Time Series Of Fluctuations Research Articles

Clustering time series under trend-oriented fuzzy information granulation

This paper proposes a trend-oriented time series granulation method to transform a long numerical time series into a relatively short granular time series which is consist of a group of unequal-size linear fuzzy information granules (LFIG). The transformed granular time series not only captures the main characteristics like trends and fluctuations of the original time series, but also saves the amount of calculation in time series clustering. Inspired by the distance measure of two equal-size LFIGs and the dynamic time warping, this paper also defines the distance measures for two unequal-size LFIGs and two LFIG time series. Based on such distance measures, the k-medoids method is employed to cluster the datasets coming from UCR time-series database. The clustering performance expressed in terms of computing time and the Rand Index demonstrates the effectiveness and advantages of the proposed time series granulation method and distance measurement. The main original aspects of this study concern the granular representation of time series with unequal-size granules, and the distance measurement of unequal-length granular time series.

A New Time-Series Fluctuation Study Method Applied to Flow and Pressure Data in a Heating Network

The key to achieving smart heating is the rational use of large amounts of data from the heating network. However, many current relevant studies based on generalized mathematical methods are unable to accurately describe the physical relationships between pipe network variables. In order to solve this problem, this paper proposes a new time-series fluctuation research method, which can be applied to the measured data of the hot water heating pipe network. This method is a new approach to identifying step data. Then, we propose the concept of time-series disturbance to quantify the degree of data anomaly. Finally, the results of a case study demonstrate the transfer process of a significant disturbance in the pipe network from the supply end to the return end. The time-series fluctuation method in this paper precisely describes two physical relationships between heating system variables and provides a feasible and convenient new research idea for self-perception and self-analysis of smart heating.

Prediction of groundwater level fluctuations under climate change based on machine learning algorithms in the Mashhad aquifer, Iran

Abstract The purpose of this study is the projection of climate change's impact on the Groundwater Level (GWL) fluctuations in the Mashhad aquifer during the future period (2022–2064). In the first step, the climatic variables using ACCESS-CM2 model under the Shared Socio-economic Pathways (SSPs) 5–8.5 scenario were extracted. In the second step, different machine learning algorithms, including Multilayer Perceptron Neural Network (MLP), Adaptive Neuro-fuzzy Inference System Neutral Network (ANFIS), Radial Basis Function Neural Network (RBF), and Support Vector Machine (SVM) were employed for the GWL fluctuations time series prediction under climate change in the future. Our results point out that temperatures and evaporation will increase in the autumn season, and precipitation will decrease by 26%. The amount of evaporation will increase in the winter due to an increase in temperature and a decrease in precipitation. The results showed that the RBFNN model had an excellent performance in predicting GWL compared to other models due to the highest value of R² (R² = 0.99) and the lowest value of RMSE, which were 0.05 and 0.06 meters in training and testing steps, respectively. Based on the result of the RBFNN model, the GWL will decrease by 6.60 meters under the SSP5-8.5 scenario.

Complexity parameters of solar-wind magnetic fluctuations at 1 AU during SC23 and SC24

Context. The solar wind develops a highly turbulent character during its expansion, where plasma and electromagnetic fluctuations coexist. Considering the presence of turbulence in the plasma as a complex system, the turbulence in the solar wind in general has been measured and studied using different techniques from a systems science point of view. These techniques provide the opportunity to obtain preliminary information even before much of the physics can be assimilated and integrated. Aims. We describe this plasma as a complex system in order to understand solar wind dynamics from a new perspective. Several missions provide a wide range of data concerning critical astrophysical phenomena. This poses a challenge to implement new effective methods to complement the characterization of the constantly new, and sometimes highly reduced information, especially when dealing with observational data with intermittent gaps. Methods. We work with magnetic fluctuation time series data obtained from the Wind mission at 1 AU in order to characterize the fast and slow solar wind behavior during solar cycles 23 (SC23) and 24 (SC24). We applied the horizontal visibility graph (HVG) method to obtain the evolution of measurements of Kullback-Leibler divergence (KLD), D, and the characteristic exponent, γ, over time. Both are complexity parameters extracted from the degree distributions of the networks. Results. By contrasting our complexity parameters, γ and D, with solar activity characterized by the number of sunspots and solar wind speed, we obtain significant intercorrelations among them during both cycles and ascending, descending, minimum, and maximum phases. According to γ values, the magnetic fluctuations of the solar wind are a correlated stochastic time series at 1 AU. Also, the divergence D recognizes SC23 as the most dissipative and identifies the slow wind as more variable than the fast wind, with a better anti-correlation in the minima phases. This study reveals that in terms of solar phases γmin > γdes > γasc > γmax, and Dmin < Ddes < Dasc < Dmax. We show that the HVG technique leads to results that are consistent with the complex nature of solar wind turbulence.

A framework to identify critical dynamics of water quality for diagnosing river basin ecosystem resilience and management

This study proposed a novel framework to identify critical water quality dynamics as early warning signals for diagnosing changes in the resilience of river basin ecosystems. We established empirical linkages between the theoretical background of three resilience capacities (robustness, adaptability and transformability) and water quality dynamics. Then, the processes of resilience degradation and their risk transfer or accumulation have been identified based on the shifts among different states. The methods of time-domain analysis and frequency-domain analysis were integrated into this framework, aiming to identify gradual and transient responses of water quality and its periodic fluctuation characteristics at multiple temporal scales. The time-domain analysis methods obtained the trend, cumulative periodic fluctuation of water quality by extracting the key characteristic parameters from the time-series data. The wavelet transform methods were introduced into the frequency-domain analysis to reveal the water quality fluctuation patterns at specific temporal scales. We tested the proposed framework in a typical agriculture-intensive watershed in eastern China. The results showed that this framework can be effectively used to identify three resilience states of river basin ecosystems. The degraded resilience regions were mainly distributed in the downstream area, which was influenced by their specific land use/cover and different agricultural soil health conditions. Particularly, urban sewage discharge was the main cause of periodic fluctuation in water quality time series at multiple high-frequency scales. The theoretical background of resilience capacities was elaborated in non-equilibrium dynamics before resilience degradation. Thus, this novel framework could reveal the pollution processes and driving mechanisms in different river reaches, and it also can provide adaptive management suggestions according to resilience dynamics traits.

Wavelet transforms based ARIMA-XGBoost hybrid method for layer actions response time prediction of cloud GIS services

Layer actions response time is a critical indicator of cloud geographical information services (cloud GIS Services), which is of great significance to resource allocation and schedule optimization. However, since cloud GIS services are highly dynamic, uncertain, and uncontrollable, the response time of layer actions is influenced by spatiotemporal intensity and concurrent access intensity, posing significant challenges in predicting layer action response time.To predict the response time of layer actions more accurately, we analyzed the data association of cloud GIS services. Furthermore, based on the characteristics of long-term stable trends and short-term random fluctuations in layer actions response time series, a wavelet transforms-based ARIMA-XGBoost hybrid method for cloud GIS services is proposed to improve the one-step and multi-step prediction results of layer actions response time.We generate a multivariate time series feature matrix using the historical value of the layer actions response time, the predicted value of the linear component, and the historical value of the non-linear component. There is no need to meet the traditional assumption that the linear and nonlinear components of the time series are additive, which minimizes the model’s time series requirements and enhances its flexibility. The experimental results demonstrate the superiority of our approach over previous models in the prediction of layer actions response time of cloud GIS services.

Длинные волны на шельфе юго-западного побережья о. Сахалин

The results of the study of long-wave motions with periods of more than 20 hours on the shelf of the southwest coast of Sakhalin Island using the time series of sea level fluctuations obtained in full-scale experiments with a discreteness of 1 s and a duration of 4 to 6 months are considered. Spectral analysis of the time series of sea level fluctuations for the period range from 8 to 200 hours revealed the presence of long-wave processes with periods from 26.1 to 46.7 hours, which significantly exceed the inertial period of 16.48 hours. Numerical modeling of shelf waves for exponentially convex profiles of the seabed, carried out using the dispersion relation of V.T. Buchwald and J. K. Adams for waves of the continental shelf, showed that the detected wave processes with periods from 31.2 h to 46.7 h are shelf waves. Their amplitudes increase during storms. The possibility of energy transferring from atmospheric disturbances to shelf waves, which contribute to the formation of sea level, is shown, that confirms the earlier made assumption. Using the calculation of the phase difference of the observed shelf waves at a distance of 12.4 km between Nevelsk and Gornazovodsk and determined by means of the theoretical model, it is shown that the second mode of the shelf wave with a frequency of 0.152 cycle/h is close to the theoretical one. It is shown for the wave with a period of 26.1 h recorded in Ilyinsky and Gornozavodsk for the distance between the points of 173.6 km, that this wave cannot be a shelf wave, but it is a Kelvin wave. This is confirmed by the calculated dispersion diagram, according to which the wavelength of about 689 km corresponds well to the phase difference for the distance Ilyinsky – Gornozavodsk. It is found that shelf waves, one of the generation mechanisms of which is the wind tension along the coast, have the different amplitudes in summer and winter, that is due to the seasonal direction of the offshore wind. In summer, the directions of shelf wave propagation and wind are opposite, which weakens the shelf waves.

Improving the Quality of MODIS LAI Products by Exploiting Spatiotemporal Correlation Information

The Moderate Resolution Imaging Spectroradiometer (MODIS) Leaf Area Index (LAI) product is critical for global terrestrial carbon monitoring and ecosystem modeling. However, MODIS LAI is calculated on a pixel-by-pixel and day-by-day basis without using spatial or temporal correlation information, which leads to its high sensitivity of LAI to uncertainties in observed reflectance resulting in an increased noise level in time series. While exploiting prior knowledge is a common practice to fill gaps in observations, little research has been conducted on reducing noisy fluctuations and improving the overall quality of the MODIS LAI product. To address this issue, we proposed a Spatio-Temporal Information Composition Algorithm (STICA), which directly introduces prior Spatio-temporal correlation and Multiple Quality Assessment (MQA) information into the existing MODIS LAI product. STICA reduces the noise level and improves the quality of the product while maintaining the original physically-based (Radiative Transfer Model, RTM) LAI production process. In our analysis, the R2 increased from 0.79 to 0.81, and the RMSE decreased from 0.81 to 0.68 compared to the ground-based LAI reference. The improvement was more pronounced with the degradation of the data quality. STICA reduced noisy fluctuations in the LAI time series to varying degrees among eight biome types. In the Amazon Forest, STICA significantly improved the time-series stability of LAI. Moreover, STICA can effectively eliminate abnormal declines in time series and correct for extreme outliers in LAI. We expect that the MODIS LAI Reanalyzed product generated by this method will better support the application of high-quality LAI datasets.

A guide for concurrent TMS-fMRI to investigate functional brain networks.

Transcranial Magnetic Stimulation (TMS) allows for the direct activation of neurons in the human neocortex and has proven to be fundamental for causal hypothesis testing in cognitive neuroscience. By administering TMS concurrently with functional Magnetic Resonance Imaging (fMRI), the effect of cortical TMS on activity in distant cortical and subcortical structures can be quantified by varying the levels of TMS output intensity. However, TMS generates significant fluctuations in the fMRI time series, and their complex interaction warrants caution before interpreting findings. We present the methodological challenges of concurrent TMS-fMRI and a guide to minimize induced artifacts in experimental design and post-processing. Our study targeted two frontal-striatal circuits: primary motor cortex (M1) projections to the putamen and lateral prefrontal cortex (PFC) projections to the caudate in healthy human participants. We found that TMS parametrically increased the BOLD signal in the targeted region and subcortical projections as a function of stimulation intensity. Together, this work provides practical steps to overcome common challenges with concurrent TMS-fMRI and demonstrates how TMS-fMRI can be used to investigate functional brain networks.

PERBANDINGAN METODE DOUBLE EXPONENTIAL SMOOTHING DAN TRIPLE EXPONENTIAL SMOOTHING PADA PERAMALAN NILAI EKSPOR DI INDONESIA

Exports are a source of foreign exchange that can affect the level of the country's economy and become a benchmark in determining the quality of the country. The value of Indonesian exports expressed in US Dollars is a monthly fluctuating time series data. In an effort to control the value of Indonesia's exports, it is necessary to have the right strategy, one of strategies is forecasting the value of exports in the future. To determine an appropriate forecasting method, the MAPE results from the Double Exponential Smoothing and Triple Exponential Smoothing are compared. From the research, the results of the prediction of the value of Indonesia's exports for the next 7 periods, from June 2022 to December 2022, the most accurate after a comparison of the MAPE value is closest to zero, the result is Triple Exponential Smoothing method more accurate for forecasting the value of Indonesian exports, this is because historical data on the value of Indonesian exports shows a trend and seasonal pattern at the same time

Analysis on the Spatio-Temporal Evolution Characteristics of the Impact of China's Digitalization Process on Green Total Factor Productivity.

Green production is an inevitable choice for China's high-quality economic development. With the rise of the digital technology revolution, China's digital transformation may play an integral and important role in increasing green total factor productivity (GTFP). Based on the panel data of 30 Chinese provinces from 2014-2020, the impact of digitization on GTFP was explored using the model of geographically and temporally weighted regression (GTWR), and the spatial and temporal distribution characteristics and development trends of such effects were further explored. The main findings are as follows: (1) China's digitalization level and GTFP has significant spatial autocorrelation and similar spatial distribution characteristics. (2) Digitalization has a significant positive impact on GTFP, but this impact decreases yearly, and there are noticeable regional differences. Digitalization in the eastern and central regions has a more significant impact on GTFP than in the west. (3) The region where China's digital development has extensively promoted GTFP has shifted from China's southern coastal region to the northwest and northeast regions. (4) The time-series fluctuations of the regression coefficients of the digitization level in each province in China also show agglomeration characteristics. That is, the regression coefficients of neighboring provinces have similar time-series fluctuations.

Time to reach the maximum for a stationary stochastic process.

We consider a one-dimensional stationary time series of fixed duration T. We investigate the time t_{m} at which the process reaches the global maximum within the time interval [0,T]. By using a path-decomposition technique, we compute the probability density function P(t_{m}|T) of t_{m} for several processes, that are either at equilibrium (such as the Ornstein-Uhlenbeck process) or out of equilibrium (such as Brownian motion with stochastic resetting). We show that for equilibrium processes the distribution of P(t_{m}|T) is always symmetric around the midpoint t_{m}=T/2, as a consequence of the time-reversal symmetry. This property can be used to detect nonequilibrium fluctuations in stationary time series. Moreover, for a diffusive particle in a confining potential, we show that the scaled distribution P(t_{m}|T) becomes universal, i.e., independent of the details of the potential, at late times. This distribution P(t_{m}|T) becomes uniform in the "bulk" 1≪t_{m}≪T and has a nontrivial universal shape in the "edge regimes" t_{m}→0 and t_{m}→T. Some of these results have been announced in a recent letter [Europhys. Lett. 135, 30003 (2021)0295-507510.1209/0295-5075/ac19ee].

Prediction of Construction and Production Safety Accidents in China Based on Time Series Analysis Combination Model

At present, the data related to construction safety accidents in China have the characteristics of small sample size, large fluctuation, large base, nonlinearity, etc. It is of great significance in accident prevention to break through the limitations of traditional prediction models in dealing with time series and reduce the contingency of prediction results. The purpose of this study was to use the trend decomposition method to reduce the fluctuation of non-stationary time series and use the combination of an autoregressive integrated moving average model and Grey model with fractional order accumulation to accurately predict construction accidents. This paper analyzed the number of production safety accidents and deaths in housing municipal engineering in China from 2009 to 2019, which makes monthly, quarterly and annual forecasts and compares forecast results. The FOAGM model is based on a genetic algorithm and can forecast individual months separately, which improves the overall forecast of accident numbers. The rolling forecast was used to provide an idea of hierarchical optimization of the forecast results for the annual accident number forecast with a small number of samples. The study also emphasized that the prediction of the death toll is affected by larger and above accidents. By using the CRITIC method, the impacts were quantified, so as to revise the prediction results.

A novel structure adaptive new information priority discrete grey prediction model and its application in renewable energy generation forecasting

• A novel accumulation generation operator is developed by weighting coefficient. • A novel structure adaptive new information priority grey model is proposed. • The robustness analysis of the new model is established. • Comparative study shows that new model is superior to other various grey models. Renewable energy has made a significant contribution to global power generation. Therefore, accurate mid-to-long term renewable energy generation forecasting is becoming more and more important for integrating renewable energy systems with smart grid and energy strategic planning. For this purpose, a novel structure adaptive new information priority discrete grey prediction model is established, and the disturbance analysis shows that it is suitable for small sample modeling. Firstly, the fractional dynamic weighted coefficient is introduced to define the accumulative generating operator satisfying the new information priority principle, which can realize the effective utilization of system information with insufficient information and accurately extract the development mode of system sequence from sparse samples. In terms of model structure, the nonlinear term and periodic fluctuation term are introduced to simulate the nonlinear and periodic fluctuation trend of renewable energy generation data, which improves the adaptability of the grey prediction model to the nonlinear and fluctuating time series. By designing the comparative experiment of optimization algorithm, Grey Wolf Optimizer (GWO) is selected to optimize the structural parameters of the model, giving the proposed model higher flexibility and stronger adaptability. In order to illustrate the performance of the model, the new model is compared with statistical econometrics technology, artificial neural network (ANN) and various existing grey models in three cases: World biofuel power generation, China's renewable energy power generation and small-scale solar photovoltaic power generation in the United States. The experimental results show that the proposed model is superior to other competitive models in terms of fitting accuracy and prediction accuracy. In addition, Monte-Carlo Simulation and probability density analysis further show that the proposed model can provide the best prediction effect and has high robustness.

Variability, skipped breeding and heavy-tailed dynamics in an Antarctic seabird.

The population dynamics of many colonially breeding seabirds are characterized by large interannual fluctuations that cannot be explained by environmental conditions alone. This variation may be particularly confounded by the use of skipped breeding by seabirds as a life-history strategy, which directly impacts the number of breeding pairs and may affect the accuracy of breeding abundance as a metric of population health. Additionally, large fluctuations in time series may suggest that the underlying population dynamics are heavy tailed, allowing for a higher likelihood of extreme events than expected under Gaussian dynamics. Here, we investigated the effect of demography on time series for abundance of the Adélie penguin Pygoscelis adeliae and explored the occurrence of heavy-tailed dynamics in observed Adélie time series. We focus this study on the Adélie penguin as it is an important bellwether species long used to track the impacts of climate change and fishing on the Southern Ocean ecosystem and shares life-history traits with many colonial seabirds. We quantified the impacts of demographic rates, including skipped breeding, on time series of Adélie abundance simulated using an age-structured model. We also used observed time series of Adélie breeding abundance at all known Antarctic colonies to classify distributions for abundance as Gaussian or non-Gaussian heavy tailed. We then identified the cause of such heavy-tailed dynamics in simulated time series and linked these to spatial patterns in Adélie food resource variability. We found that breeding propensity drives observed breeding fluctuations more than any other vital rate, with high variability in skipped breeding decoupling true abundance from observed breeding abundance. We also found several Antarctic regions characterized by heavy-tailed dynamics in abundance. These regions were often also characterized by high variability in zooplankton availability. In simulated time series, heavy-tailed dynamics were strongly linked to high variability in adult survival. Our results illustrate that stochastic variability in abundance dynamics, particularly the presence of variable rates of skipped breeding, can challenge our interpretation of fluctuations in abundance through time and obscure the relationship between key environmental drivers and population abundance.

A Novel Fault Diagnosis Method for Rolling Bearing Based on Hierarchical Refined Composite Multiscale Fluctuation-Based Dispersion Entropy and PSO-ELM

This paper proposes a novel fault diagnosis method for rolling bearing based on hierarchical refined composite multiscale fluctuation-based dispersion entropy (HRCMFDE) and particle swarm optimization-based extreme learning machine (PSO-ELM). First, HRCMFDE is used to extract fault features in the vibration signal at different time scales. By introducing the hierarchical theory algorithm into the vibration signal decomposition process, the problem of missing high-frequency signals in the coarse-grained process is solved. Fluctuation-based dispersion entropy (FDE) has the characteristics of insensitivity to noise interference and high computational efficiency based on the consideration of nonlinear time series fluctuations, which makes the extracted feature vectors more effective in describing the fault information embedded in each frequency band of the vibration signal. Then, PSO is used to optimize the input weights and hidden layer neuron thresholds of the ELM model to improve the fault identification capability of the ELM classifier. Finally, the performance of the proposed rolling bearing fault diagnosis method is verified and analyzed by using the CWRU dataset and MFPT dataset as experimental cases, respectively. The results show that the proposed method has high identification accuracy for the fault diagnosis of rolling bearings with varying loads and has a good load migration effect.

Multi-scale dynamics for a lean-burn spark ignition natural gas engine under low load conditions

This work investigates the multi-scale dynamics of the combustion system in a lean-burn spark ignition natural gas engine using different gas injection timings (GIT). The in-cylinder pressure time series are measured, and the indicated mean effective pressure (IMEP) time series are calculated. The influence of GIT on the combustion system is investigated through wavelet analysis, multi-resolution analysis, and the return maps with the GIT covering from 0 to 90°CA after top dead center at the intake stage. Results show that the combustion system has multi-scale chaotic characteristics, and it is sensitive to the change of GIT. The unreasonable GIT will result in serious combustion variations. Meanwhile, the combustion instability of the engine evolves on multiple time scales, showing evident multi-scale oscillation characteristics. All the wavelet power spectrums (WPS) present the characteristics of intermittent short-time periodic oscillations and persistent large-scale periodic oscillations concealed inside the IMEP time series. When the GIT approaches the medium value, the persistence of large-scale periodic oscillations is weakened, while the characteristics of high-frequency intermittent oscillations are enhanced. Under all working conditions, the contribution rate of high-frequency signal D1 decomposed by the IMEP time series to the overall time series fluctuation is about 40% or even higher. The contribution rate of the signal D1 also increases with the aggravation of combustion instability. The return map structures of high-frequency signals D1 and D2 show bifurcation structures, and the bifurcation characteristics of the signal D1 are more evident under medium GIT conditions, indicating a certain correlation between the 2–8 engine cycles, and the correlation between the adjacent engine cycles is stronger. The deterministic relationship between the multiple engine cycles found can help in developing a more reasonable and efficient combustion control strategy, which provides a theoretical basis for improving the stability of a natural gas engine.

Identifying critical transitions in major depression with fuzzy recurrence entropy

ABSTRACT Detection of mood changes can be useful for the efficacy of drug treatment of psychiatric disorders in major depression. This paper presents an approach for detecting major time points of mood transitions under the influence of antidepressants. These time points, which are indicated by critical fluctuations in time series, can be discovered using change point detection and time-frequency analysis of fuzzy recurrence entropy of mood data. The mood transitions identified by the fuzzy recurrence entropy agree with results that were clinically validated and reported in literature. The proposed approach can detect early mood transitions in a fully objective way that does not require the manual analysis of fluctuations in the signal dynamics.

Numerical Investigation of Flow Structure and Turbulence Characteristic around a Spur Dike Using Large-Eddy Simulation

Spur dikes provide significant control for flow regimes in river regulation engineering, which can help in the regeneration of stream habitats. However, the narrowing of the flow by spur dike changes the turbulence characteristics. To clarify the turbulence characteristics around the spur dike, the method of large eddy simulation (LES) was used to investigate the horizontal turbulence structure around spur dikes with different discharges in an open-channel flume. The simulations were an exact reproduction of large-scale laboratory experiments, which showed agreement with the experimental results. The distributions of time-averaged streamwise velocity, bed shear stress, and second-order turbulence statistics obtained from the LES were analyzed. An examination of the time series of velocity fluctuation as the probability density function, quadrant analysis, the power density spectra, flow instability, and the vortex separation created in the detached shear layer were estimated. The results accurately revealed the flow field under flow separation, the turbulence statistics inside the separated shear layer, and the vortex structure and emphasized the variation in the different water depths. The results demonstrated that the form of turbulence was not significantly affected by discharge. Moreover, vortex and energy transmission displayed the same periodicity, despite variances in the structural form of turbulence at different water depths. Overall, the results of the study provide an efficient basis for understanding the turbulence around spur dikes, which is crucial for their safe design.

A Data-Driven Self-Supervised LSTM-DeepFM Model for Industrial Soft Sensor

Soft sensor, as an important paradigm for industrial intelligence, is widely used in industrial production to achieve efficient monitoring and prediction of production status including product quality. Data-driven soft sensor methods have attracted attention, which still have challenges because of complex industrial data with diverse characteristics, nonlinear relationships, and massive unlabeled samples. In this article, a data-driven self-supervised long short-term memory–deep factorization machine (LSTM-DeepFM) model is proposed for industrial soft sensor, in which a framework mainly including pretraining and finetuning stages is proposed to explore diverse industrial data characteristics. In the pretraining stage, an LSTM-autoencoder is first unsupervised pretrained. Then, based on two self-supervised mask strategies, LSTM-deep can explore the interdependencies between features as well as the dynamic fluctuation in time series. In the finetuning stage, relying on pretrained representation, the temporal, high-dimensional, and low-dimensional features can be extracted from the LSTM, deep, and FM components, respectively. Finally, experiments on the real-world mining dataset demonstrate that the proposed method achieves state of the art comparing with stacked autoencoder-based models, variational autoencoder-based models, semisupervised parallel DeepFM, etc.