Distributed optical fiber temperature sensor based on deep learning

Abstract

The invention provides a distributed optical fiber temperature sensor based on deep learning, and relates to the technical field of distributed optical fiber sensing. According to the sensor, a multimode optical fiber is adopted as a temperature sensing unit, a broadband laser light source is used for outputting laser, and a spectrograph is used for collecting spectral information of the multimode optical fiber at different positions and different temperatures. The collected spectral data is processed through the data preprocessing module, so that the spectral data is suitable for input of a neural network. And designing and training a deep learning neural network model, demodulating the spectral information by using the deep learning neural network model, and predicting the temperature of each position of the multimode optical fiber. The sensor overcomes the problems that a traditional distributed optical fiber temperature sensor is prone to noise interference, low in temperature resolution and slow in time response, has the advantages of being high in precision and high in real-time performance, and is widely suitable for temperature monitoring requirements in the fields of industrial automation, environment monitoring, electric power facility maintenance and the like.

Description

Technical Field

[0001] The present invention relates to the field of distributed optical fiber sensing technology, and in particular to a distributed optical fiber temperature sensing method based on deep learning. Background Art

[0002] Since optical fiber has the ability to transmit and sense information, optical fiber sensors can be used for temperature measurement and sensing at any location. With the increasing demand for temperature monitoring in industrial automation, environmental monitoring, power facility maintenance and other fields, distributed optical fiber temperature sensors have attracted widespread attention due to their non-invasiveness, anti-electromagnetic interference, compact size and excellent corrosion resistance.

[0003] However, traditional distributed fiber optic temperature sensors are susceptible to noise interference in complex environments, resulting in reduced measurement accuracy. Traditional distributed fiber optic temperature sensors also have problems such as low temperature resolution and slow time response.

[0004] At present, the application of machine learning in distributed fiber optic sensing has made significant progress. Machine learning plays an important role in solving the problems related to spectral information extraction, spectral noise suppression and multi-parameter identification of distributed sensors. In addition, machine learning can be combined with interferometric sensing systems to not only realize distributed temperature measurement, but also improve the resolution and real-time performance of temperature measurement. As a powerful machine learning method, deep learning has the ability of automatic feature extraction and pattern recognition, and is expected to play an important role in the signal processing of distributed fiber optic temperature sensors. Summary of the invention

[0005] The purpose of the present invention is to provide a distributed optical fiber temperature sensor based on deep learning, which is used to solve the problems of traditional distributed optical fiber sensors such as susceptibility to noise interference, low temperature resolution and slow time response.

[0006] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions: a distributed optical fiber temperature sensor based on deep learning, including obtaining training data: wherein, a multimode optical fiber is used as the temperature sensing unit of the distributed optical fiber temperature sensor, and a spectrometer is used to collect spectral information at different positions and temperatures of the multimode optical fiber; data preprocessing: preprocessing operations are performed on the collected spectral data so that it can correctly output the neural network; neural network model design and training: design a suitable deep learning neural network model, and use the collected data set for network training.

[0007] Furthermore, to obtain training data, the present invention uses a multimode optical fiber as a temperature sensing unit, uses a broadband laser light source to output laser, and uses a spectrometer to collect spectral information. A fiber optic fusion splicer is used to fuse a section of single-mode optical fiber on both sides of the multimode optical fiber used as the temperature sensing unit, and then one end is connected to the laser light source and the other end is connected to the spectrometer to form a spectral information collection device. When a heating plate is used to heat at multiple positions of the temperature sensing unit, the resulting spectrum will change, and the distributed temperature sensing function is realized through subsequent operations of the present invention. In the case of multiple temperature combinations at multiple positions of the temperature sensing unit, the spectral information is collected by a spectrometer to construct a data set for training a neural network.

[0008] Further, data preprocessing, the spectral data collected in the above steps are saved as CSV format files through the spectrum, which contain data information and wavelength configured by the spectrometer and the corresponding light intensity information at the wavelength. The deep learning neural network used in the present invention only requires light intensity information, so it is necessary to perform batch data preprocessing on the data collected in the above steps. The present invention uses the Pycharm integrated development environment and uses the python language to perform batch preprocessing operations on the raw data: batch read CSV files, delete useless spectrometer configuration information, delete wavelength data columns, only save light intensity data columns, and save the processed data to a specified folder. In order to facilitate network training, the preprocessed spectral data is classified by temperature label.

[0009] Further, the neural network model is designed and trained. The spectral data we collected and preprocessed is a one-dimensional array containing intensity values ​​of different wavelengths. Therefore, for spectral data obtained from highly multimode interference, the MLP designed based on matrix multiplication is the preferred architecture for describing the interactions between all features in the layer. In order to train and evaluate the MLP, the data set is divided into three subsets: training, validation, and testing. The training and validation subsets are used for training and cross-validation to fine-tune hyperparameters such as learning rate and mini-batch size, while the test subset is retained for evaluation purposes. The MLP network model used in the present invention is a regression network model, the input is the preprocessed one-dimensional spectral intensity data, and the output is the predicted temperature value of each position, which is determined to be accurate within a given error range. After the model is adjusted and optimized, the trained network model is saved and can be used for real-time temperature demodulation of distributed optical fiber temperature sensors. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of a process of a distributed optical fiber temperature sensor based on deep learning provided by an embodiment of the present invention;

[0011] Figure 2 It is a schematic diagram of a distributed optical fiber temperature sensor data acquisition system provided by an embodiment of the present invention;

[0012] Figure 3 It is a schematic diagram of a deep learning network model structure provided by an embodiment of the present invention;

[0013] Reference numerals: 1 is a broadband laser source; 2 is a first single-mode optical fiber; 3 is a multi-mode optical fiber; 4 is a second single-mode optical fiber; 5 is a heating device; and 6 is a spectrometer. DETAILED DESCRIPTION

[0014] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The steps in the following embodiments are only for the convenience of explanation and description, and the order between the steps is not limited. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0015] See also Figure 1 The embodiment of the present invention provides a process diagram of a distributed optical fiber temperature sensor based on deep learning, including obtaining training data: wherein, a multimode optical fiber is used as the temperature sensing unit of the distributed optical fiber temperature sensor, and a spectrometer is used to collect spectral information at different positions and temperatures of the multimode optical fiber; data preprocessing: preprocessing operations are performed on the collected spectral data so that it can correctly output the neural network; constructing a deep neural network model: the deep neural network model constructed by the present invention is a multi-layer perceptron; using training data to train the deep neural network model: using the preprocessed data, inputting the neural network to complete the training process of the deep neural network; using the optical fiber spectrum and the trained network model to predict the temperature value in real time: after the network model training is completed, it can be used in the distributed optical fiber temperature sensor to realize the function of real-time temperature prediction.

[0016] See also Figure 2 An embodiment of the present invention provides a distributed optical fiber temperature sensor data acquisition system, comprising a broadband laser source 1, a first single-mode optical fiber 2, a multimode optical fiber 3, a second single-mode optical fiber 4, a heating device 5 and a spectrometer 6; the broadband laser source 1 is connected to one end of the multimode optical fiber 3 through the first single-mode optical fiber 2, and the other end of the multimode optical fiber 3 is connected to the spectrometer 6 through the second single-mode optical fiber 4, and the heating device 5 is placed at the lower end of the multimode optical fiber 3. The heating device 5 can set a constant temperature to heat the multimode optical fiber 3, and multiple heating devices 5 can heat the multimode optical fiber 3 at different positions.

[0017] Specifically, a fiber fusion splicer is used to complete the fusion of the multimode fiber and the single-mode fibers at both ends, and then the two sections of the fiber are connected to a broadband laser source and a spectrometer, respectively, to form a complete spectral information collection device. A heating device is used to heat the temperature sensing unit, i.e., the multimode fiber, and the spectrum collected by the spectrometer will undergo corresponding changes, i.e., the required temperature information is hidden in the collected spectral information. Based on this principle, a heating device is used to quantitatively heat the multimode fiber at multiple locations of the multimode fiber, and the corresponding spectral information is collected and saved at the same time.

[0018] Specifically, the temperature variation range of the heating device is 30°C to 100°C, the variation step is 5°C, and there are 3 heating positions; the wavelength range collected by the spectrometer is set to 1510nm-1590nm, that is, the bandwidth is 80nm, and the wavelength collection resolution is 0.08nm, so the collected spectral information contains 1001 wavelengths and their corresponding light intensity information, and the save format is a CSV file.

[0019] Further, data preprocessing, the spectral information of the CSV format file collected in the above steps contains the configuration information of the spectrometer, which cannot be directly used for the training of the deep learning network model, and the data preprocessing step must be performed first; the embodiment of the present invention uses the Pycharm integrated development environment, uses the Python language to write programs, and batch preprocesses the CSV format files collected in the above steps. The specific operation is to import the os library for file and directory operations, and import the pandas library for data processing and analysis; specify the folder path containing the original CSV file and the new folder path for saving the processed CSV file; traverse all CSV format files in the original folder, delete the spectrometer configuration information therein, delete the wavelength information column, and retain the light intensity information column; finally, save the modified data to the specified new folder.

[0020] See also Figure 3 , an embodiment of the present invention provides a schematic diagram of the structure of a deep learning network model. A deep neural network can be regarded as a highly complex fitting function that can map high-dimensional raw inputs (such as images) to outputs (such as another image, the category of an object, or a measurement value). Training is usually performed by presenting many example inputs to the deep neural network and requiring it to optimize a loss function (i.e., the difference between the true value and the output of the deep neural network model). Once the loss function is minimized with a large number of example inputs, the model is considered to be trained and should be able to generalize to inputs that have never been seen during training. In an embodiment of the present invention, a deep neural network is trained using spectral information of a multimode optical fiber output under the influence of different temperatures and its corresponding temperature labels. In this case, the deep neural network is trained to accurately understand the relationship between the measured value and the spectral information while ignoring noise.

[0021] Specifically, see Figure 3 A schematic diagram of the structure of a deep learning neural network model provided in an embodiment of the present invention. In the embodiment of the present invention, the input is spectral data containing temperature change information. A multilayer perceptron (MLP) is used, which consists of 4 hidden layers, each of which has 512, 256, 128 and 64 nodes respectively. The activation function of each layer output is ReLU, and dropout is not used. In the embodiment of the present invention, the deep neural network model is used to complete the regression task, and the output value is the predicted value of the temperature at each point. In this regression case, the optimized loss function is the standard mean square error (MSE), and the fixed learning rate of the Adam optimizer is set to 0.001. All codes are written in Python using the Pycharm integrated development environment.

[0022] Specifically, to train and evaluate the MLP (Multi-layer Perceptron), the dataset is divided into three subsets: training set, validation set, and test set. The training set and validation set are used for training and cross-validation to fine-tune hyperparameters such as learning rate and batch size, while the test subset is used for evaluation purposes. The MLP is trained according to a standard iterative process, which has the following steps: (1) A small batch of spectra are randomly selected from the training set and the predicted temperature values ​​are generated through the MLP; (2) The MSE between the predicted temperature values ​​and the actual temperature values ​​is calculated; (3) The network weights are adjusted according to the standard back-propagation method to minimize the error; (4) When all the small batches in the training dataset are completed, that is, after one epoch, the training set is re-randomized and the next iterative training begins; (5) This process is repeated until the error or loss reaches a minimum.

[0023] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A distributed optical fiber temperature sensor based on deep learning, characterized in that: The sensor includes: a multimode optical fiber as a temperature sensing unit; a broadband laser source for outputting laser to one end of the multimode optical fiber; a first single-mode optical fiber for connecting the broadband laser source and the multimode optical fiber; a second single-mode optical fiber for connecting the other end of the multimode optical fiber to a spectrometer; a heating device, placed at the lower end of the multimode optical fiber, capable of setting a constant temperature to heat the multimode optical fiber; a spectrometer for collecting spectral information at different positions and temperatures of the multimode optical fiber; a deep learning neural network for receiving preprocessed spectral data and predicting the temperature value of each position, wherein the deep learning neural network is a multilayer perceptron (MLP) comprising four hidden layers, an activation function of ReLU, no dropout is used, the optimization loss function for the regression task is the mean square error (MSE), the optimizer is Adam, and the fixed learning rate is 0.

001.

2. The distributed optical fiber temperature sensor according to claim 1, characterized in that: The temperature variation range that can be set by the heating device of the multimode optical fiber is 30° C. to 100° C., the variation step is 5° C., and there are 3 heating positions.

3. The distributed optical fiber temperature sensor according to claim 1, characterized in that: The wavelength range collected by the spectrometer is set to 1510nm to 1590nm, the bandwidth is 80nm, the wavelength collection resolution is 0.08nm, and the collected spectrum information contains 1001 wavelengths and their corresponding light intensity information, and the storage format is a CSV file.

4. The distributed optical fiber temperature sensor according to claim 1, characterized in that: The data preprocessing step includes using Python language to batch read spectral data files in CSV format, deleting useless spectrometer configuration information and wavelength data columns, saving only the light intensity information column, and saving the processed data to a specified folder to facilitate network training.

5. The distributed optical fiber temperature sensor according to claim 1, characterized in that: The training process of the deep learning neural network includes training, validation and testing using the preprocessed spectral dataset, wherein the training and validation subsets are used to fine-tune hyperparameters such as learning rate and mini-batch size, and the test subset is used to evaluate the generalization ability of the model.

6. The distributed optical fiber temperature sensor according to claim 1, characterized in that: After training, the deep learning neural network model can predict the temperature value of the distributed optical fiber temperature sensor in real time, and the prediction error is considered accurate within a given range.

7. The distributed optical fiber temperature sensor according to claim 1, characterized in that: The sensor can be applied to the fields of industrial automation, environmental monitoring, power facility maintenance, etc. It has the advantages of non-invasiveness, resistance to electromagnetic interference, compact size and corrosion resistance, and is particularly suitable for high-precision temperature measurement in complex environments.

8. The distributed optical fiber temperature sensor according to claim 1, characterized in that: When the sensor trains the deep learning neural network model, the data set used is divided into three subsets: training, validation and testing. The training and validation subsets are used for network training and parameter optimization, and the testing subset is used for final model performance evaluation.

Images