, based on the loss value and the weights, iteratively optimizing the hyperparameters of the recurrent neural network model using the Bayesian adaptive method to obtain the optimal recurrent neural network model.

[0069] The recurrent neural network model includes an input layer, a hidden layer, and an output layer. The output of the input layer is connected to the input of the hidden layer, and the output of the hidden layer is connected to the input of the output layer. The input layer includes two neurons, and the hidden layer includes several neurons. In this embodiment, the number of neurons in the hidden layer is set to 16, and the sigmoid function is used as the activation function. The output layer includes one neuron.

[0070] In some specific embodiments, groundwater level data exhibits temporal continuity and continuous fluctuations, while RNNs can flexibly handle input sequences of varying lengths and capture dependencies within the sequences. Therefore, this embodiment of the invention utilizes an RNN model to couple groundwater level and atmospheric pressure data as two-dimensional time series data, enabling automatic classification of earthquake-related and non-earthquake-related groundwater level data.

[0071] Specifically, such as Figure 2As shown, the RNN model in this embodiment of the invention consists of three layers: an input layer, a hidden layer, and an output layer. The input layer contains two neurons and is used to acquire groundwater level data and air pressure data from the dataset. The second layer is the hidden layer, which contains M neurons (M is an adjustable parameter). The output of the hidden layer at each time step (second) is h. t ={h1,h2,….h M In each 20-second sequence, the hidden layers at adjacent time points are connected by weights W. ih The hidden layers are interconnected in a loop, where i represents the i-th time point in the time series. Therefore, the outputs of the hidden layers are {…,h}. t-1 ,h t ,h t+1 ,…} is a vector sequence. The output layer consists of a single neuron; an output of 1 indicates earthquake-related, and an output of 0 indicates non-earthquake-related.

[0072] Furthermore, the number of neurons M in the hidden layer was set to 16. 80% of the data was extracted from the dataset as training data to train the RNN model, and 20% as validation data. To optimize the hyperparameters, a Bayesian adaptive method was used. The accuracy of the trained RNN model reached between 96% and 98%.

[0073] It should be noted that the RNN model is created with 2 neurons in the input layer, 16 neurons in the hidden layer, and 1 neuron in the output layer. Every 20 input data points form a recurrent sequence. The input layer passes the received data to the next layer and converts the input measurement time series data into vector data for feature extraction in the next neural network layer (hidden layer). The hidden layer is the feature extraction layer; through repeated training on the training set data, it establishes the relationship between groundwater level features and earthquake-induced water level changes (training set), thereby classifying the input groundwater level. The output layer, the layer below the hidden layer, outputs the classification results of the hidden layer in the form of labels (non-earthquake-induced, earthquake-induced).

[0074] Furthermore, in some embodiments, step S220 may include: S221, inputting the two-dimensional time series data into the recurrent neural network model; S222, inputting the two-dimensional time series data into the hidden layer based on the input layer of the recurrent neural network model; S223, performing feedforward calculation on the two-dimensional time series data based on the hidden layer of the recurrent neural network model to obtain the weights between the two-dimensional time series data before and after the label to be assigned and ...

[0075] Specifically, the sequence data of the dataset is input into the RNN model for training. The input is a vector {…,X} over a sequence time t. t-1 ,X t ,X t+1 The sequence data pruned in step S1 is input into the input layer, with each sequence lasting 20 seconds. Therefore, the input for one sequence is {X1, X2, ..., X}. 20 Each X represents a vector consisting of groundwater level data and air pressure data. The input consists of a sequence of 300 earthquake times and a sequence of 200 non-earthquake times.

[0076] The hidden layer has 16 neurons and uses the sigmoid function as the activation function. The output of the hidden layer depends on the input data X at the current time t. t It also depends on the hidden layer result of the previous time step t-1 in the same sequence. The formula for calculating the hidden layer result is:

[0077] h t =sig(W ih X t +b ih +W hh h t-1 +b hh )

[0078] In the above formula, h t h represents the weights of the two-dimensional time series vector data output at the current time. t-1W represents the weights of the two-dimensional time series vector data output at the previous time step, sig(·) represents the sigmoid function, and W ih W represents the weight matrix between the input layer and the hidden layer. hh b represents the weight matrix connecting the states of two sequences in the hidden layer at different time steps. ih b represents the bias vector of the input unit. hh X represents the bias vector of the hidden unit. t This indicates the input data.

[0079] Where sig(·) is the sigmoid function, and its function expression is:

[0080] sig(x)=(1+e -x ) -1

[0081] The output layer has one neuron. If the input layer contains an earthquake sequence, the neuron's value is set to 1, meaning the target label is set to 1; if the input layer contains a non-earthquake sequence, the neuron's value is set to 0, meaning the target label is set to 0. The calculation formula for the output layer is:

[0082] y t =sig(W ho h t +b o )

[0083] In the above formula, y t W represents the output of the recurrent neural network model. ho b represents the weight matrix between the hidden layer and the output layer. o h represents the bias vector in the output layer. t This represents the output of the hidden layer in a recurrent neural network model, and sig(·) represents the sigmoid function.

[0084] The binary cross-entropy loss function is used to measure the loss between the model output and the target label. The W parameter between each layer is continuously adjusted based on the descent gradient calculated from the new loss value. The binary cross-entropy loss function measures the difference between the probability distribution predicted by the model and the probability distribution of the true label; specifically, it calculates the logarithmic difference between the true label and the predicted probability. The mathematical expression of the binary cross-entropy loss function is usually:

[0085]

[0086] In the above formula, BCELoss(·) represents the binary cross-entropy loss function, n represents the number of samples, and y i Let p(y) represent the true label of the i-th sample. i =1) represents the predicted probability of the i-th sample.

[0087] Furthermore, in some embodiments, step S240 may include: S241, setting the hyperparameter adjustment range of the recurrent neural network model; S242, testing the hyperparameters of each group of the recurrent neural network models within the given hyperparameter adjustment range using the Bayesian adaptive method; S243, comparing the loss value of the recurrent neural network model trained using the hyperparameters of each group of the recurrent neural network models with the loss value and the weights, selecting the hyperparameters of the recurrent neural network model corresponding to the minimum loss value, and outputting the optimal recurrent neural network model.

[0088] In some specific embodiments, a Bayesian adaptive method is used to optimize the model's hyperparameters. The Bayesian optimization method is invoked during the training process, automatically adjusting the hyperparameters within a given range based on the model's loss value during iterative training until the optimal hyperparameters are found.

[0089] S300. Based on the optimal recurrent neural network model, the groundwater level change data is classified to obtain the classification results;

[0090] It should be noted that the classification results include groundwater level fluctuation data caused by earthquakes and groundwater level fluctuation data not caused by earthquakes.

[0091] In summary, this invention utilizes an RNN model to automatically identify groundwater level fluctuation data caused by earthquakes, overcoming the shortcomings of previous studies that relied heavily on manual screening. It employs a time-series data reconstruction method to extract groundwater level change characteristics. Based on the temporal persistence of earthquakes, the sampled source data is segmented into sequential data. Leveraging the flexibility of the RNN model in processing sequential data, the temporal variation characteristics of groundwater levels are extracted.

[0092] The innovative technical points of this invention include:

[0093] This invention utilizes an RNN model to identify earthquake-related groundwater level changes. Compared to traditional methods such as mathematical modeling and filtering, it offers advantages such as less data distortion and higher screening efficiency. This invention learns earthquake-related water level features from a dataset and outputs judgment labels of 0 and 1. This adds judgment labels to the input measured water level data, reducing computation and avoiding errors caused by multiple water level calculations. It also shows good recognition performance for water level fluctuations caused by small-scale earthquakes. Furthermore, in the RNN model learning process, each 20-second sequence in the dataset is randomly input. This method ensures the accuracy of model learning while improving the model's efficiency in screening water level data. Compared to traditional methods such as filtering, which require repeated filtering based on a fixed time sequence, this invention divides the time series into 20-second groups for direct judgment, without strict requirements on the fixed order of the data. Therefore, it is faster than traditional methods. Moreover, after model learning, the judgment label for each group can be directly output, eliminating the need for manual comparison of earthquake data and thus improving overall data screening efficiency.

[0094] Please see Figure 3 This application also provides a groundwater level change data classification system based on a recurrent neural network, which can implement the above-mentioned groundwater level change data classification method based on a recurrent neural network. The system includes:

[0095] The first module is used to collect groundwater level change data and atmospheric pressure change data and perform data preprocessing to obtain two-dimensional time series data.

[0096] The second module is used to train the recurrent neural network model based on the two-dimensional time series data to obtain the optimal recurrent neural network model.

[0097] The third module is used to classify groundwater level change data based on the optimal recurrent neural network model and obtain the classification results.

[0098] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0099] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A groundwater level change data classification method based on recurrent neural networks, characterized in that, The method includes: Data on groundwater level changes and atmospheric pressure changes were collected and preprocessed to obtain two-dimensional time series data. The recurrent neural network model is trained based on the two-dimensional time series data to obtain the optimal recurrent neural network model; The groundwater level change data is classified based on the optimal recurrent neural network model to obtain the classification results; the classification results are output in the form of labels, and the labels include groundwater level changes not caused by earthquakes or groundwater level changes caused by earthquakes. The process of collecting groundwater level change data and atmospheric pressure change data, and performing data preprocessing to obtain two-dimensional time series data includes: Collect data on groundwater level changes and atmospheric pressure changes; Set the data periodic sequence; The groundwater level change data and the atmospheric pressure change data are processed according to the data period sequence to obtain groundwater level time change data and atmospheric pressure time change data. The groundwater level time variation data and the atmospheric pressure time variation data are integrated to obtain the two-dimensional time series data.

2. The method according to claim 1, characterized in that, The step of training the recurrent neural network model based on the two-dimensional time series data to obtain the optimal recurrent neural network model includes: Construct a recurrent neural network model; The two-dimensional time series data is input into the recurrent neural network model for training. The weights between the two-dimensional time series data and the weights between each network layer of the recurrent neural network model are obtained through feedforward calculation. The loss value of the recurrent neural network model is obtained by performing loss calculation on the recurrent neural network model during the training process using the binary cross-entropy loss function. Based on the loss value and the weights, the hyperparameters of the recurrent neural network model are iteratively optimized using a Bayesian adaptive method to obtain the optimal recurrent neural network model.

3. The method according to claim 2, characterized in that, The recurrent neural network model includes an input layer, a hidden layer, and an output layer. The output of the input layer is connected to the input of the hidden layer, and the output of the hidden layer is connected to the input of the output layer, wherein: The input layer comprises two neurons; The hidden layer comprises several neurons and uses the sigmoid function as the activation function. The output layer consists of one neuron.

4. The method according to claim 2, characterized in that, The step of inputting the two-dimensional time series data into the recurrent neural network model for training, and obtaining the weights between the preceding and following two-dimensional time series data and the weights between each network layer of the recurrent neural network model through feedforward calculation, includes: The two-dimensional time series data is input into the recurrent neural network model; Based on the input layer of the recurrent neural network model, the two-dimensional time series data is input into the hidden layer; Based on the hidden layer of the recurrent neural network model, feedforward calculation is performed on the two-dimensional time series data to obtain the weights between the two-dimensional time series data before and after the label to be assigned and the weights between each network layer of the recurrent neural network model. Based on the output layer of the recurrent neural network model, the weights between the two-dimensional time series data before and after the data to be labeled and the weights between each network layer of the recurrent neural network model are labeled to obtain the preliminary weights between the two-dimensional time series data before and after the data and the weights between each network layer of the recurrent neural network model. Repeat the above steps of inputting the two-dimensional time series data into the recurrent neural network model for training until the loss function of the recurrent neural network model reaches its minimum value, and output the weights between the two-dimensional time series data and the weights between each network layer of the recurrent neural network model.

5. The method according to claim 4, characterized in that, The expression for calculating the weights of the hidden layer in the recurrent neural network model is as follows: In the above formula, This represents the weights of the two-dimensional time series vector data output at the current moment. This represents the weights of the two-dimensional time series vector data output at the previous time step. This represents the sigmoid function. This represents the weight matrix between the input layer and the hidden layer. This represents the weight matrix that connects the states of two sequences at different time steps in the hidden layer. This represents the bias vector of the input cell. This represents the bias vector of the hidden unit. This indicates the input data.

6. The method according to claim 4, characterized in that, The expression for the output layer of the recurrent neural network model is: In the above formula, This represents the output of the recurrent neural network model. This represents the weight matrix between the hidden layer and the output layer. This represents the bias vector in the output layer. This represents the output of the hidden layer in a recurrent neural network model. This represents the sigmoid function.

7. The method according to claim 2, characterized in that, The expression for the binary cross-entropy loss function is: In the above formula, This represents the binary cross-entropy loss function. Indicates the number of samples. Indicates the first The true label of each sample Indicates the first The predicted probability of a sample.

8. The method according to claim 2, characterized in that, The step of iteratively optimizing the hyperparameters of the recurrent neural network model using a Bayesian adaptive method based on the loss value and the weights to obtain the optimal recurrent neural network model includes: Set the hyperparameter adjustment range of the recurrent neural network model; The hyperparameters of each group of the recurrent neural network models within the given hyperparameter adjustment range are tested using the Bayesian adaptive method. Based on the loss value and the weights, compare the loss value of the recurrent neural network model obtained by training with the hyperparameters of each group of recurrent neural network models, select the hyperparameters of the recurrent neural network model with the smallest loss value, and output the optimal recurrent neural network model.

9. A groundwater level change data classification system based on recurrent neural networks, characterized in that, The system includes: The first module is used to collect groundwater level change data and atmospheric pressure change data and perform data preprocessing to obtain two-dimensional time series data. The second module is used to train the recurrent neural network model based on the two-dimensional time series data to obtain the optimal recurrent neural network model. The third module is used to classify groundwater level change data based on the optimal recurrent neural network model and obtain classification results; the classification results are output in the form of labels, and the labels include groundwater level changes not caused by earthquakes or groundwater level changes caused by earthquakes. The process of collecting groundwater level change data and atmospheric pressure change data, and performing data preprocessing to obtain two-dimensional time series data includes: Collect data on groundwater level changes and atmospheric pressure changes; Set the data periodic sequence; The groundwater level change data and the atmospheric pressure change data are processed according to the data period sequence to obtain groundwater level time change data and atmospheric pressure time change data. The groundwater level time variation data and the atmospheric pressure time variation data are integrated to obtain the two-dimensional time series data.

Images