Fiber optic angular velocity sensor facilitating assembly

Abstract

The present invention relates to the technical field of sensors, and particularly to an optical fiber angular velocity sensor that is convenient for assembly, including: a sensor data collection module for collecting the sensing data of the sensor; a data preprocessing module for preprocessing the sensing data to obtain preprocessed sensor data; a whale optimization algorithm module for obtaining the optimal design parameters by iteratively optimizing the design parameters of the sensor; a model training module for establishing a sensor performance prediction model using a deep belief network and predicting the performance indicators of the sensor according to the optimal design parameters; a result verification module for manufacturing a sensor sample according to the optimized design parameters and performing a performance test, and determining whether the performance test result is within a preset range. If so, the optimal performance indicators are output. The present invention obtains the optimal design parameters through the whale optimization algorithm and the deep belief network, enhancing the versatility and robustness of the sensor and enabling it to better adapt to various complex environments.

Description

Technical Field

[0001] The present invention relates to the technical field of sensors, and in particular to a fiber optic angular velocity sensor that is convenient for assembly. Background Art

[0002] A fiber optic angular velocity sensor is a high-precision angular velocity measurement device based on the optical principle. The working principle of the fiber optic angular velocity sensor is based on the Sagnac effect. When light enters the fiber optic loop from the light source, it will be reflected multiple times in the loop and form an interference loop.

[0003] Existing theoretical models cannot fully and accurately describe the behavior of the sensor in actual applications. Parameters such as the refractive index, loss, and nonlinear effects of the optical fiber may be affected by various factors in the actual environment, resulting in a deviation between the theoretical prediction and the actual performance. The characteristics of the optical fiber material, such as the thermal expansion coefficient and the refractive index temperature coefficient, may vary due to slight differences in the manufacturing process. These variations may lead to unstable sensor performance, making the prediction results inconsistent with the actual performance. The fiber optic angular velocity sensor is sensitive to changes in environmental factors such as temperature, humidity, and pressure. These changes may cause unstable sensor performance or even failure.

[0004] The present invention provides a fiber optic angular velocity sensor that is convenient for assembly, which can efficiently search the parameter space of the sensor structure and find the optimal or approximately optimal design solution. It enables the designer to more comprehensively understand the performance of the sensor under different environments, thereby making more reasonable decisions. Summary of the Invention

[0005] The present invention provides a fiber optic angular velocity sensor that is convenient for assembly to solve the defect of inaccurate performance prediction of the fiber optic angular velocity sensor in the prior art.

[0006] The present invention provides a fiber optic angular velocity sensor that is convenient for assembly, including: a sensor data collection module for collecting the sensing data of the sensor, and the sensing data includes design parameters and performance indicators.

[0007] A data preprocessing module for preprocessing the sensing data to obtain preprocessed sensor data.

[0008] A whale optimization algorithm module for obtaining the optimal design parameter combination by iteratively optimizing the design parameters of the sensor.

[0009] A model training module for establishing a sensor performance prediction model using a deep belief network and predicting the performance indicators of the sensor according to the optimal design parameter combination.

[0010] The result verification module manufactures sensor samples according to the optimized design parameters, conducts performance tests, and determines whether the performance test results are within the preset range. If so, it outputs the optimal performance indicators.

[0011] For the fiber optic angular velocity sensor provided by the present invention that is easy to assemble, in the sensor data collection module, the design parameters and performance indicators include:

[0012] Design parameters: shape, size, material properties, stress distribution, thermal deformation, strength.

[0013] Performance indicators: stability, resolution, accuracy, power consumption.

[0014] For the fiber optic angular velocity sensor provided by the present invention that is easy to assemble, the sensor preprocessed data includes:

[0015] Data cleaning, which is used to clean the sensed data and remove duplicate, incorrect, or abnormal data points.

[0016] Feature selection, which is used to extract features that affect the sensor performance from the sensed data.

[0017] Data normalization, which is used to normalize the design parameters and performance indicators on the same scale.

[0018] For the fiber optic angular velocity sensor provided by the present invention that is easy to assemble, the optimization process in the whale optimization algorithm module includes:

[0019] Set the number of combinations of design parameters to N, and determine the maximum number of iterations to T max , initialize the position of each parameter combination in the design parameter combination, denoted as X i (i = 1, 2, 3, …, n).

[0020] Evaluate the fitness value of each parameter combination, and select the n parameter individuals with the optimal fitness value as the initial population position. Record the optimal fitness value and its corresponding position at this time, and update the parameters a, A, C, I, p simultaneously.

[0021] Use the formula to update the position of the parameter combination to obtain the position update stage.

[0022] Judge whether the algorithm termination condition is reached. If so, output the optimal objective value; otherwise, repeat the position update stage until the termination condition is satisfied.

[0023] For the fiber optic angular velocity sensor provided by the present invention that is easy to assemble, the position update stage includes:

[0024] If p is less than 0.5 and |A| is less than 1, then update the individual position of the parameter combination to position one.

[0025] If p is less than 0.5 and |A| is greater than or equal to 1, then update the individual position of the parameter combination to Position Two.

[0026] If p is greater than or equal to 0.5, then update the individual position of the parameter combination to Position Three.

[0027] Recalculate the fitness value of the updated whale and update the current optimal solution.

[0028] According to the fiber optic angular velocity sensor provided by the present invention that is easy to assemble, in the position update stage, the update formula for Position One is:

[0029]

[0030] where p is the probability of state selection, t is the current iteration number, A and C are coefficients, X * (t) is the design parameter combination of the group of sensors with the optimal fitness currently, X(t) represents the design parameter combination in the current iteration, D is the distance between the current design parameter combination and the target parameter combination, I is a random value, and b is the parameter of the exponential function.

[0031] According to the fiber optic angular velocity sensor provided by the present invention that is easy to assemble, in the position update stage, the update formula for Position Two is:

[0032]

[0033] where X rand is the position of the random design parameter combination.

[0034] According to the fiber optic angular velocity sensor provided by the present invention that is easy to assemble, the model training module includes:

[0035] A model establishment unit for constructing a model based on the optimal design parameter combination, optimal parameters, and sensing data.

[0036] A model training unit for training the DBN model by combining unsupervised learning and supervised learning.

[0037] A model evaluation unit for evaluating the prediction performance of the model using a test set and comparing the performance differences between different models.

[0038] According to the fiber optic angular velocity sensor provided by the present invention that is easy to assemble, the model establishment unit includes:

[0039] Design the network structure of the DBN according to the optimal design parameter combination and sensing data. The network structure includes the number of nodes and connection methods of the input layer, hidden layer, and output layer.

[0040] The number of nodes in the input layer is the number of design parameter combinations.

[0041] The number of nodes in the output layer is the number of performance indicators.

[0042] The number of nodes in the hidden layer is determined by historical sensor data.

[0043] The fiber optic angular velocity sensor facilitating assembly provided by the present invention includes: The model training unit includes:

[0044] An unsupervised pre-training sub-unit, which is used to train the network layer by layer using the unsupervised learning method. First, train the first layer of RBM, and use the features learned by it as the input of the second layer of RBM. Then, train the second layer of RBM until all layers are trained.

[0045] A supervised fine-tuning sub-unit. After the unsupervised pre-training is completed, DBN uses the supervised learning method to fine-tune the entire network, and uses the backpropagation algorithm to adjust the weights and bias terms of the network to minimize the prediction error.

[0046] The fiber optic angular velocity sensor facilitating assembly provided by the present invention iteratively optimizes the design parameters of the sensor through the whale optimization algorithm to obtain the optimal design parameters, and uses the deep belief network to establish a sensor performance prediction model. According to the optimal design parameters, the performance indicators of the sensor are predicted, solving the problem of inaccurate performance prediction of the fiber optic angular velocity sensor. The beneficial effects obtained are:

[0047] Traditional sensor design methods often rely on a large number of experiments and trial and error. By using the whale optimization algorithm for simulation and emulation, the number of experiments can be greatly reduced, and the design cost and time cost can be reduced. The whale optimization algorithm has the characteristic of fast convergence and can find an approximate optimal solution in a short time. This enables designers to quickly iterate and optimize the sensor structure and improve the design efficiency.

[0048] The Whale Optimization Algorithm (WOA) can comprehensively simulate the working states of sensors under different assembly conditions, thus more accurately predicting the actual performance of sensors. Through simulation and emulation, potential performance issues can be identified and resolved during the design phase, thereby reducing the number and cost of experimental verifications. The algorithm can automatically search for the optimal combination of structural parameters, enabling the sensor to achieve the best balance in terms of stress, thermal deformation, strength, etc., thus improving the overall performance. The Whale Optimization Algorithm can automatically search for the optimal solution in the sensor structural parameter space without the need for manual trial one by one, greatly improving the design efficiency. Through simulation and emulation, the effects of different design schemes can be quickly evaluated, thereby accelerating the design iteration process and shortening the overall design cycle. The number and complexity of experimental verifications are reduced, and the material consumption and labor costs during the design process are lowered. This helps to evaluate the adaptability and stability of sensors under different environments. Through the optimization algorithm, the structural parameters of the sensor that perform well under different conditions can be found. This enhances the generality and robustness of the sensor, enabling it to better adapt to various complex environments.

[0049] The Deep Belief Network (DBN) can automatically learn meaningful feature representations from sensor data, which are crucial for subsequent performance prediction tasks. It avoids the uncertainty and difficulty of manual feature extraction, improving the accuracy and robustness of the prediction model. The DBN can model the data generation process. In sensor performance prediction, this helps to understand the distribution and variation patterns of sensor data, providing a more accurate basis for prediction. The DBN can learn and understand the data distribution and generate new data samples based on the learned model. When performing sensor performance prediction, this can simulate different scenarios and conditions to further verify and optimize the prediction model.

[0050] WOA can efficiently search for the optimal parameter configuration of the DBN, such as the number of network layers, the number of nodes in each layer, the learning rate, etc. This optimization process can significantly improve the performance of the DBN, making it more accurate and efficient in handling complex tasks. By optimizing the parameters of the DBN using WOA, a set of optimal parameter configurations can be obtained. This set of configurations not only performs well on the current dataset but also maintains good performance on unknown datasets. After combining the DBN and WOA, the generalization ability of the model is enhanced, enabling it to better adapt to different application scenarios and data distributions. The traditional DBN training process may involve a large amount of parameter adjustment and experimentation to find the best model configuration. As an efficient optimization algorithm, WOA can significantly reduce this trial-and-error process and accelerate model training. By quickly converging to the optimal solution, WOA enables the DBN to reach a stable performance level faster, thus saving training time and computing resources. Description of the Drawings

[0051] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0052] Figure 1 It is a module diagram of the fiber optic angular velocity sensor that is easy to assemble provided by an embodiment of the present invention. Detailed implementation manners

[0053] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0054] The following will describe in conjunction with Figure 1 the fiber optic angular velocity sensor that is easy to assemble of the present invention.

[0055] As Figure 1 shown, the fiber optic angular velocity sensor that is easy to assemble provided by an embodiment of the present invention includes: a fiber optic coil, a laser, a beam splitter, a signal detection system, and auxiliary components.

[0056] Fiber optic coil: The fiber optic coil is composed of multiple turns of single-pass optical fiber wound around a skeleton. When the system has an angular velocity relative to the Earth inertial system, two beams of light propagating in the fiber optic coil in the clockwise and counterclockwise directions respectively will generate an optical path difference or a phase difference, that is, the Sagnac effect. The fiber optic connector is a standardized interface that can provide mechanical stability and optical alignment, thus ensuring low loss of optical signals during transmission. These connections are usually achieved through fiber optic connectors or fiber optic adapters to ensure efficient transmission of optical signals.

[0057] Laser: It is used to emit a collimated beam, which is split into two beams of equal intensity by the beam splitter. The performance of the laser directly affects the measurement accuracy and stability of the sensor.

[0058] Beam splitter: The beam splitter is usually an optical element, such as a semi-transparent and semi-reflective mirror or a fiber optic coupler, used to split the beam emitted by the laser into two beams of equal intensity and guide them to the two ends of the fiber optic coil respectively, so as to propagate in the fiber optic coil in the clockwise and counterclockwise directions.

[0059] Focusing lens: By placing the lens near the end of the fiber optic coil and adjusting the position and focal length of the lens, the focusing lens focuses the two split light beams onto the end face of the fiber optic coil to ensure that light can efficiently enter the fiber optic coil and propagate along it. The focusing lens can improve the efficiency of the light signal entering the fiber optic coil and reduce the loss and reflection of the light signal.

[0060] Signal detection system: Used to detect the phase difference or optical path difference between the two light beams propagating out of the fiber optic coil and convert it into an electrical signal for further processing and analysis. The signal detection system usually includes components such as a photodetector and a signal processing circuit. The signal detection system includes components such as a photodetector and a signal processing circuit. The photodetector is used to convert the optical signal into an electrical signal, and the signal processing circuit is used to amplify, filter, and digitize the electrical signal for further analysis and calculation.

[0061] Other auxiliary components: The fiber optic angular velocity sensor also includes other auxiliary components such as fiber optic connectors, adapters, and housings. The auxiliary components are used to protect the optical and electronic components inside the sensor from external environmental interference and damage and facilitate installation and integration into a larger system or device.

[0062] The laser is connected to the beam splitter, and the two ends of the fiber optic coil are respectively connected to the two output ports of the beam splitter. When the two light beams propagate in the fiber optic coil, they will return to the beam splitter again, and the light signal after interference here is then guided to the signal detection system. The fiber optic connector is electrically connected to the adapter, and the housing is threadedly connected to the front connector.

[0063] Sensor data collection module, used to collect the sensing data of the sensor. The sensing data includes design parameters and performance indicators.

[0064] Sensor performance prediction and optimization aims to improve its performance indicators by adjusting the design parameters of the sensor. The design parameters of the sensor include, for example, shape, size, material properties, stress distribution, thermal deformation, and strength. The performance indicators include stability, resolution, accuracy, power consumption, etc.

[0065] Data preprocessing module, used to preprocess the sensor sensing data to obtain the preprocessed sensor data. The preprocessed sensor data includes: data cleaning, feature selection, and data normalization.

[0066] Data cleaning, used to clean the collected sensing data and remove duplicate, incorrect, or abnormal data points.

[0067] Feature selection: Extract features from the original data that have a significant impact on the sensor performance, which helps to reduce the complexity of the model and improve the prediction accuracy.

[0068] Data normalization: Normalize or standardize the design parameters and performance indicators to ensure they are on the same scale, which helps machine learning algorithms learn and predict better.

[0069] Whale optimization algorithm optimization module, which obtains the optimal design parameter combination by iteratively optimizing the design parameters of the sensor.

[0070] Use the whale optimization algorithm to simulate and model the structure of the sensor to predict and optimize its performance. By simulating the stress distribution, thermal deformation, strength, stability, accuracy, and resolution under different assembly conditions, find the optimal structure design scheme.

[0071] The whale optimization algorithm uses random or optimized search agents to simulate the predation strategy of whales in nature, and then realizes the exploration and optimization of the search space. It uses a spiral model to simulate the attack mechanism of whales, which enables the algorithm to explore the solution space more flexibly and comprehensively during the search process. Different positions in the predation process of the whale population represent different feasible solutions. This is like the process of finding the optimal solution in the solution space. Each iteration is like a whale swimming one step in the ocean. By continuously adjusting the position, the global optimal solution is finally found. There are three search methods in this algorithm, namely encircling prey, bubble net hunting, and searching for prey.

[0072] In the initial stage of encircling prey, the design parameters of a group of sensors represent a feasible solution. Since the position of the optimal individual is unknown in the search space, it is necessary to first determine the position of the prey for encirclement. The whale optimization algorithm uses the currently best potential solution as the target object and adopts a siege strategy to update its search position.

[0073] The formula for encircling prey is expressed as:

[0074]

[0075] In the formula, t is the current iteration number, A and C are coefficients, X * (t) is the design parameters of the group of sensors with the best fitness currently, X(t) represents the design parameters in the current iteration, and D is the distance between the current design parameters and the target parameters. If there is a better candidate solution, the value of X*(t) is updated in each iteration. The specific calculation formulas for A and C are expressed as:

[0076]

[0077] In the formula, r1 and r2 are random values in the interval (0, 1), and during the entire iteration process, the value of parameter a will linearly decrease from 2 to 0, and this change process runs through the exploration stage and exploitation stage of the algorithm. t represents the current iteration number, and T max is the maximum iteration number set initially.

[0078] During the hunting stage, whales hunt their prey by shrinking the encirclement and releasing bubbles to form a bubble net in a spiral rotation. That is, when humpback whales make a spiral movement, they will simultaneously shrink the encirclement to ensure that they can effectively contract and surround the prey. According to the calculation formula of A, the value of A at this time is randomly selected within the interval [-a, a]. When the random value range of A is limited to [-1, 1], the updated design parameter X(t + 1) will fluctuate between the current design parameter X(t) and the optimal design parameter X*(t), thus achieving an effective enclosure of the target parameter. The formula expression for updating the current design parameter is as follows:

[0079]

[0080] In the formula, is the distance between the design parameter and the target design parameter, b is a constant, defined as the logarithmic spiral form presented when surrounding the prey, I is a random value, and its value range is between [-1, 1], and b is the parameter of the exponential function.

[0081] When whales hunt their prey, the whales carry out the surrounding of the prey and the spiral ascent simultaneously. It is assumed that the state selection probabilities between the surrounding mechanism and the spiral ascent are both 0.5. The surrounding formula is expressed as:

[0082]

[0083] In the formula, p is the probability of state selection, and the random range is [0, 1]. By adjusting the probability p, the parameter a, and the variation range of A, the design parameter migrates to a better parameter. When the number of iterations increases continuously, the parameters A and the convergence factor a in the above formula will continuously decrease until |A| < 1, indicating that the current local optimal solution is obtained.

[0084] Searching for prey stage: In order to enable the whale population to search more fully in the solution space, the algorithm updates the position based on the distances between different whale individuals to ensure random search. During the search process, whale individuals tend to approach randomly selected conspecifics and use their position information to update their own positions. This method actually guides the whales away from the current position of the prey, thereby enhancing the global optimization ability of the algorithm.

[0085] The mathematical model is expressed as:

[0086]

[0087] In the formula, D is the distance between the current design parameter and the randomly selected design parameter, X randis the position of the random design parameters. When |A| ≥ 1, the algorithm randomly selects a design parameter and updates the positions of other design parameters accordingly, aiming to increase the randomness of the search and help discover better-quality prey.

[0088] When the maximum number of iterations is reached or the termination condition is satisfied, the algorithm iteration stops. The current optimal solution is output, that is, the optimal sensor design scheme.

[0089] The model training module uses a deep belief network to establish a sensor performance prediction model and predicts the performance indicators of the sensor according to the optimal design parameters.

[0090] The model building unit is used to build a model according to the optimal design parameter combination and sensing data.

[0091] Network structure design: Design the network structure of the DBN according to the optimal design parameters and sensing data, including the number of nodes and connection methods in the input layer, hidden layer, and output layer. The number of input layer nodes corresponds to the number of design parameters, and the number of output layer nodes corresponds to the number of performance indicators. The number of hidden layers and nodes needs to be analyzed and determined through experiments to find the optimal network structure.

[0092] The model training unit trains the DBN model by combining unsupervised learning and supervised learning.

[0093] Unsupervised pre-training: Use unsupervised learning methods to train the network layer by layer. First, train the first-layer RBM, and use the features it learns as the input of the second-layer RBM, then train the second-layer RBM, and so on until all layers are trained. This process enables each layer of the RBM to learn an effective representation of the input data. Use the contrastive divergence algorithm to train the RBM layer by layer to ensure that the feature vectors retain as much feature information as possible when mapped to different feature spaces.

[0094] Supervised fine-tuning: After the unsupervised pre-training is completed, the DBN uses supervised learning methods to fine-tune the entire network. In this stage, the backpropagation algorithm is usually used to adjust the weights and bias terms of the network to minimize the prediction error. Through fine-tuning, the DBN can further improve its performance on specific tasks. Use the backpropagation algorithm for global parameter optimization to adjust the weights and bias terms of the entire network to minimize the prediction error.

[0095] The model evaluation unit is used to evaluate the prediction performance of the model using the test set and compare the performance differences between different models. Use techniques such as cross-validation to evaluate the model to select the best combination of hyperparameters.

[0096] The result verification module manufactures sensor samples according to the optimized design parameters and conducts performance tests to determine whether the performance test results are within the preset range. If so, it outputs the optimal performance indicators.

[0097] The fiber optic angular velocity sensor provided by the present invention is easy to assemble. By iteratively optimizing the design parameters of the sensor through the whale optimization algorithm to obtain the optimal design parameters, and using the deep belief network to establish a sensor performance prediction model, and predicting the performance indicators of the sensor according to the optimal design parameters, the problem of inaccurate performance prediction of the fiber optic angular velocity sensor is solved. The beneficial effects obtained are as follows:

[0098] WOA can efficiently search for the optimal parameter configuration of DBN, such as the number of network layers, the number of nodes in each layer, the learning rate, etc. This optimization process can significantly improve the performance of DBN, making it more accurate and efficient in dealing with complex tasks. By optimizing the parameters of DBN through WOA, a set of optimal parameter configurations can be obtained. This set of configurations not only performs well on the current dataset but also maintains good performance on unknown datasets. This means that after the combination of DBN and WOA, the generalization ability of the model is enhanced, and it can better adapt to different application scenarios and data distributions. The traditional DBN training process may involve a large number of parameter adjustments and experiments to find the best model configuration. As an efficient optimization algorithm, WOA can significantly reduce this trial-and-error process and accelerate model training. By quickly converging to the optimal solution, WOA enables DBN to reach a stable performance level faster, thus saving training time and computing resources.

[0099] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, located in one place or distributed to multiple network units. Select some or all of the modules according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0100] Through the description of the above embodiments, those skilled in the art clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on such an understanding, the above technical solutions are essentially or the part that contributes to the prior art is embodied in the form of a software product. This computer software product is stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (personal computer, server, or network device, etc.) to execute the methods of each embodiment or some parts of the embodiments.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fiber optic angular velocity sensor facilitating assembly, characterized in that, Including: A sensor data collection module, which is used to collect the sensing data of the sensor. The sensing data includes design parameters and performance indicators. Among them, the design parameters are: shape, size, material properties, stress distribution, thermal deformation, strength; the performance indicators are: stability, resolution, accuracy, power consumption; A data preprocessing module, which is used to preprocess the sensing data to obtain preprocessed sensor data; A whale optimization algorithm module, which is used to obtain the optimal design parameter combination by iteratively optimizing the design parameters of the sensor; A model training module, which uses a deep belief network to establish a sensor performance prediction model and predicts the performance indicators of the sensor according to the optimal design parameter combination; A result verification module, which manufactures a sensor sample according to the optimized design parameters and conducts a performance test, and judges whether the performance test result is within the preset range. If so, it outputs the optimal performance indicator; Among them, the model training module includes: A model establishment unit, which is used to construct a model according to the optimal design parameter combination and the sensing data; A model training unit, which trains the DBN model by combining unsupervised learning and supervised learning; A model evaluation unit, which is used to evaluate the prediction performance of the model using a test set and compare the performance differences between different models.

2. The fiber optic angular velocity sensor facilitating assembly according to claim 1, wherein The preprocessed sensor data includes: Data cleaning, which is used to clean the sensing data and remove duplicate, incorrect or abnormal data points; Feature selection, which is used to extract features that affect the sensor performance from the sensing data; Data normalization, which is used to normalize the design parameters and performance indicators on the same scale.

3. The fiber optic angular velocity sensor facilitating assembly according to claim 1, characterized in that, The model establishment unit includes: Design the network structure of the DBN according to the optimal design parameter combination and the sensing data. The network structure includes the number of nodes and connection methods of the input layer, hidden layer and output layer; The number of nodes in the input layer is the number of design parameter combinations; The number of nodes in the output layer is the number of performance indicators; The number of nodes in the hidden layer is determined by historical sensor data.

4. The fiber optic angular velocity sensor facilitating assembly according to claim 1, wherein, The model training unit includes: An unsupervised pre-training subunit, which is used to train the network layer by layer using an unsupervised learning method. First, train the first layer of RBM, and use the features learned by it as the input of the second layer of RBM, and then train the second layer of RBM until all layers are trained; A supervised fine-tuning subunit. After the unsupervised pre-training is completed, the DBN uses a supervised learning method to fine-tune the entire network, and uses the backpropagation algorithm to adjust the weights and bias terms of the network to minimize the prediction error.

Images