CN-122018021-A - Ground-air transient electromagnetic data rapid inversion method based on long-short-time memory network

Abstract

The invention relates to the technical field of geophysical exploration and discloses a ground-air transient electromagnetic data rapid inversion method based on a long-short-term memory network, which comprises the following steps of constructing a ground-air transient electromagnetic training data set, generating resistivity and thickness parameters through a random layered model, and generating corresponding transient electromagnetic response data based on forward modeling; the training data set is preprocessed, including normalization, time sequence remodeling and data set division, and a CNN-BiLSTM-Attention mixed neural network model is constructed, wherein the model comprises a CNN feature extraction module, a BiLSTM time sequence modeling module, an Attention mechanism module and a Seq2Seq decoder module which are connected in sequence. According to the ground-air transient electromagnetic data rapid inversion method based on the long-short-time memory network, the trained network is utilized for reasoning, the single-measuring-point inversion time can be shortened to the second level, the traditional interpretation period of several days is compressed to the hour level, and the response speed of geological disaster monitoring and investigation is greatly improved.

Inventors

• LI BING
• LIU JIANLI
• CHEN XIAOLONG
• LI XINSHENG
• LI MING

Assignees

• 陕西地矿物化探队有限公司

Dates

Publication Date

20260512

Application Date

20251225

Claims (10)

1. 1. A ground-air transient electromagnetic data rapid inversion method based on a long-short-time memory network is characterized by comprising the following steps: Constructing a ground-air transient electromagnetic training data set, which comprises the steps of generating resistivity and thickness parameters through a random layered model, and generating corresponding transient electromagnetic response data based on forward modeling; Preprocessing the training data set, including normalization, time sequence remodeling and data set division; constructing a CNN-BiLSTM-Attention mixed neural network model, wherein the model comprises a CNN feature extraction module, a BiLSTM time sequence modeling module, an Attention mechanism module and a Seq2Seq decoder module which are connected in sequence; Training the hybrid neural network model by using the training data set until a loss function converges; And inputting the ground-to-air transient electromagnetic observation data to be inverted into a trained model, and outputting the resistivity and thickness parameters of the underground medium.
2. 2. The method for quickly inverting the ground-to-air transient electromagnetic data based on the long-short-term memory network according to claim 1, wherein the CNN feature extraction module comprises at least one-dimensional convolution layer and a pooling layer and is used for extracting local spatial features of the transient electromagnetic signals.
3. 3. The method for rapidly inverting ground-to-air transient electromagnetic data based on long and short term memory network according to claim 1, wherein the BiLSTM time sequence modeling module comprises a bidirectional LSTM layer for modeling a bidirectional long-range time sequence dependency of transient electromagnetic signals.
4. 4. The method for rapidly inverting ground-air transient electromagnetic data based on long-short-time memory network according to claim 1, wherein the attention mechanism module is Bahdanau attention mechanisms for dynamically weighting BiLSTM output and focusing on critical time channel information.
5. 5. The method for rapidly inverting ground-to-air transient electromagnetic data based on long and short term memory network according to claim 1, wherein the Seq2Seq decoder module comprises an LSTM layer and a fully connected layer for autoregressive generation of a sequence of formation parameters.
6. 6. The method for rapidly inverting ground-to-air transient electromagnetic data based on long and short term memory network according to claim 1, wherein the construction of the training data set comprises performing log space uniform sampling in a preset resistivity and thickness range and injecting segment differentiation noise to simulate actual measurement conditions.
7. 7. The method for fast inversion of ground-to-air transient electromagnetic data based on long and short term memory network according to claim 1, wherein the loss function of the model comprises Mean Square Error (MSE) and Mean Absolute Error (MAE) for the optimization of the training phase and the evaluation of the verification phase, respectively.
8. 8. The method for quickly inverting the ground-to-air transient electromagnetic data based on the long-short-term memory network according to claim 1, wherein the method is suitable for the ground-to-air transient electromagnetic data inversion in goaf, aquifer or mineral exploration.
9. 9. The method for rapidly inverting ground-to-air transient electromagnetic data based on long-short-term memory network according to claim 1, wherein the inversion time of the model to single-measurement-point data in an inference stage is less than 0.5 seconds.
10. 10. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method of any of claims 1 to 9.

Description

Ground-air transient electromagnetic data rapid inversion method based on long-short-time memory network Technical Field The invention relates to the technical field of geophysical exploration, in particular to a ground-air transient electromagnetic data rapid inversion method based on a long-short-time memory network. Background The goaf in the northern region of Shaanxi is widely distributed, the geological disaster risk is prominent, and a high-precision and high-efficiency geophysical detection means are needed to realize effective disaster prevention and reduction. The ground-to-air transient electromagnetic method combines the advantages of ground emission and air reception, has the characteristics of large detection depth, high resolution, strong terrain adaptability and the like, and becomes an effective method for finely detecting the goaf in a complex environment. However, the method has the problems of large data acquisition amount, complex time sequence characteristics, low calculation efficiency, long interpretation period, easy trapping in local optimization and the like because the traditional inversion method depends on initial model selection and repeated iterative optimization, and is difficult to meet the requirements of real-time engineering detection. The existing inversion method is mostly based on regularization or linear approximation theory, is sensitive to noise and depends on manual experience, and particularly has low efficiency when processing multi-measuring-point and large-range data. In recent years, although research attempts are made to use fully-connected neural networks for electromagnetic inversion, the problems of weak feature extraction capability, insufficient long-term dependence modeling and the like generally exist, and the inversion precision and generalization capability are limited. Therefore, an intelligent inversion method capable of achieving both efficiency and precision is urgently needed to achieve rapid and reliable interpretation of the ground-to-air transient electromagnetic data. Disclosure of Invention In order to solve the problems in the background technology, the invention provides the following technical scheme that the ground-to-air transient electromagnetic data rapid inversion method based on the long-short-time memory network comprises the following steps: Constructing a ground-air transient electromagnetic training data set, which comprises the steps of generating resistivity and thickness parameters through a random layered model, and generating corresponding transient electromagnetic response data based on forward modeling; Preprocessing the training data set, including normalization, time sequence remodeling and data set division; constructing a CNN-BiLSTM-Attention mixed neural network model, wherein the model comprises a CNN feature extraction module, a BiLSTM time sequence modeling module, an Attention mechanism module and a Seq2Seq decoder module which are connected in sequence; Training the hybrid neural network model by using the training data set until a loss function converges; And inputting the ground-to-air transient electromagnetic observation data to be inverted into a trained model, and outputting the resistivity and thickness parameters of the underground medium. Preferably, the CNN feature extraction module includes at least one-dimensional convolution layer and a pooling layer, for extracting local spatial features of the transient electromagnetic signal. Preferably, the BiLSTM timing modeling module includes a bidirectional LSTM layer for modeling a bidirectional long-range timing dependency of the transient electromagnetic signal. Preferably, the attention mechanism module is Bahdanau attention mechanisms for dynamically weighting BiLSTM output and focusing on critical time-channel information. Preferably, the Seq2Seq decoder module comprises an LSTM layer and a fully connected layer for autoregressive generation of a sequence of formation parameters. Preferably, the construction of the training data set includes performing log-space uniform sampling within a preset resistivity and thickness range, and injecting segment-differentiated noise to simulate the actual measurement condition. Preferably, the loss function of the model comprises a Mean Square Error (MSE) and an average absolute error (MAE) for the optimization of the training phase and the evaluation of the verification phase, respectively. Preferably, the method is suitable for ground-to-air transient electromagnetic data inversion in goaf, aquifer or mineral exploration. Preferably, the model has an inversion time of less than 0.5 seconds for single-station data in the reasoning stage. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method. Compared with the prior art, the invention provides a ground-air transient electromagnetic data rapid inversion method based on a long-short-tim