A machine learning-based method for identifying mudslide seismic signals

Abstract

The present application relates to the technical field of debris flow identification, and discloses a debris flow seismic signal identification method based on machine learning. The method comprises the following steps: obtaining a debris flow seismic signal; constructing a debris flow seismic signal identification model based on a slice attention mechanism; using the debris flow seismic signal identification model based on the slice attention mechanism to firstly perform instance normalization processing on the debris flow seismic signal, then cutting the debris flow seismic signal through a sliding window with overlap, performing Fourier transform processing on each slice, then extracting slice time sequence features of the slice sequence subjected to the Fourier transform through an LSTM model, combining position encodings of different position slices, calculating attention weights by using a multi-head attention mechanism, and finally obtaining a debris flow seismic signal identification result through a full connection layer.

Description

Technical Field

[0001] This invention relates to the field of geological hazard identification, specifically to a method for identifying debris flow ground motion signals based on machine learning. Background Technology

[0002] Debris flows are a common surface material process formed by the rapid sliding of large amounts of water, soil, rocks, and other debris under the influence of gravity. They are characterized by extremely high fluidity and impact force, making them highly destructive. Because debris flows are often sudden and highly unpredictable, rapid and accurate identification is a crucial prerequisite for geological disaster prevention and mitigation.

[0003] Traditional debris flow identification methods mainly include: using mud level gauges or current meters to monitor sudden increases in the depth or velocity of the flow in the riverbed to identify debris flows. While these methods are effective, they often lack the conditions for installing such instruments in the riverbed. Another method involves listening for abnormal sounds, such as a rumbling sound emanating from a ravine or deep valley, which may indicate a debris flow. However, this only provides a rough guess and cannot pinpoint the exact location of the debris flow, and its timeliness is poor. Video surveillance technology can be used to monitor the riverbed in real time to identify debris flows, but its effectiveness is limited by factors such as sunlight, power supply, and video signal transmission in mountainous areas.

[0004] In recent years, machine learning methods have been widely used in debris flow seismic signal identification, but they still face many challenges: 1) Traditional models cannot directly process the original waveform signals, relying on manual feature extraction, which easily introduces redundant or insufficient features, leading to unreasonable inductive bias; 2) Convolutional neural networks (CNNs) require multiple layers during dimensionality reduction, which can easily lead to overfitting. To address these challenges, this invention proposes an improved Transformer-based model—Patch Fourier Transformer (PFT). PFT, through its patch attention mechanism, can effectively perceive energy changes in the time-frequency map and exhibit attention weights highly consistent with its spatiotemporal distribution. Tested using over a dozen debris flow events from Illgraben in 2020, the PFT model achieved an accuracy exceeding 96% in debris flow seismic waveform identification. These results demonstrate the significant potential and advantages of PFT in debris flow identification tasks. Summary of the Invention

[0005] To address the aforementioned shortcomings in the existing technology, this invention provides a method for identifying debris flow ground motion signals based on machine learning.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0007] A method for identifying debris flow seismic signals based on machine learning includes the following steps:

[0008] S1. Obtain the ground motion signal of debris flow;

[0009] S2. Construct a debris flow seismic signal identification model based on slice attention mechanism;

[0010] S3. Based on the constructed slice attention mechanism, the debris flow ground motion signal recognition model is used to extract the temporal features of the debris flow ground motion signal and embed the location codes at different locations.

[0011] S4. The attention weight is calculated using a multi-head attention mechanism, and the debris flow ground motion signal identification result is obtained through the fully connected layer.

[0012] Furthermore, step S3 specifically includes the following steps:

[0013] S31. Debris flow ground motion signal recognition model based on slice attention mechanism is used to perform instance normalization processing on debris flow ground motion signal.

[0014] S32. The debris flow ground motion signal is cut by overlapping sliding windows, and the Fourier transform of each cut slice is performed to obtain the slice sequence.

[0015] S33. Extract the temporal features of the slices from the Fourier transform sequence using an LSTM model and embed the positional codes of the corresponding slices.

[0016] Furthermore, the specific method of normalization in S21 is as follows:

[0017]

[0018] In the formula, x(t) is the original value at time t, μ and σ are the mean and standard deviation of the waveform, respectively, and γ and β are the learnable scaling and offset parameters, respectively.

[0019] Furthermore, step S33 specifically includes the following steps:

[0020] S331. The Fourier transform slice sequence is cut into a smaller-dimensional slice sequence by using overlapping sliding windows.

[0021] S332. Extract the temporal features of each slice in the obtained smaller dimension slice sequence through the LSTM model, so that the original slice sequence becomes a new slice sequence with smaller dimension and the same number of slices, and make the LSTM model share weights in all new slice sequences.

[0022] S333. The location information of the new slice sequence is embedded into the constructed debris flow ground motion signal identification model through location encoding.

[0023] Furthermore, the specific method of embedding in S333 is as follows:

[0024]

[0025] In the formula, PE represents the embedding of the corresponding position and the corresponding vector dimension, pos represents the current position, i is the dimension index in the position vector, and d... model It is the dimension of the input vector.

[0026] Furthermore, the slice with embedded position encoding in S333 is represented as follows:

[0027]

[0028] In the formula, For slices after embedding position encoding, W pos For position encoding, For slices in the new slice sequence.

[0029] Furthermore, step S4 specifically includes the following steps:

[0030] S41. Use the slice with embedded position encoding as the input to the encoder;

[0031] S42. A multi-head attention mechanism is used to transform the encoder input into a query matrix, a key matrix, and a value matrix, respectively.

[0032] S43. Calculate attention using the transformed query matrix, key matrix, and value matrix;

[0033] S44. Segment the outputs of all heads and map the segmented result to the set dimension of the constructed debris flow ground motion signal identification model through a linear layer.

[0034] S45. Design a loss function to measure the difference between the predicted probability and the true probability label, and then output the debris flow ground motion signal identification result.

[0035] Furthermore, the query matrix, key matrix, and value matrix in S42 are respectively represented as follows:

[0036]

[0037] In the formula, The lookup matrix, key matrix, and value matrix for the h-th head are, in order. These are the linear transformation matrices of the query matrix, key matrix, and value matrix of the h-th head, respectively.

[0038] Furthermore, the specific method by which the spliced ​​results are mapped to the defined dimensions of the constructed debris flow seismic signal identification model through a linear layer in S44 is as follows:

[0039] MultiHead(Q i ,K i V i = Concat(head1, head2, ..., head) H W O

[0040] Q i K i V i The matrix consists of, in order: query matrix, key matrix, and value matrix; head H For each header's output, the subscript H is its number, and W... O This is the weight matrix of the linear layer.

[0041] Furthermore, the loss function in S45 is expressed as:

[0042]

[0043] In the formula, N is the number of samples, and y i It is the true label of the i-th sample. is the predicted probability of the i-th sample, and BCE is the binary cross-entropy loss function.

[0044] The present invention has the following beneficial effects:

[0045] This invention is used to identify debris flows in continuous vibration signals. The PFT attention weight distribution has a high similarity to the time-frequency plot, which is a key factor for high-precision identification. The model exhibits high accuracy in debris flow waveform identification and can accurately identify multiple debris flows with short time intervals, demonstrating its potential in practical applications. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the process of the debris flow ground motion signal identification method based on machine learning according to the present invention.

[0047] Figure 2 This is a schematic diagram illustrating the performance of the PFT model at different wavelengths in an embodiment of the present invention.

[0048] Figure 3a This is a schematic diagram of the original waveform in an embodiment of the present invention.

[0049] Figure 3b This is a time-frequency diagram of the original waveform in an embodiment of the present invention.

[0050] Figure 3c This is the distribution of attention weights in the first-layer attention mechanism of the PFT model in an embodiment of the present invention.

[0051] Figure 3dThis is the attention weight distribution of the PFT model in the second-layer attention mechanism according to an embodiment of the present invention.

[0052] Figure 3e This is the attention weight distribution of the PFT model in the third-layer attention mechanism according to an embodiment of the present invention. Detailed Implementation

[0053] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0054] A machine learning-based method for identifying debris flow seismic signals, such as Figure 1 As shown, it includes the following steps:

[0055] S1. Obtain the ground motion signal of debris flow;

[0056] Because the entire dataset is imbalanced between debris flow (positive) and noise (negative) samples, to ensure the model can fully learn the waveform features of debris flows, this embodiment randomly samples noise samples from 2017-2019 to achieve a 1:1 ratio of positive to negative samples. Subsequently, the dataset is divided into training, validation, and test sets, with proportions of 80%, 10%, and 10%, respectively. The training set is primarily used for the model to learn high-level feature representations of waveform data. The validation set is used to prevent overfitting and to save the optimal model state in a timely manner through an early stopping strategy. The test set is used for the final model performance evaluation to ensure the unbiasedness of the results. Data from 2020 is used to evaluate the model's continuous identification and early warning performance of vibration signals in actual operation.

[0057] This study employs a supervised learning method. The start and end times of each debris flow event in the training set are defined by timestamps. Within these time intervals, the labels of relevant samples are set to 1. Considering the uncertainty of manual annotation and the specific characteristics of debris flow events, this embodiment sets linearly increasing and decreasing labels before and after the start and end of the debris flow, respectively, gradually increasing from 0 to 1 and decreasing from 1 to 0. This blurs the specific start and end times of the debris flow, enhancing the model's ability to identify event boundaries.

[0058] In order to obtain sufficient contextual information, in this invention, the three-component waveform is divided into 5-minute long waveforms with an overlap rate of 80%, that is, the step size of each division is 1 minute.

[0059] S2. Construct a debris flow seismic signal identification model based on slice attention mechanism;

[0060] The Transformer architecture using a pure encoder is primarily used for tasks that do not require sequence generation, such as text classification, entity recognition, or sentence embedding generation. This architecture contains only the encoder portion of the Transformer, with each encoder layer consisting of a self-attention mechanism and a feedforward network. By capturing the global dependencies between elements in the input sequence, this architecture generates high-quality feature representations, effectively supporting downstream classification or other tasks.

[0061] We define the problem as follows: Given a three-component data sample, Where L is the length of the waveform, our model uses... express, Where C represents a classification probability for the waveform at each minute, and the model uses a Transformer encoder as the core architecture of this embodiment.

[0062] S3. Based on the constructed slice attention mechanism, the debris flow ground motion signal recognition model is used to extract the temporal features of the debris flow ground motion signal and embed the location codes at different locations.

[0063] This embodiment specifically includes the following steps:

[0064] S31. Debris flow ground motion signal recognition model based on slice attention mechanism is used to perform instance normalization processing on debris flow ground motion signal.

[0065] Instance normalization performs independent data normalization on each channel of each sample to adjust the scale and distribution of the data, making the model training process more stable and helping the model converge quickly in the early stages of training.

[0066]

[0067] Where x(t) is the original value at time t, μ and σ are the mean and standard deviation of the waveform, respectively, and γ and β are learnable scaling and offset parameters.

[0068] S32. The debris flow ground motion signal is cut by overlapping sliding windows, and the Fourier transform of each cut slice is performed to obtain the slice sequence.

[0069] For the data of each channel in the input sample In this embodiment, it is divided into overlapping slices, and the length of each slice is denoted as in this embodiment. And the step size of the slice is denoted as in this embodiment. Then in this process x i Will become Where N is the number of slices, and

[0070] In time-series models, the Fourier transform effectively captures the frequency characteristics of the data, helping the model better identify and understand the patterns hidden in the time dimension. Simultaneously, the Fourier transform can effectively separate noise, especially high-frequency noise, in the frequency domain, thereby improving signal quality. Furthermore, by transforming complex time-series data from the time domain into simple frequency components in the frequency domain, the Fourier transform simplifies data processing and reduces model complexity.

[0071] S33. Extract the temporal features of the slices from the Fourier transform sequence using an LSTM model and embed the positional codes of the corresponding slices.

[0072] The specific method is as follows:

[0073] S331. The Fourier transform slice sequence is cut into a smaller-dimensional slice sequence by using overlapping sliding windows.

[0074] S332. Extract the temporal features of each slice in the obtained smaller dimension slice sequence through the LSTM model, so that the original slice sequence becomes a new slice sequence with smaller dimension and the same number of slices, and make the LSTM model share weights in all new slice sequences.

[0075] As the sequence length increases, the dimensionality of each patch in the sequence also increases accordingly. This embodiment uses LSTM to capture the dependencies in the patch sequence and preserves this key information during dimensionality reduction. For each patch: This embodiment reduces it to a smaller patch: Where j = 1, 2, ..., K are the indices of the small patches. For each small patch, the dimensions satisfy... Each patch becomes after passing through LSTM. Finally, the original sequence becomes LSTM shares weights across all patches.

[0076] S333. The location information of the new slice sequence is embedded into the constructed debris flow ground motion signal identification model through location encoding.

[0077] In the Transformer's attention mechanism, the calculation of attention weights is unordered, and it cannot directly capture the sequential relationship between patches in the waveform. The order of patches is particularly important for waveform recognition. Therefore, this embodiment injects the positional information of patches in the waveform explicitly into the model through positional encoding. The embedding method in this embodiment is as follows:

[0078]

[0079] Where pos is the current position (index in the sequence), i is the dimension index in the position vector, and d model It is the dimension of the input vector.

[0080] S4. The attention weight is calculated using a multi-head attention mechanism, and the debris flow ground motion signal identification result is obtained through the fully connected layer.

[0081] This embodiment specifically includes the following steps:

[0082] S41. Use the slice with embedded position encoding as the input to the encoder;

[0083] The Transformer architecture using a pure encoder is primarily used for tasks that do not require sequence generation, such as text classification, entity recognition, or sentence embedding generation. This architecture contains only the encoder portion of the Transformer, with each encoder layer consisting of a self-attention mechanism and a feedforward network. By capturing the global dependencies between elements in the input sequence, this architecture generates high-quality feature representations, effectively supporting downstream classification or other tasks.

[0084] S42. A multi-head attention mechanism is used to transform the encoder input into a query matrix, a key matrix, and a value matrix, respectively.

[0085] For each input flag, compute the Query, Key, and Value vectors. These vectors are obtained through a linear transformation (weight matrix), and the matrix size is N×d. Therefore, the complexity of computing the Query, Key, and Value is O(N·d). 2 ).

[0086] For each query, a dot product is performed with all other keys to obtain an attention score matrix of size N×N. Therefore, the complexity of the dot product is O(N). 2 ).

[0087] Calculating the Softmax weight for each query also has a time complexity of O(N). 2 ).

[0088] The output of each query is a weighted sum of all values, and the time complexity of this weighted summation is O(N). 2 This is because the output at each position needs to be weighted and summed with N Values.

[0089] Therefore, the total complexity of a single Self-Attention computation is: O(N·d) 2 )

[0090] S43. Calculate attention using the transformed query matrix, key matrix, and value matrix;

[0091] Each Transformer layer also contains a feedforward network, typically consisting of two linear transformations and an activation function (such as ReLU). Assume the hidden layer dimension of each layer is d. ff (usually d) ff >>d), then the computation process of the Feedforward network includes:

[0092] The first linear transformation: from dimension d to d ff The computational complexity is O(N·d·d). ff ).

[0093] Activation function: Assuming the complexity of the activation function is a constant O(1), the complexity of this operation is O(N·d). ff ).

[0094] The second linear transformation: from d ff Converting back to d, the computational complexity is O(N·d). ff ·d).

[0095] Therefore, the total complexity of the Feedforward network is: O(N·d) ff ·d).

[0096] Space complexity mainly consists of the following parts:

[0097] Input and output storage: The input and output of each token need to be stored, so the space complexity of input and output is O(N·d).

[0098] The weight matrix for Self-Attention: Self-Attention requires storing an N×N attention score matrix, with a space complexity of O(N). 2 ).

[0099] Parameters of the Feedforward network: The Feedforward network has two linear transformation weight matrices, d×d ff and d ff×d, therefore the space complexity is O(d) ff ·d).

[0100] Therefore, the space complexity of each layer is O(N). 2 +N·d+d ff ·d).

[0101] Where L is the number of Transformer layers, N is the length of the input sequence, and d is the dimension of the hidden layers. ff It is the hidden layer dimension of the Feedforward network.

[0102] When this embodiment divides the one-dimensional sequence into slices and considers a stride of 's', the division method differs from the case without a stride. The stride 's' represents the distance between two consecutive slices, and is typically less than the slice length 'L'. patch (Also known as "overlapping"). This method is similar to the operation of a sliding window.

[0103] To calculate the number of patches after partitioning, this embodiment derives a formula based on the step size and patch length. The sequence length N is divided into multiple patches, with each patch starting at position 0, s, 2s, ..., until the last patch starts at position NL. patch Therefore, the number of patches after partitioning is:

[0104] When considering the step size s, the time complexity of Self-Attention depends on the new sequence length N. patch And the dimension d of each patch. The time complexity of each layer of Self-Attention is still O(N). patch 2 ·d), with a space complexity of O(N). patch 2 +N patch ·d)

[0105] S44. Segment the outputs of all heads and map the segmented result to the set dimension of the constructed debris flow ground motion signal identification model through a linear layer.

[0106] After calculating the positional encoding of patches at different locations, this embodiment first uses a linear layer to... Projected onto the feature space d of the model in this embodiment model And embed positional encoding, i.e. here Then As input to the encoder, the attention mechanism in this embodiment employs multi-head attention (Figure c) to enable the model to capture different features and dependencies of the input data from different angles or subspaces. In this embodiment, h = 1, 2, 3…H represents the number of heads, where each head transforms its input into a Query, Key, and Value matrix respectively. as well as in Then, by calculating its attention

[0107]

[0108] The outputs of all the heads are then concatenated, and the concatenated result is mapped to the required dimension of the model through a linear layer:

[0109] MultiHead(Q i ,K i V i = Concat(head1, head2, ..., head) H W O

[0110] head i The output of each head, Finally, this embodiment uses a feedforward layer to obtain the classification results of the waveform for each minute: Where y k ∈(0,1),k=0,1…C.

[0111] S45. Design a loss function to measure the difference between the predicted probability and the true probability label, and then output the debris flow ground motion signal identification result.

[0112] BCELoss (Binary Cross-Entropy Loss) is used as the loss function in the classification problem of this embodiment to measure the difference between the predicted probability and the true probability label, and it is expressed as follows:

[0113]

[0114] Where N is the number of samples, y i It is the true label of the i-th sample. It is the predicted probability of the i-th sample.

[0115] Because debris flows differ from earthquakes (point sources) in that they are moving line sources, collisions between particles within the fluid, as well as collisions between particles and the riverbed, cause debris flows to continuously radiate surface waves as they move through the channel. To capture contextual information of varying lengths, the experiment maintained the same training samples for different models, but the sample lengths varied across models. Validation results on the 2017-2019 dataset are as follows: Figure 2 As shown.

[0116] pass Figure 2 It can be observed that the F1 score and accuracy generally show an upward trend between 1 and 5 minutes. This indicates that when the waveform is appropriately extended, the waveform itself can provide more contextual information to help the model learn. When the waveform is longer than 5 minutes, the overall performance decreases. This shows that excessively long waveforms firstly lead to a significant increase in the number of model parameters, and due to the limited training samples, they are prone to underfitting. Secondly, larger windows may introduce more noise or cause information overload in some cases, thereby reducing the model's recognition accuracy.

[0117] When using short waveforms (<3 min) as input, the 1 min model exhibits high recall but low precision, indicating that it performs well in identifying debris flow events at this waveform length. However, it also misclassifies a large number of negative samples (non-debris flow events) as positive samples. The 2 min model shows high precision but low recall, indicating that it is accurate in identifying positive samples but misses a significant number of positive samples. When the waveform is appropriately extended (3-5 min), the difference between recall and precision tends to decrease. However, when the waveform becomes too long (>5 min), this difference gradually amplifies, meaning that the model is very cautious in identifying positive samples and only identifies the waveform as positive when it is highly certain. Among the tested waveform lengths, the 5 min waveform strikes a good balance between recall (0.955) and precision (0.959), indicating that a good balance is achieved between the false negative rate and the false positive rate in actual early warning systems.

[0118] These experimental results clearly demonstrate the advantages of the PFT model in several aspects of debris flow early warning tasks: Unlike traditional models that require complex feature engineering, the PFT model can directly extract key features of the time series from the original waveform without human intervention. This gives the model a stronger ability to capture latent patterns in the data, and the PFT model can not only adapt to different time windows of varying lengths but also maintain relatively stable performance at appropriate time scales (2-5 minutes), with PFT_5min showing the best performance.

[0119] To more intuitively visualize the overall performance of the PTF model and its classification performance for each category, this embodiment calculates the ROC curve and the confusion matrix: the former is a tool used to evaluate the performance of a classification model by showing the relationship between the model's false positive rate (FPR) and true positive rate (TPR); the latter displays the comparison between the model's predictions and actual results in matrix form to analyze the model's correct and incorrect classifications.

[0120] The True Positive Rate (TPR) represents the proportion of positive classes that are correctly classified, also known as recall or sensitivity; the False Positive Rate is the proportion of incorrect model judgments to the true negative values.

[0121]

[0122] Where TP is the number of correctly predicted positive classes, FN is the number of classes that were actually positive but were predicted as negative classes, FP is the number of classes that were actually negative but were predicted as positive classes, and TN is the number of correctly predicted negative classes.

[0123] The confusion matrix reveals that when the waveform is less than 5 minutes long, appropriately extending the waveform is beneficial for identifying positive examples (debris flow) (0.93->0.95). When the waveform is longer than 5 minutes long, its ability to identify positive examples decreases, but the model performs more accurately in identifying negative samples (0.97->0.99). This may be because when the debris flow waveform is too long, the dynamic process of the debris flow itself changes (such as erosion), causing differences in the debris flow signal before and after in the waveform, thus reducing the model's accuracy.

[0124] Overall, the PFT-based model demonstrates strong robustness in debris flow classification tasks, especially at a wavelength of 5 minutes, where it exhibits high ability to identify both positive and negative samples.

[0125] The model's attention mechanism learns the importance of different patches in the original waveform, i.e., the distribution of attention weights. The model can form a global understanding of the waveform, thus "sensing" the energy changes in the waveform, and the distribution of attention weights shows spatiotemporal consistency with the time-frequency plot of the waveform.

[0126] Specifically, we visualized the attention weights at different levels of the model. Figure 3(c) shows that when extracting features in the first layer of the model, the weight difference between different patches is only 0.93% (weight range: 0.00845-0.00853). This result indicates that in the initial stage, the model focuses on the overall features of the waveform, forming a global understanding, and its weight distribution is highly consistent with the energy changes in the time-frequency graph. In the second layer of the model (Figure 3(d)), the attention mechanism assigns higher weights to the parts of the waveform with larger energy changes (the weights are increased by about 2 times compared to the first layer), and the weight difference between different patches is amplified by about 70%. This phenomenon indicates that the model begins to focus on regions of significant energy change in the waveform. In the last layer of the model (Figure 3(e)), the weight difference between patches is further amplified (about 80%), and the weight distribution becomes clearer. Overall, during the feedforward process, the change in the attention distribution resembles a denoising process, and its attention weight distribution always maintains a high degree of spatiotemporal consistency with the energy changes in the time-frequency graph.

[0127] We further analyzed the impact of different Fourier transform positions on the attention distribution. The results show that directly performing a Fourier transform on the original waveform leads to the model over-focusing on a specific frequency band. Without a Fourier transform, the model is almost unable to effectively distinguish energy changes in different parts at the first layer, indicating that the model struggles to effectively learn waveform features. Both situations result in a significant decrease in model performance, demonstrating that accurate attention weight distribution is crucial for high-precision recognition.

[0128] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0129] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0130] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0131] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0132] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A machine learning-based method for identifying a mudslide seismic signal, characterized by, It comprises the following steps: S1, acquiring a debris flow seismic signal; S2, constructing a debris flow seismic signal recognition model based on a slicing attention mechanism, the recognition model comprising an encoder part of a Transformer, each encoder layer comprising a self-attention mechanism and a feedforward network, and generating high-quality feature representations by capturing global dependencies between elements in the input sequence; S3, extracting time sequence features of the debris flow seismic signal based on the constructed debris flow seismic signal recognition model based on the slicing attention mechanism, and embedding position encodings at different positions, specifically comprising the following steps: S31, performing instance normalization processing on the debris flow seismic signal by using the debris flow seismic signal recognition model based on the slicing attention mechanism; S32, cutting the debris flow seismic signal by using a sliding window with overlap, and performing Fourier transform on each cut slice to obtain a slice sequence; S33, extracting slice time sequence features by using an LSTM model on the slice sequence after Fourier transform, and embedding position encodings of the corresponding position slices, specifically comprising the following steps: S331, cutting the slice sequence after Fourier transform by using a sliding window with overlap to make it a slice sequence with smaller dimensions; S332, extracting time sequence features of each slice in the obtained slice sequence with smaller dimensions by using an LSTM model, making the original slice sequence into a new slice sequence with smaller dimensions and unchanged number of slices, and making the LSTM model share weights in all new slice sequences; S333, embedding position information of the new slice sequence into the constructed debris flow seismic signal recognition model by using position encodings; S4, calculating attention weights by using a multi-head attention mechanism, and obtaining a debris flow seismic signal recognition result by using a fully connected layer.

2. The machine learning based identification method of mudslide seismic ground motion signals according to claim 1, characterized in that, The specific way of normalization in S31 is: where is the original value at time point t, and are the mean and standard deviation in the waveform, respectively, and are the learnable scaling and offset parameters, respectively. 3.The machine learning based identification method of mudslide seismic ground motion signal according to claim 1, characterized in that, The specific way of embedding in S333 is: wherein is an embedding for the corresponding position and corresponding vector dimension, is a current position, is a dimension index in the position vector, is a dimension of the input vector.

4. The machine learning based identification method of mudslide seismic ground motion signals according to claim 1, characterized in that, The slice representation after embedding position encodings in S333 is: wherein is an embedded position encoded slice, is a position encoding, is a slice in a new slice sequence, is a linear layer projection matrix.

5. The machine learning based identification method of mudslide seismic ground motion signals according to claim 1, characterized in that, S4 specifically comprises the following steps: S41, taking the slice after embedding position encodings as the input of the encoder; S42, converting the encoder input into a query matrix, a key matrix and a value matrix by using a multi-head attention mechanism; S43, calculating attention by using the converted query matrix, key matrix and value matrix; S44, concatenating the outputs of all heads, and mapping the concatenated result to the set dimension of the constructed debris flow seismic signal recognition model by using a linear layer; S45, designing a loss function to measure the difference between the predicted probability and the true probability label, and then outputting a debris flow seismic signal recognition result. 6.The machine learning based identification method of mudslide seismic ground motion signal according to claim 1, wherein, The query matrix, key matrix and value matrix in S42 are respectively represented as: In the formula, in the formula, Linear transformation matrix of the query matrix, the key matrix and the value matrix of the hth head in turn, Linear transformation matrix of the query matrix, the key matrix and the value matrix of the hth head in turn, 7. The machine learning based identification method of mudslide seismic ground motion signals according to claim 5, characterized in that, The specific way of mapping the concatenated result to the set dimension of the constructed debris flow seismic signal recognition model in S44 is: wherein are, in order, the query matrix, the key matrix, and the value matrix; is the output of each head, with subscript H its number, is the weight matrix of the linear layer. 8.The machine learning based identification method of mudslide seismic ground motion signal according to claim 5, characterized in that, The loss function in S45 is represented as: wherein, N is the number of samples, is the true label of the th sample, is the predicted probability of the th sample, is the binary cross-entropy loss function.

Images