Conventional Mathematical Models Research Articles

Transformer hot spot temperature estimation through adaptive neuro fuzzy inference system approach

Transformer performance and efficiency can be enhanced by effectively address the properties of its insulation system. The power transformer insulation system weakens as a result of operational thermal stresses brought on by dynamic loading and shifting environmental patterns. Winding hot spot temperature is a crucial metric that must be maintained below the prescribed limit while power transformers are operating so as to maintained power system reliability. This is due to the fact that, among other variables, the time-dependent aging effect of insulation depends on transitions in hot spot temperatures. Due to the non‐linear nature of the conventional mathematical models used to determine these temperatures, and complexity of thermal phenomena, investigations still need to be exercised to fully understand the variables that associate with hot spot temperature computation with minimum error. This paper explores the possibilities of enhancing top oil and hot spot temperature estimation accuracy through the use of an adaptive neuro-fuzzy inference (ANFIS) technique. The paper presents an adaptive neuro fuzzy model to approximate the hot spot temperature of a mineral oil-filled power transformer based on loading, and established top oil temperature. Initially, a sub-ANFIS top oil temperature estimation model based on loading and ambient temperature as inputs is established. Using a hybrid optimization technique, the ANFIS membership functions were fine-tuned throughout the training process to reduce the difference between the actual and anticipated outcomes. The correctness and reliability of the created adaptive neural fuzzy model have been verified using real-world field data from a 60/90MVA, 132 kV power transformers under dynamic operating regimes. The ANFIS model results were validated against field measured values and literature-based electrical-thermal analogous models, establishing a precise input-output correlation. The developed ANFIS model achieves the highest coefficient of determination for both TOT and HST (0.98 and 0.96) and the lowest mean square error (7.8 and 10.3) among the compared thermal models. Correct determination of HST can help asset managers in thermal analysis trending of the in-service transformers, helping them to make proper loading recommendations for safeguarding the asset.

Model of cognitive activity of the human brain based on the mathematical apparatus of quantum mechanics

Describing the transmission and processing of information by an individual is a fundamental issue in modern cognitive science. Previously, unique scientific models for information transfer between individuals [1], as well as models for cognitive activity [2], were developed. However, numerous presented models lack scalability and formalization, hindering their ability to provide a comprehensive explanation for information transfer processes and their distortion due to interaction with the external communicative environment. One aspect of human cognitive activity is that individuals think not based on a code, as a computer does, but through the interaction of various mental images. Despite the fact that these images have a specific material foundation in the form of electrical and chemical activity in the human brain, describing them using conventional mathematical models presents several difficulties [3]. Therefore, this paper suggests novel techniques for describing information images/representations’ functioning, which simulate an individual’s cognitive activity. These methods rely on the mathematical models of self-oscillatory and conventional quantum physics (including potential wells and virtual particles conventionally used in the physical domain) for characterizing basic interactions [4]. The authors adopt a phenomenological perspective and do not regard cognitive systems as quantum. The aim of this study ject is to establish a model for cognitive function in the human brain using the mathematical principles of self-oscillating quantum mechanics from the perspective of information imaging and representation. Methods. The theory proposes [5] that information images/representations (IR) share characteristic properties with virtual Feymann particles and other elementary particles. The human mind is presented as a one-dimensional potential well with finite walls of varying sizes and an internal potential barrier that simulates the boundary between consciousness and subconsciousness. The authors have applied parametrization based on this theory. As a result, we propose the foundations of the mathematical apparatus based on classical quantum mechanics, followed by the mathematical apparatus of self-oscillating quantum mechanics. The latter, though little known, may allow to predict certain cognitive functions of the human brain by modifying and applying it to non-quantum settings. An equation is derived for the state function of the information image of an individual engaging in cognitive activity. Primary calculations were conducted on the state functions of information images using a computer model. The movement patterns of the information image within and outside were then deduced.

Release kinetics approach of stimuli-responsive mesoporous silica nanocarriers: pH-sensitive linker versus pH-sensitive framework

Silica-based stimuli-responsive nanomaterials have attracted significant attention in the field of drug delivery because they can achieve controlled release of anticancer drugs. For this, a gatekeeper is usually used, thus avoiding unwanted leakage at pH 7.4, while modulating the release under more acidic pH conditions in the tumor cell microenvironment. To optimize the efficiency of these nanostructures, it is crucial to study the release kinetics of anticancer drugs, predict their behavior, and modify the transport process mechanism. In the present study, we synthesized and characterized two types of pH-responsive mesoporous silica nanocarriers, which have transferrin conjugates on the surface serving as a gatekeeper. One nanocarrier was a conventional mesoporous silica nanoparticle (MSN) with transferrin attached through a pH-sensitive linker (MSN-Tf), while the other had a pH-sensitive diimine bridge in its framework, giving it degradable characteristic (dMSN-Tf). The use of conventional mathematical models and a three-parameter model that considers the drug-matrix interaction allowed the evaluation and elucidation of the different mechanisms involved in the kinetics of doxorubicin release from both materials at pH 7.4 and pH 5.0. The change in the kinetics of doxorubicin release from zero-order to first-order when it is in a degradable system such as dMSN-Tf, can be very useful for rapid drug delivery in acidic media such as inside cancer cells. Cell viability experiments in lung cancer cell lines A549 and H1299, showed enhanced anticancer activity by the nanomaterial with the pH-sensitive framework compared to the material that has a pH-sensitive linker. Overall, these findings provide important insights for optimizing the delivery of anticancer drugs.

A framework for modeling of catalyst deactivation and coking free of main-reaction kinetics in methanol to gasoline process

Optimizing the methanol-to-gasoline technology necessitates a profound comprehension of catalyst deactivation and coke formation kinetics. However, conventional mathematical models cannot capture catalyst deactivation without relying on complex main-reaction kinetic models, and they treat the prediction of coking as a separate problem. Addressing these challenges, this research introduces a computational framework to identify the optimal model for the kinetics of deactivation and coking. The key novelty is leveraging the concept of active site loss to eliminate the requirement for main-reaction kinetic models while facilitating the prediction of coking according to the site loss model. Experimental data of an HZSM-5 catalytic bed under diverse operating conditions (WHSV: 10–100 h−1, temperature: 325–375 °C) were employed to validate method efficiency for predicting reactor outputs, including oxygenates, light olefins, and gasoline. The proposed approach demonstrates a 0.52 % Root-Mean-Square-Error in simulating product weights and a 66 % reduction in deactivation kinetic constants. This study reveals assuming a critical volume for each grown coke precursor justifies experimental data by R2 of 99.71 %. This volume was computed around 100 Å3 and exhibits a linear Arrhenius temperature dependency. These results highlight the high accuracy and extreme simplicity this framework offers. Moreover, the method transforms deactivated catalyst data at varying WHSVs into a larger unified dataset for the fresh catalyst at each isotherm. Hence, it also positively impacts the kinetic study of main reactions. Consequently, the developed framework is a promising tool for direct assessment of catalyst deactivation and coking in similar processes toward advancing catalyst design strategies and fuel production efficiency.

Implementation of deep neural networks and statistical methods to predict the resilient modulus of soils

ABSTRACT The Resilient Modulus (Mr) is perhaps the most relevant and widely used parameter to characterise the soil behaviour under repetitive loading for pavement applications. Accordingly, it is a crucial parameter controlling the mechanistic-empirical pavement design. Nonetheless, determining the Mr by laboratory tests is not always possible due to the high consumption of time and financial resources. Thus, developing new indirect approaches for estimating the MR is necessary. Precisely, this article investigates the application of Deep Neural Networks (DNNs) and statistical methods to predict the Mr of soils. For that purpose, the Long-Term Pavement Performance (LTPP) database was implemented. It includes 64 701 datasets resulting from coarse-grained and fine-grained soil samples considering a wide range of grain size distribution and subjected to different stress levels. The input parameters were the bulk stress, octahedral shear stress, and the percentage of soil particles passing through the different sieves (3”, 2”, 3/2”, 1”, 3/4”, 1/2”, 3/8”, No. 4, No. 10, No. 40, No. 80, and No. 200) and the output was the Mr. The results suggest that while conventional mathematical models are unable to predict the influence of the grain size distribution and stress level on the Mr, the proposed DNNs were able to reproduce very accurate predictions. Notably, the proposed computational models have been uploaded to a GitHub repository and have become a valuable tool for forecasting the Mr when experimental measurements are not feasible.

What goes up...

Financial bonds are usually seen as safe and dependable investments, but a recent crash in the value of “ultra-long” bonds was bigger than conventional mathematical models could have predicted. They should have used rollercoaster physics, as Jon Cartwright explains.

Interpretable machine learning learns complex interactions of urban features to understand socio‐economic inequality

AbstractInequality in cities is a phenomenon arising from the complex interactions among urban systems and population activities. Conventional statistics and mathematical models like multiple regression models require assumptions of feature interactions with specified mathematical forms that may fail to fully capture complex interactions of heterogeneous urban components, creating challenges in systematically assessing socio‐economic inequality in cities. To overcome the limitations of these conventional mathematical models, in this work, we propose an interpretable machine learning model to capture the complex interactions of urban variables and the main interaction effects on socio‐economic statuses. We extract urban features from high‐resolution anonymized mobile phone data with billions of activity records related to people and facilities in 47 US metropolitan areas and predict the attributes of urban areas from six income and race groups. We show that socio‐economic inequality in cities can be effectively measured by the predictability of trained machine learning models in controlled experiments. We also examine the tradeoff between spatial resolution, sample size, and model accuracy; test the presence of influential features; and measure the transferability of the trained models to identify the optimal values for controlled factors. The results show that metropolitan areas share similar patterns of inequality, which could be moderated by improved polycentric facility distribution and road density. The generality of associated factors and transferability of machine learning models can help bridge data gaps between cities and inform about inequality alleviation strategies. Despite similarities, 50% to 90% of variations among cities are still present, which shows the need for localized policies for inequality alleviation and mitigation. Our study shows that machine learning models could be an effective approach to examine inequality, which opens avenues for more data‐centric and complexity‐informed planning, design, policymaking, and engineering toward equitable cities.

A GPS Distance Error Forecast Model Based on IIR Filter Denoising and LSTM

Wireless channel prediction has attracted significant attention in recent years but is still considered an open topic. Global Positioning Systems (GPS) error prediction is one such example. Conventional mathematical and statistical models cannot predict time-varying channel variables within coherent time windows. Their mechanisms are geared towards minimizing mean square errors over all datasets rather than time-sequential prediction within coherent windows. On the other hand, Artificial Neural Network (ANN) based approaches have been introduced and studied for this purpose. ANNs have the capability to learn and identify underlying patterns in data. However, extracting and predicting wireless channel variables is a non-trivial task that presents challenges to researchers. Moreover, the widely applied Long Short-term Memory (LSTM) does not demonstrate satisfactory results because of unavoidable interferences, outliers and noise. To address these concerns, a novel LSTM with Infinite Impulse Response (IIR) filtering is proposed, which performs preprocessing of data smoothing. The proposed model reconfigures LSTM cells by adding new IIR gates, dedicated to removing noise and interferences for each cell in the hidden layers of ANN, and are self-adaptive through backpropagation during the training phase to achieve optimization. The model is evaluated with several commonly used time series forecast models, and the GPS error prediction results demonstrate that the proposed model is able to well adapt to the datasets with the best performance. The average prediction accuracy at the 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> step is increased by 34% compared to the second-best model.

Evaluation of medical services from the perspective of COVID-19 vaccine demand satisfaction in Hangzhou, China.

The outbreak of COVID-19 has had a huge global impact, and it continues to test the resilience of medical services to emergencies worldwide. In the current post-epidemic era, vaccination has become a highly effective strategy to prevent the spread of COVID-19. However, using conventional mathematical models to evaluate the spatial distribution of medical resources, including vaccination, ignore people's behaviors and choices and make simplifications to the real world. In this study, we use an enhanced model based on the Theory of People Behavior (TPB) to perform a macro analysis of the satisfaction ability of medical resources for vaccination in Hangzhou, China, and attribute the city to a three-level structure. According to the allocation, the supply capacity of vaccination sites is calculated and divided into four categories (good, normal, not bad, and bad). Meanwhile, we raise an assumption based on the result and the general development law of the city and analyze the reasons for the impact of personal behavior on the spatial distribution of medical resources, as well as the relationship between the demand distribution and spatial distribution of medical resources and future development strategies. It is considered that the overall medical resources, especially vaccination in Hangzhou, feature the situation of central supply overflow, and are found to hardly meet the needs of population points in surrounding areas, requiring a more flexible strategy to allocate facilities in these areas.

Deep Learning Model for Global Spatio-Temporal Image Prediction

Mathematical methods are the basis of most models that describe the natural phenomena around us. However, the well-known conventional mathematical models for atmospheric modeling have some limitations. Machine learning with Big Data is also based on mathematics but offers a new approach for modeling. There are two methodologies to develop deep learning models for spatio-temporal image prediction. On these bases, two models were built—ConvLSTM and CNN-LSTM—with two types of predictions, i.e., sequence-to-sequence and sequence-to-one, in order to forecast Aerosol Optical Thickness sequences. The input dataset for training was NASA satellite imagery MODAL2_E_AER_OD from Terra/MODIS satellites, which presents global Aerosol Optical Thickness with an 8 day temporal resolution from 2000 to the present. The obtained results show that the ConvLSTM sequence-to-one model had the lowest RMSE error and the highest Cosine Similarity value. The advantages of the developed DL models are that they can be executed in milliseconds on a PC, can be used for global-scale Earth observations, and can serve as tracers to study how the Earth’s atmosphere moves. The developed models can be used as transfer learning for similar image time-series forecasting models.

COVID-19 forecasting using new viral variants and vaccination effectiveness models

Recently, a high number of daily positive COVID-19 cases have been reported in regions with relatively high vaccination rates; hence, booster vaccination has become necessary. In addition, infections caused by the different variants and correlated factors have not been discussed in depth. With large variabilities and different co-factors, it is difficult to use conventional mathematical models to forecast the incidence of COVID-19. Machine learning based on long short-term memory was applied to forecasting the time series of new daily positive cases (DPC), serious cases, hospitalized cases, and deaths. Data acquired from regions with high rates of vaccination, such as Israel, were blended with the current data of other regions in Japan such that the effect of vaccination was considered in efficient manner. The protection provided by symptomatic infection was also considered in terms of the population effectiveness of vaccination as well as the vaccination protection waning effect and ratio and infectivity of different viral variants. To represent changes in public behavior, public mobility and interactions through social media were also included in the analysis. Comparing the observed and estimated new DPC in Tel Aviv, Israel, the parameters characterizing vaccination effectiveness and the waning protection from infection were well estimated; the vaccination effectiveness of the second dose after 5 months and the third dose after two weeks from infection by the delta variant were 0.24 and 0.95, respectively. Using the extracted parameters regarding vaccination effectiveness, DPC in three major prefectures of Japan were replicated. The key factor influencing the prevention of COVID-19 transmission is the vaccination effectiveness at the population level, which considers the waning protection from vaccination rather than the percentage of fully vaccinated people. The threshold of the efficiency at the population level was estimated as 0.3 in Tel Aviv and 0.4 in Tokyo, Osaka, and Aichi. Moreover, a weighting scheme associated with infectivity results in more accurate forecasting by the infectivity model of viral variants. Results indicate that vaccination effectiveness and infectivity of viral variants are important factors in future forecasting of DPC. Moreover, this study demonstrate a feasible way to project the effect of vaccination using data obtained from other country.

Lattice-and-Plate Model: Mechanics Modeling of Physical Origami Robots.

Origami robots are characterized by their compact design, quasi-two-dimensional manufacturing process, and folding joint-based transmission kinematics. The physical requirements in terms of payload, range of motion, and embedding core robotic components have made it unrealistic to rely on conventional mathematical models for designing these new robots. Therefore, origami robots require a comprehensive approach to model their mechanics. Currently, there is no generalized mechanics model to achieve this goal. Therefore, in this work, we propose a nonlinear lattice-and-plate model to simulate the mechanics of physical origami robots within several seconds, including the localized bending on flexible hinges, global displacements of rigid panels, and trajectory of predefined outputs. Moreover, this proposed model captures the large displacement and self-contact of adjacent panels during locomotion. We validate the efficiency of the model on various origami actuators, grippers, and metamaterials. To conclude, the computational model can help to accelerate the design iteration of origami robots.

Correlating mechanical properties of polyurethane-organoclay nanocomposite coatings with processing

In this study, mechanical properties were computed for polyurethane (PU)/ Cloisite20A (C20A) coatings. These coatings material were synthesized by three processing methods at variable clay weight percentages (0–3 wt%). Dispersion of C20A platelets in pre-polymer suspension were assessed by the transmission electron microscopy (TEM). In order to scrutinize the conformational changes in the functional groups at the molecular level, deconvolution of Fourier transform infrared spectroscopy – attenuated total reflectance (FTIR-ATR) spectrum was employed. An enhancement in elongation at break with the reduction in stiffness for the nanocomposites was quantified from dynamic tensile testing. For instance, 3 wt% C20A PU nanocomposites processed by sonication and subsequent homogenization showed a decrease of Young's modulus by 53% but exhibits an increase in both elongation at break and toughness by 97% and 93% respectively. The amount of shear force exerted during the processing protocols played a crucial role w.r.t dispersion of C20A. Even at low wt% (1–2) nanocomposite samples exhibit better mechanical properties in simultaneous processing using sonication and homogenization. To quantify the relationship between theoretical and experimental observations, three conventional mathematical models have been applied. Deviations in model behavior was observed, due to the high sensitivity of the nanocomposites to processing schemes, even with same chemical formulations.

Quasiconformal model with CNN features for large deformation image registration

&lt;p style='text-indent:20px;'&gt;Image registration has been widely studied over the past several decades, with numerous applications in science, engineering and medicine. Most of the conventional mathematical models for large deformation image registration rely on prescribed landmarks, which usually require tedious manual labeling. In recent years, there has been a surge of interest in the use of machine learning for image registration. In this paper, we develop a novel method for large deformation image registration by a fusion of quasiconformal theory and convolutional neural network (CNN). More specifically, we propose a quasiconformal energy model with a novel fidelity term that incorporates the features extracted using a pre-trained CNN, thereby allowing us to obtain meaningful registration results without any guidance of prescribed landmarks. Moreover, unlike many prior image registration methods, the bijectivity of our method is guaranteed by quasiconformal theory. Experimental results are presented to demonstrate the effectiveness of the proposed method. More broadly, our work sheds light on how rigorous mathematical theories and practical machine learning approaches can be integrated for developing computational methods with improved performance.&lt;/p&gt;

Hysteresis modeling and compensation of a rotary series elastic actuator with nonlinear stiffness.

Series elastic actuators (SEAs) have widely been adapted in robots where safe human-robot interaction is required for accurate and robust force control. Recent research on the SEAs has shown that the SEA with a user-defined variable stiffness possesses several advantages over the constant stiffness SEA, such as large force range and bandwidth while keeping low output impedance and high force fidelity. However, a limitation of this type of SEA is that an obvious hysteresis effect exists and the associated torque curves are nonlinear and vary with amplitudes. Conventional mathematical hysteresis models are usually developed with some kind of black-box modeling, and the model parameters are adjusted through parameter identification methods. It is challenging to tune the model parameters to match the experimental data well among inputs with different amplitudes, let alone the inverse model of the hysteresis, which is necessary to compensate the hysteresis effect in control. In this paper, a rotary SEA (rSEA) with nonlinear stiffness is proposed. A concept called "virtual deformation" is introduced to mathematically transform the nonlinear curve into a polyline hysteresis model. This eases torque estimation with respect to the deformation of the rSEA. A hysteresis compensation torque controller is implemented for precise torque control. A prototype of the rSEA was fabricated, and the experimental results verified modeling accuracy of the proposed model. Our results showed that, with the new model, the computation cost was greatly reduced while keeping the modeling accuracy almost the same compared with the nonlinear backlash model.

Enhancement of membrane system performance using artificial intelligence technologies for sustainable water and wastewater treatment: A critical review

In recent years, membrane technologies are widely utilized in water and wastewater treatment processes. However, controlling and improving these systems still need to be investigated and, therefore, are attracting increasing amounts of attention from researchers worldwide. Industry 4.0 has increased in importance over the past few years, and artificial intelligence (AI) technology has demonstrated its strength in supporting decision-making in various fields, including environmental systems and especially membrane processes. AI allows for cost-effective operation of systems, including better planning and tracking as well as comprehensive understanding of resource-loss in real-time, then maximizing revenue capture and water quality satisfaction. This study therefore aims to provide a comprehensive review of the current application of AI-based tools in simulating membrane processes as well as the feasibility of applying these models to other fields in which membranes are to be used in the future. The existing conventional mathematical models are illustrated along with their advantages and shortcomings. The definition and classification of state-of-the-art AI models, as well as the benefits of these over conventional models, are also discussed. Furthermore, the basic principle of membrane processes and current application of AI-based technologies in simulating the performance of these membrane systems are systematically reviewed. Finally, the implications and recommendations for future studies are discussed.

Methodology of developing mathematical-ANN hybrid model based mill setup model for hot strip rolling

Hot strip mill, one of the most important units of an integrated steel plant, is operated by mill setup model. The conventional mill setup models calculate thermal, reduction and speed schedules of the material being rolled using mathematical models derived from fundamental principles of heat transfer and plastic deformation. However, such mill setup models often compute inaccurate schedules leading to quality issues and operational problems. This paper describes a novel technique of developing a hybrid model by integrating mathematical models with artificial neural network (ANN) model. The trained hybrid models use a multivariable optimization algorithm to calculate the thermal, reduction and speed schedules during hot strip rolling. More than six hundred coils were successfully rolled in an industrial hot strip mill using the mill setup model developed under the present work. It is found that the mill setup model developed using the hybrid models is more accurate and faster than the mill setup models that use conventional mathematical models.

A genetic algorithm for the fuzzy shortest path problem in a fuzzy network

The shortest path problem (SPP) is an optimization problem of determining a path between specified source vertex s and destination vertex t in a fuzzy network. Fuzzy logic can handle the uncertainties, associated with the information of any real life problem, where conventional mathematical models may fail to reveal proper result. In classical SPP, real numbers are used to represent the arc length of the network. However, the uncertainties related with the linguistic description of arc length in SPP are not properly represented by real number. We need to address two main matters in SPP with fuzzy arc lengths. The first matter is how to calculate the path length using fuzzy addition operation and the second matter is how to compare the two different path lengths denoted by fuzzy parameter. We use the graded mean integration technique of triangular fuzzy numbers to solve this two problems. A common heuristic algorithm to solve the SPP is the genetic algorithm. In this manuscript, we have introduced an algorithmic method based on genetic algorithm for determining the shortest path between a source vertex s and destination vertex t in a fuzzy graph with fuzzy arc lengths in SPP. A new crossover and mutation is introduced to solve this SPP. We also describe the QoS routing problem in a wireless ad hoc network.

Viscoelastic Wave Simulation with High Temporal Accuracy Using Frequency-Dependent Complex Velocity

In recent decades, the study of seismic attenuation has received more and more concerns because it can stimulate the development of wave propagation simulation and improve the accuracy of structure imaging and reservoir prediction. In this paper, we review the attenuation theory and the development of high temporal accuracy wave simulation. The conventional mathematical models to describe the characteristics of viscoelastic are based on constant-Q model or standard linear solids theory. However, these approaches possess some noticeable shortcomings. Therefore, we introduce a frequency-dependent complex velocity to derive the novel viscoelastic wave equations with decoupled amplitude dissipation and phase dispersion. To obtain high temporal accuracy viscoelastic wave simulation, we adopt the normalized pseudo-Laplacian to compensate for the temporal dispersion errors caused by the second-order finite-difference discretization in the time domain. During the implementation, we incorporate the normalized pseudo-Laplacian into the optimized staggered-grid finite-difference coefficients. Therefore, it can greatly reduce the times of low-rank decomposition and Fourier transform and largely improve the computational efficiency. Based on this strategy, we can implement the high temporal accuracy viscoelastic wavefield extrapolation by comprehensively exploiting the staggered-grid finite-difference scheme, pseudo-spectral method and low-rank decomposition algorithm. Meanwhile, a linear velocity model is employed to evaluate the accuracy of low-rank approximation. Furthermore, we use several numerical examples to carry out the comparison between our scheme and other conventional methods. The numerical results reveal that our proposed scheme can effectively compensate for temporal dispersion errors and help generate high temporal accuracy viscoelastic wave solutions.

Forecasting of Coalbed Methane Daily Production Based on T-LSTM Neural Networks

Accurately forecasting the daily production of coalbed methane (CBM) is important forformulating associated drainage parameters and evaluating the economic benefit of CBM mining. Daily production of CBM depends on many factors, making it difficult to predict using conventional mathematical models. Because traditional methods do not reflect the long-term time series characteristics of CBM production, this study first used a long short-term memory neural network (LSTM) and transfer learning (TL) method for time series forecasting of CBM daily production. Based on the LSTM model, we introduced the idea of transfer learning and proposed a Transfer-LSTM (T-LSTM) CBM production forecasting model. This approach first uses a large amount of data similar to the target to pretrain the weights of the LSTM network, then uses transfer learning to fine-tune LSTM network parameters a second time, so as to obtain the final T-LSTM model. Experiments were carried out using daily CBM production data for the Panhe Demonstration Zone at southern Qinshui basin in China. Based on the results, the idea of transfer learning can solve the problem of insufficient samples during LSTM training. Prediction results for wells that entered the stable period earlier were more accurate, whereas results for types with unstable production in the early stage require further exploration. Because CBM wells daily production data have symmetrical similarities, which can provide a reference for the prediction of other wells, so our proposed T-LSTM network can achieve good results for the production forecast and can provide guidance for forecasting production of CBM wells.