Traditional Back Propagation Research Articles

Prediction of the corrosion depth of oil well cement corroded by carbon dioxide using GA-BP neural network

A corrosion prediction model was established based on the genetic algorithm (GA) and back propagation (BP) neural network to predict the long-term corrosion changes of oil well cement, Considering that the cement sheath is susceptible to corrosion and its corrosion degree is not easy to observe in acid gas wells and geological storage wells containing carbon dioxide (CO2). The initial weights and thresholds of the neural network were optimized by GA. The number of hidden layer nodes was selected by error verification, and the network was trained with an improved algorithm. The sample data was regression processed based on the empirical formula and was used in the network training. The simulation results shows that: The improved GA-BP network model (3–5-6–1) has a higher prediction accuracy with faster convergence and better fitting effect compared with the traditional BP neural network and the regression model (REG) in long-term prediction of corrosion depth in oil well cement. The regression coefficient (R2) of the prediction model is 0.9913, and the mean square error (MSE) of test samples is 0.0026. The modeling idea proposed in this paper can be applied to improve the accuracy of prediction models in predicting the corrosion of oil well cement.

Developing a Hybrid Algorithm Based on an Equilibrium Optimizer and an Improved Backpropagation Neural Network for Fault Warning

In today’s rapidly evolving manufacturing landscape with the advent of intelligent technologies, ensuring smooth equipment operation and fostering stable business growth rely heavily on accurate early fault detection and timely maintenance. Machine learning techniques have proven to be effective in detecting faults in modern production processes. Among various machine learning algorithms, the Backpropagation (BP) neural network is a commonly used model for fault detection. However, due to the intricacies of the BP neural network training process and the challenges posed by local minima, it has certain limitations in practical applications, which hinder its ability to meet efficiency and accuracy requirements in real-world scenarios. This paper aims to optimize BP networks and develop more effective fault warning methods. The primary contribution of this research is the proposal of a novel hybrid algorithm that combines a random wandering strategy within the main loop of an equilibrium optimizer (EO), a local search operator inspired by simulated annealing, and an adaptive learning strategy within the BP neural network. Through analysis and comparison of multiple sets of experimental data, the algorithm demonstrates exceptional accuracy and stability in fault warning tasks, effectively predicting the future operation of equipment and systems. This innovative approach not only overcomes the limitations of traditional BP neural networks, but also provides an efficient and reliable solution for fault detection and early warning in practical applications.

An improved filtering algorithm for indoor localization based on DE-PSO-BPNN

Among the indoor localization algorithms, the algorithm based on traditional Back Propagation Neural Network (BPNN) has the problems of slow convergence and easy to fall into local optimum. It is difficult to apply the algorithm in noisy environments. Therefore, in this paper, we propose a novel indoor localization algorithm where the whole localization process is divided into two parts: data preprocessing and localization output. Data preprocessing means using filtering algorithm to process the Received Signal Strength Indication (RSSI) sequence. It is considered that the initial value of the received sequence has a significant impact on the performance of Kalman Filter (KF). An improved Kalman Filtering algorithm (DBSCAN-KF) is proposed based on the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm. First, the RSSI values that are seriously disturbed by noise in the sequence are removed using the DBSCAN algorithm, and then the RSSI sequences are processed using KF so that the RSSI values can be closer to the theoretical values. The localization output part is to reduce the localization error caused by the BPNN. In this paper, the Differential Evolution (DE) algorithm and Particle Swarm Optimization (PSO) algorithm are combined, and the Differential Evolution Particle Swarm Optimization (DE-PSO) algorithm is proposed. The BPNN weights and thresholds are optimized in parallel, which improves the speed and ability of global optimization search and further avoids the shortcomings of traditional BPNNs that are prone to fall into local optimization in the training process. Experimental results show that the BPNN localization algorithm based on DBSCAN-KF improves the average localization accuracy by 0.26m compared with the BPNN localization algorithm without filtering. After filtering, the localization algorithm based on DE-PSO improved BPNN (DE-PSO-BP) improves the average localization accuracy by about 24% compared with the localization algorithm based on DE-PSO-BP. The localization algorithm based on DE-PSO-BP improves the average localization accuracy by about 61% compared with the traditional BPNN.

A method of using back propagation neural network to estimate orbital lifetime of LEO satellites

The “IADC space debris mitigation guidelines” requires that the post-mission orbit lifetime of low Earth orbiting (LEO) satellites should not be longer than 25 years, which is the key to debris mitigation. To analyze the degree of global compliance to this 25-year guideline, a fast and relatively accurate orbit life estimation algorithm is required. The traditional method uses numerical orbit propagation to accomplish the task, although its accuracy is high when a high-precision force model is used, the calculation is time-consuming. This paper proposes a new method for orbit lifetime estimation without relying on orbit propagation. A back propagation (BP) neural network is used to directly fit the relationship from the satellite parameters to the orbit lifetime. The space objects that have reentered the Earth’s atmosphere are selected from the catalog database of the US Space Surveillance Network (SSN), and are used to generate samples for network training. The genetic algorithm is combined with the traditional back propagation algorithm for the training of the neural network. The results show that for the specific application background of identifying whether a satellite complies with the 25-year guideline, the orbit lifetime estimation accuracy of the BP network meets the demand to some extent. Moreover, because it does not involve orbit propagation, the BP network has a high computational efficiency in estimating orbit lifetime, showing a good application prospect.

Prediction of Ultimate Bearing Capacity of Pile Foundation Based on Two Optimization Algorithm Models

The determination of the bearing capacity of pile foundations is very important for their design. Due to the high uncertainty of various factors between the pile and the soil, many methods for predicting the ultimate bearing capacity of pile foundations focus on correlation with field tests. In recent years, artificial neural networks (ANN) have been successfully applied to various types of complex issues in geotechnical engineering, among which the back-propagation (BP) method is a relatively mature and widely used algorithm. However, it has inevitable shortcomings, resulting in large prediction errors and other issues. Based on this situation, this study was designed to accomplish two tasks: firstly, using the genetic algorithm (GA) and particle swarm optimization (PSO) to optimize the BP network. On this basis, the two optimization algorithms were improved to enhance the performance of the two optimization algorithms. Then, an adaptive genetic algorithm (AGA) and adaptive particle swarm optimization (APSO) were used to optimize a BP neural network to predict the ultimate bearing capacity of the pile foundation. Secondly, to test the performance of the two optimization models, the predicted results were compared and analyzed in relation to the traditional BP model and other network models of the same type in the literature based on the three most common statistical indicators. The models were evaluated using three common evaluation metrics, namely the coefficient of determination (R2), value account for (VAF), and the root mean square error (RMSE), and the evaluation metrics for the test set were obtained as AGA-BP (0.9772, 97.8348, 0.0436) and APSO-BP (0.9854, 98.4732, 0.0332). The results show that compared with the predicted results of the BP model and other models, the test set of the AGA-BP model and APSO-BP model achieved higher accuracy, and the APSO-BP model achieved higher accuracy and reliability, which provides a new method for the prediction of the ultimate bearing capacity of pile foundations.

Deep learning of electromechanical impedance for concrete structural damage identification using 1-D convolutional neural networks

Common damages in concrete materials and structures are usually in small sizes at initial stage, which induce small stiffness and mass loss being difficult to evaluate severity level merely depending on traditional electromechanical admittance (EMA, inverse of impedance) technique. In fact, several state-of-the-art computerized techniques have been incorporated with the EMA for automated evaluation of concrete damages. However, complicated data preprocessing still limits the efficiency of these techniques in terms of accuracy and computational cost. To this end, this paper proposed a novel deep learning approach using one-dimensional (1-D) convolutional neural networks (CNNs) for exploiting the raw EMA signatures to automatically identify tiny damages in concrete structures, which eliminated tedious data preprocessing for network training and testing. Two independent EMA databases measured by smart piezoelectric sensors were established based on proof-of-concept experiment of slight/severe mass-loss damage detection on a concrete cube, as well as practical application of bolt-looseness identification on a full-scaled shield tunnel segment assembled structure. For both tests, effect of varied dimensions of input data on the efficiency of CNN models were evaluated as well. The well-trained CNN model perfectly quantified the severity degrees of mass-loss and bolt-looseness damages, which demonstrated significant superiority than traditional back propagation neural network and the model with less dimensions of input data exhibited higher accuracy. Delightful results in this work potentially provided a potential paradigm of EMA data-driven tiny damage identification for real-life concrete infrastructures.

Deep Retrieval Architecture of Temperature and Humidity Profiles from Ground-Based Infrared Hyperspectral Spectrometer

Temperature and humidity profiles in the atmospheric boundary layer are essential for climate studies. The ground-based infrared hyperspectral spectrometer has the advantage of measuring radiances emitted from the atmosphere at a high temporal and moderate vertical resolution. In this article, the retrieval of temperature and humidity profiles from ground-based infrared hyperspectral observations is exploited. Although existing inversion algorithms based on physical models or statistical learning have made some progress, they still suffer from high computational complexity or poor performance. Motivated by the strength of the deep learning, we present a deep retrieval architecture (DReA) by skillfully designing a light-weight one-dimensional convolution neural network (CNN) to retrieve the temperature and humidity profiles. Experiments were conducted using atmospheric emitted radiance interferometer (AERI) and radiosonde data to demonstrate the superiority of the proposed DReA. The validation of the DReA with the radiosonde, using 802 profiles with 37 layers below 3 km, presents an excellent retrieval ability with a root mean square error (RMSE) of 0.87 K for the temperature and 1.06 g/kg for the water vapor mixing ratio. Furthermore, a thorough comparison with commonly used inversion methods such as the traditional back propagation (BP) and the eigenvector (EV) regression method, shows that our proposed DReA method obtains a leading solution in retrieving temperature and humidity profiles.

Novel Cuckoo Search-Based Metaheuristic Approach for Deep Learning Prediction of Depression

Depression is a common illness worldwide with doubtless severe implications. Due to the absence of early identification and treatment for depression, millions of individuals worldwide suffer from mental illnesses. It might be difficult to identify those who are experiencing mental health illnesses and to provide them with the early help that they need. Additionally, depression may be associated with thoughts of suicide. Currently, there are no clinically specific diagnostic biomarkers that can identify the severity and type of depression. In this research paper, the novel particle swarm-cuckoo search (PS-CS) optimization algorithm is proposed instead of the traditional backpropagation algorithm for training deep neural networks. The backpropagation algorithm is widely used for supervised learning in deep neural networks, but it has limitations in terms of convergence speed and the possibility of getting trapped in local optima. These problems were addressed by using a deep neural network architecture for depression detection tasks along with the PS-CS optimization technique. The PS-CS algorithm combines the strengths of both particle swarm optimization and cuckoo search algorithms, which allows for a more efficient and effective optimization of the network parameters. We also evaluated how well the suggested methods performed against the most widely used classification models, including (K-nearest neighbor) KNN, (support vector regression) SVR, and decision trees, as well as the most widely used deep learning models, including residual neural network (ResNet), visual geometry group (VGG), and simple neural network (LeNet). The findings show that the suggested method, PS-CS, in conjunction with the CNN model, outperformed all other models, achieving the maximum accuracy of 99.5%. Other models, such as the KNN, decision trees, and logistic regression, achieved lower accuracies ranging from 69% to 97%.

A reliability-oriented genetic algorithm-levenberg marquardt model for leak risk assessment based on time-frequency features

Since leaks in high-pressure pipelines transporting crude oil can cause severe economic losses, a reliable leak risk assessment can assist in developing an effective pipeline maintenance plan and avoiding unexpected incidents. The fast and accurate leak detection methods are essential for maintaining pipeline safety in pipeline reliability engineering. Current oil pipeline leakage signals are insufficient for feature extraction, while the training time for traditional leakage prediction models is too long. A new leak detection method is proposed based on time-frequency features and the Genetic Algorithm-Levenberg Marquardt (GA-LM) classification model for predicting the leakage status of oil pipelines. The signal that has been processed is transformed to the time and frequency domain, allowing full expression of the original signal. The traditional Back Propagation (BP) neural network is optimized by the Genetic Algorithm (GA) and Levenberg Marquardt (LM) algorithms. The results show that the recognition effect of a combined feature parameter is superior to that of a single feature parameter. The Accuracy, Precision, Recall, and F1score of the GA-LM model is 95%, 93.5%, 96.7%, and 95.1%, respectively, which proves that the GA-LM model has a good predictive effect and excellent stability for positive and negative samples. The proposed GA-LM model can obviously reduce training time and improve recognition efficiency. In addition, considering that a large number of samples are required for model training, a wavelet threshold method is proposed to generate sample data with higher reliability. The research results can provide an effective theoretical and technical reference for the leakage risk assessment of the actual oil pipelines.

Hybrid Simulated Annealing Particle Swarm Optimization Support Vector Machine Based Temperature-Pressure Error Compensation Approach for TDLAS Gas Detection

ABSTRACT The target gas concentration in TDLAS (Tunable Diode Laser Absorption Spectroscopy) gas detection is nonlinearly related to the photodetector output signal, which is effortlessly disturbed by the temperature and pressure inside the gas absorption cell, and it will decrease the detection accuracy of the system. This paper proposes an SA-PSO-SVM (Simulate Anneal-Particle Swarm Optimization-Support Vector Machine) based temperature-pressure error compensation approach to improve the accuracy of TDLAS gas detection. First, an experimental system concerning the effects of gas temperature-pressure on the absorption line distortion was built. Meanwhile, a method for processing and exploring normalized data about the temperature and pressure effects on the gas concentration was proposed. Second, an SA-PSO-SVM-based model of temperature-pressure compensation for TDLAS gas detection was constructed, and a multi-point correction was performed. Finally, the developed model was compared with traditional BP (Back Propagation), SVM, and PSO-SVM-based temperature-pressure compensation models for TDLAS gas detection, to verify the effectiveness in terms of mean absolute error, mean relative error, and mean square error, which could significantly improve the detection accuracy of TDLAS system. The results provide guidance and lay a technical foundation for the study of multi-component gas cross-aliasing absorption line separation.

Improving performance prediction of evacuated tube solar collector through convolutional neural network method

The Convolutional Neutral Network (CNN) method is used for predicting the performance of an all-glass straight-through evacuated tube solar collector based on operational data collected over several days. The input layer of the CNN-model includes the mass flow rate, inlet temperature, solar irradiance, wind speed and outdoor temperature. The temperature difference of the collector fluid inlet and outlet, collected heat and efficiency of the collector are the main output. The CNN-model outperforms a Multiple Linear Regression (MLR) and Back Propagation (BP) neural network model by having the best prediction accuracy with the lowest Root Mean Squared Error (RMSE = 0.00577), the lowest Mean Absolute Error (MAE = 0.00357) and the highest value of Coefficient of Determination (R2 = 0.9534) for predicting the collector efficiency, and the same also applied to the temperature difference and useful solar heat. The BP-model performed better than the MLR model. MLR is therefore not recommended for performance evaluation of vacuum tube solar collectors due its poor capability to handle non-linear problems leading to lower accuracy. The prediction precision of CNN outperformed BP. A traditional BP algorithm employs a large number of weights, it is calculation intensive, and is more complex than the CNN algorithm. The CNN-model showed that solar collector performance is more affected by the solar irradiance and mass flow rate than by the wind speed and outdoor temperature.

Optimization of solid oxide fuel cell power generation voltage prediction based on improved neural network

Abstract This paper proposes a method for predicting the generation voltage of a solid oxide fuel cell based on the data results of a stand-alone solid oxide fuel single cell simulation model under ideal conditions, with the aim of improving the generation efficiency and extending the service life of the solid oxide fuel cell. In this paper, a modified back propagation (BP) neural network algorithm is used to improve the prediction accuracy of the solid oxide fuel cell generation voltage by using the whale algorithm to optimize the BP neural network model to improve its convergence and achieve the effect of improving the prediction accuracy. First, the characteristics of the independent solid oxide fuel cell are introduced and simulated. Second, the long short-term memory network model, linear regression network model and BP neural network are simulated and compared, and the results show that the BP neural network prediction model is more accurate and can be optimized and improved. Finally, the BP neural network is optimized and simulated using the whale algorithm, and the simulation results show that the method has better convergence and higher prediction accuracy than the traditional BP neural network prediction model.

Surface Feature Prediction for Laser Ablated 40Cr13 Stainless Steel Based on Extreme Learning Machine.

Determining an optimal combination of laser process parameters can significantly improve the efficiency and quality of 40Cr13 steel surface processing. In this study, two machine learning models (ELMSS and ELMPS) were proposed to predict the processing results of surface features to optimize process parameters. The prediction accuracies of the proposed models were always higher than those of traditional back propagation (BP) and radial basis function (RBF) neural networks, and the calculation time of the proposed models was significantly reduced. In comparison, the prediction accuracy ranking for ablation depth was ELMSS (92.6%), BP (89.8%), and RBF (89.6%), and for the ablation width, it was ELMSS (98.3%), BP (97.4%), and RBF (96.1%). The material removal rate was 92.4%, 91.1%, and 89.1% for ELMSS, BP, and RBF, respectively. Finally, the prediction accuracy ranking for surface roughness was 86.8%, 80.7%, and 79.5% for ELMPS, BP, and RBF, respectively. After optimization by the genetic algorithm, the prediction accuracies of the proposed models for the depth, width, material removal rate, and surface roughness reached 94.0%, 99.0%, 93.2%, and 91.2%, respectively. With the support of ELMSS and ELMPS, the results of the surface features can be predicted before machining and the appropriate process parameters can be selected in advance.

Forecasting the gross domestic product using a weight direct determination neural network

<abstract><p>One of the most often used data science techniques in business, finance, supply chain management, production, and inventory planning is time-series forecasting. Due to the dearth of studies in the literature that propose unique weights and structure (WASD) based models for regression issues, the goal of this research is to examine the creation of such a model for time-series forecasting. Given that WASD neural networks have been shown to overcome limitations of traditional back-propagation neural networks, including slow training speed and local minima, a multi-function activated WASD for time-series (MWASDT) model that uses numerous activation functions, a new auto cross-validation method and a new prediction mechanism are proposed. The MWASDT model was used in forecasting the gross domestic product (GDP) for numerous nations to show off its exceptional capacity for learning and predicting. Compared to previous WASD-based models for time-series forecasting and traditional machine learning models that MATLAB has to offer, the new model has produced noticeably better forecasting results, especially on unseen data.</p></abstract>

Application of carbon emission prediction based on a combined neural algorithm in the control of coastal environmental pollution in China

The marine ecosystem provides the environment, resources, and services necessary for the development of every human society. In recent years, China's coastal zone has been polluted to varying degrees, which has seriously affected its development. The characteristics of marine environmental data include the variety of data types, the complexity of factors affecting the marine environment, and the unpredictability of marine pollution. Currently, there are few studies applying the clustering analysis algorithm to marine environmental monitoring. Then, carbon emissions (CEs) from coastal areas are predicted using marine environmental data. Therefore, this paper mainly studies the spatial and temporal accumulation characteristics of marine environmental data and uses the fuzzy c-means (FCM) algorithm to mine the data monitored by the marine environment. Meanwhile, it has been focused on the prediction of coastal CEs, and the grey model-back propagation (GM-BP) algorithm has been developed to predict CEs from coastal areas, which solves the problem that the traditional back propagation neural network (BPNN) cannot fully learn data features, which leads to a decline in accuracy. The experimental results showed that the FCM algorithm can divide the marine sample data into corresponding categories to distinguish polluted and unpolluted samples. The improved neural network model has a higher degree of non-linear fit and lower prediction error than a back propagation (BP) neural network. The main contribution of this paper is to first study the spatial and temporal accumulation characteristics of marine environmental data. The academic contribution of this study is to substitute the predictions of the three gray models (GMs) with the neural network structure simulation to finally obtain more accurate predictions. From a practical point of view, this study is helpful to a certain extent in alleviating the pressure of climate change due to increased CEs in global coastal zones. This study can also provide a new method of measuring environmental governance for marine environmental regulatory authorities.

Research on the Prediction Method of Monthly Hidden Danger Quantity in Coal Mine Based on BP Neural Network Periodic Combination Model

To better prevent the occurrence of hidden dangers of coal mine accidents and ensure the safety production of coal mine enterprises. This paper mines and analyses the pattern of historical monthly hidden danger quantity in the coal mine and constructs three models: the traditional backpropagation (BP) neural network model, the BP neural network based on the adaptive moment estimation optimization algorithm (Adam-BP) model, and the BP neural network prediction model with the introduction of monthly moderators (Month-Adam-BP). The experimental results show that the Adam-BP model can improve the prediction accuracy, in which the mean absolute percentage error (MAPE) improves by 8.93%, the root mean square error (RMSE) improves by 8.15%, the postdifference ratio C improves by 0.04, and the small error probability P improves by 0.12; the Month-Adam-BP model with the introduction of the monthly adjustment factor further improves the prediction accuracy, in which MAPE improves by 2.61%, RMSE improves by 5.41%, the postdifference ratio C improves by 0.06, and the small error probability P improves by 0.03. And the Month-Adam-BP model prediction accuracy reaches the level 2 standard with credible prediction effect; it can also be used to predict coal mines with periodic characteristics of hidden hazard data. Our prediction results show that the predicted number of hidden hazards in this coal mine for the next month is 29, which is an increase compared to the number of hidden dangers in the previous month. Thus, the coal mine safety managers need to strengthen the management of hidden hazards further to prevent accidents, which can better serve the standardization of coal mine safety production and ensure the smooth production of the coal mine.

The formulation of irrigation and nitrogen application strategies under multi-dimensional soil fertility targets based on preference neural network

With the aim of improving soil fertility, it is of great significance to put forward optimal irrigation and nitrogen fertilizer application strategies for improving land productivity and alleviating non-point source pollution effects. To overcome this task, a 6-hidden layer neural network with a preference mechanism, namely Preference Neural network (PNN), has been developed in this study based on the field data from 2018 to 2020. PNN takes soil total nitrogen, organic matter, total salt, pH, irrigation time and target soil depth as input, and irrigation amount and nitrogen application rate (N rate) as output, and the prior preference matrix was used to adjust the learning of weight matrix of each layer. The outcomes indicated that the predictive accuracy of PNN for irrigation amount were (R2 = 0.913, MAE = 0.018, RMSE = 0.022), and for N rate were (R2 = 0.943, MAE = 0.009, RMSE = 0.011). The R2 predicted by PNN at the irrigation amount and N rate were 40.03% to more than 99% and 40.33% to more than 99% higher than those obtained using support vector regression (SVR), linear regression (LR), logistic regression (LOR) and traditional back propagation neural network (BPNN), respectively. In addition, compared with the neural network (Reverse Multilayer Perceptron, RMLP) with the same structure but no preference structure, the R2 of the predicted irrigation amount and N rate by PNN increased by 25.81% and 27.99%, respectively. The results showed that, through the irrigation of 93 to 102, 92 to 98 and 92 to 98 mm, along with nitrogen applications of 65 to 71, 64 to 73 and 72 to 81 kg/hm2 at 17, 59 and 87 days after sowing, respectively, the organic matter, total nitrogen, total salt content and pH of the soil would reach high fertility levels simultaneously.

Optimization of Performance Traditional Back-propagation with Cyclical Rule for Forecasting Model

The traditional Back-propagation algorithm has several weaknesses, including long training times and significant iterations to achieve convergence. This study aims to optimize traditional Back-propagation using the cyclical rule method to cover these weaknesses. Optimization is done by changing the training function and standard Back-propagation parameters using the training function and cyclical rule parameters. After that, a comparison of the two results will be carried out. This study uses quantitative method of time-series data on coronavirus cases sourced from the Worldometer website, then analyzed using three forecasting models with five input layers, one hidden layer (5, 10, and 15 neurons) and one output layer. The results showed that the 5-10-1 model with the training function and cyclical rule parameters and the tansig and purelin activation functions could perform well in optimization, including faster training time and smaller iterations (epochs), MSE training performance, and better tests. Low and high accuracy (92%) with an error rate of 0.01. So it was concluded that the training function and cyclical rule parameters with the tansig and purelin activation functions were able to optimize the traditional Back-propagation method, and the 5-10-1 model could be used for forecasting active cases of the coronavirus in Asia

Lapping Quality Prediction of Ceramic Fiber Brush Based on Gaussian-Restricted Boltzmann Machine.

Although ceramic fiber brushes have been widely used for deburring and surface finishing, the associated relationship between process parameters and lapping quality is still unclear. In order to optimize the lapping process of ceramic fiber brushes, this paper proposes a multi-layer neural network based on the Gaussian-restricted Boltzmann machine (GRBM), and verified its prediction effectiveness. Compared with a traditional back-propagation neural network, its prediction error was reduced from 7.6% to 4.5%, and the determination coefficient was increased from 0.96 to 0.98, respectively. The comparison results showed that the proposed model can better grasp the relationship between process parameters and machining quality, which can be used as a decision-making foundation for lapping-process optimization.

Temperature compensation methods of spin-exchange relaxation-free co-magnetometer

According to the temperature characteristics of spin-free exchange relaxation (SERF) co-magnetometer, three temperature compensation methods are proposed in this paper, including particle swarm optimization radial basis function (PSO-RBF) neural network, Gaussian regression and least squares support vector basis. The effectiveness of the three compensation methods is verified by experiments and compared with the back-propagation (BP) neural network optimized by a genetic algorithm compensation method. In order to improve the effect of temperature compensation, this paper also conducts correlation and cluster analysis on the different positions temperature and output signals of the SERF co-magnetometer, and selects the data of temperature points that are closely related to signal bias changes for model training. Through experimental comparison with traditional linear compensation and BP neural network compensation methods, it is found that PSO-RBF neural network has advantages in training speed, compensation accuracy and robustness. Experiments show that PSO-RBF neural network temperature compensation algorithm improves the stability of the SERF co-magnetometer by more than 53 at room temperature or under artificially imposed temperature changes.