False Neighbors Research Articles

Forecasting macroscopic dynamics in adaptive Kuramoto network using reservoir computing.

Forecasting a system's behavior is an essential task encountering the complex systems theory. Machine learning offers supervised algorithms, e.g., recurrent neural networks and reservoir computers that predict the behavior of model systems whose states consist of multidimensional time series. In real life, we often have limited information about the behavior of complex systems. The brightest example is the brain neural network described by the electroencephalogram. Forecasting the behavior of these systems is a more challenging task but provides a potential for real-life application. Here, we trained reservoir computer to predict the macroscopic signal produced by the network of phase oscillators. The Lyapunov analysis revealed the chaotic nature of the signal and reservoir computer failed to forecast it. Augmenting the feature space using Takkens' theorem improved the quality of forecasting. RC achieved the best prediction score when the number of signals coincided with the embedding dimension estimated via the nearest false neighbors method. We found that short-time prediction required a large number of features, while long-time prediction utilizes a limited number of features. These results refer to the bias-variance trade-off, an important concept in machine learning.

General circulation models for rainfall simulations: Performance assessment using complex networks

Reliable assessment of the impact of climate change on hydrology in a region requires proper selection of General Circulation Models (GCMs). The present study introduces the concepts of complex networks for evaluating the performance of GCMs in simulating rainfall and selecting an ensemble of the best-performing GCMs. The performance of 49 GCMs from the Coupled Model Intercomparison Project phase 6 (CMIP6) in simulating monthly rainfall over India is assessed. The observed (interpolated and gridded rainfall provided by the India Meteorological Department) rainfall data and GCM-simulated rainfall data over the period 1961–2014 at a spatial resolution of 1° x 1° (a total of 288 grids) are studied. The GCMs are evaluated for two cases: (1) Case 1 – whole year rainfall; and (2) Case 2 – summer monsoon rainfall (June–September). The clustering coefficient of the rainfall network is used as a network measure to evaluate the ability of the GCMs in simulating rainfall. For rainfall network construction, each grid is considered as a network. The phase space reconstruction concept is used to reconstruct the single-variable rainfall time series in a multi-dimensional phase space to represent the rainfall dynamics. The optimal dimension for reconstruction is determined using the false nearest neighbor (FNN) algorithm. For network construction, each reconstructed vector is considered as a node, and the connections between them, identified using a distance threshold for the reconstructed vectors in the phase space, serve as the links. For each of the 288 grids, the GCMs are ranked based on the difference in the clustering coefficient between the observed and GCM-simulated rainfall networks. The group decision-making (GDM) approach is employed to identify the ensemble of the best-performing GCMs for the entire study area considering all the grids. The results suggest different models perform well for the whole-year rainfall and summer monsoon rainfall. For the whole-year rainfall, the models CMCC-CM2-HR4, GFDL-ESM4, CMCC-ESM2, EC-Earth3-AerChem, and CMCC-CM2-SR5 rank as the top five. For the summer monsoon rainfall, FIO-ESM-2-0, E3SM-1-1, CESM2-FV2, CMCC-CM2-HR4, and CMCC-CM2-SR5 perform the best. Considering the rank of each model for both the cases, the best-performing models are identified to be CMCC-CM2-HR4, CMCC-CM2-SR5, CMCC-ESM2, FIOESM-2-0, and E3SM-1-1. The rainfall simulated from these models also show close resemblance to the observed rainfall, especially in terms of monthly mean rainfall. Therefore, even among the many GCMs that show good agreement with the monthly mean observed rainfall, the clustering coefficient-based analysis helps to narrow down the models for the ensemble of best-performing GCMs.

Adopting a dendritic neural model for predicting stock price index movement

Financial time series forecasting has been an attractive application of machine learning techniques because an advanced forecasting method can help to accurately predict price changes in markets and make good trading profits. In this study, an emerging machine learning approach, named the dendritic neuron model (DNM), is innovatively applied to forecast financial time series. To pursue better prediction performance, a novel scale-free differential evolution (SFDE) is defined as the training algorithm of the DNM, which can well control the balance between exploration and exploitation. In addition, the maximum Lyapunov exponent is used to detect the chaotic property of financial time series; then, the series is reconstructed into a phase space with high dimension before the prediction, where the time delay of the phase space is calculated by a mutual information method and the embedding dimension is separately determined by a false nearest neighbors approach. In our experiments, eight benchmark stock price indices selected from developed markets and emerging markets are used to validate the effectiveness and efficiency of the proposed forecasting model. Overall, the experimental results illustrate that the DNM trained by the SFDE algorithm yields better forecasting performances than other prevailing models and that it can be considered a reliable and satisfactory forecasting tool for predicting price changes in financial markets for practical applications.

Hopf and quasi-periodic Hopf bifurcations and deterministic coherence in coupled Duffing-Holmes and Van der Pol oscillators: the Arnol’d resonance web

A survey on coupled Van der Pol (VDP) and Duffing-Holmes (DH) oscillators, a model widely encountered in various branches of Physics and engineering is done. A bifurcation analysis of the model is presented when both unidirectional and mutual couplings are considered. It is found that stable, unstable and chaotic behaviors appear in the models. Using the method of charts of dynamics regimes in parameter planes, numerical study of the parameters space of the initial differential equations is done. Results from both approaches are compared and discussed. Features of the bifurcation picture are discussed when varying control parameters and analysis of slow-flow equations is presented. We show that the local bifurcation transition from an invariant one-torus (IT1) to an invariant two-torus (IT2) is caused by a Neimark-Saker (NS) bifurcation, also known as a one dimension-higher quasi-periodic Hopf (QH) bifurcation, these by analysing the graph of Lyapunov exponents. We observe that in the mutual coupling case, a complex structure generally-called Arnold’s resonance web phenomenon appears. Numerical simulations are compared to experimental measurements to illustrate the above behaviors. We also note the birth of coherence resonance in the slave oscillator for a certain coupling strength in the case of unidirectional coupling. We also analyze the effect of the coupling strength on the generalized synchronization between the master system and the slave system in the case of unidirectional coupling, through the concept of mutual false nearest neighbors. It allows us to understand when and how closeness in response space implies closeness in driving space.

Largest Lyapunov Exponent Optimization for Control of a Bionic-Hand: A Brain Computer Interface Study.

This paper introduces a brain control bionic-hand, and several methods have been developed for predicting and quantifying the behavior of a non-linear system such as a brain. Non-invasive investigations on the brain were conducted by means of electroencephalograph (EEG) signal oscillations. One of the prominent concepts necessary to understand EEG signals is the chaotic concept named the fractal dimension and the largest Lyapunov exponent (LLE). Specifically, the LLE algorithm called the chaotic quantifier method has been employed to compute the complexity of a system. The LLE helps us to understand how the complexity of the brain changes while making a decision to close and open a fist. The LLE has been used for a long time, but here we optimize the traditional LLE algorithm to attain higher accuracy and precision for controlling a bionic hand. In the current study, the main constant input parameters of the LLE, named the false nearest neighbor and mutual information, are parameterized and then optimized by means of the Water Drop (WD) and Chaotic Tug of War (CTW) optimizers. The optimized LLE is then employed to identify imaginary movement patterns from the EEG signals for control of a bionic hand. The experiment includes 21 subjects for recording imaginary patterns. The results illustrated that the CTW solution achieved a higher average accuracy rate of 72.31% in comparison to the traditional LLE and optimized LLE by using a WD optimizer. The study concluded that the traditional LLE required enhancement using optimization methods. In addition, the CTW approximation method has the potential for more efficient solutions in comparison to the WD method.

Sea-Surface Small Floating Target Recurrence Plots FAC Classification Based on CNN

In this paper, we propose the False-Alarm-Rate-Controllable (FAC) classification of sea clutter Recurrence Plots (RPs) based on Convolutional Neural Networks (CNN), which is shortened as RPs-CNN. Sea clutter data is a non-linear and recursive time series, and RPs provide qualitative analysis of non-linear and recursive dynamic systems. Thus, we construct the RPs datasets to extract the recursive feature of sea clutter. In the RPs datasets, RPs parameters, embedding delay τ and embedding dimension m, are obtained by the average mutual information (AMI) and false nearest neighbor (FNN) algorithms, respectively. In addition, in order to extract the local features of RPs, CNN is applied. CNN makes full use of local features of the datasets, has the advantages of translation invariance and strong generalization ability, and shares the available weights to simplify network structure. We implement a proper LeNet-5 CNN training on the constructed RPs datasets, and verity it by IPIX measured datasets. The experimental results demonstrate that the CNN successfully classifies the RPs of targets and clutter. Moreover, the feasibility of RPs-CNN is verified by seven aspects, i.e., Accuracy, Precision, False Alarm, Miss Rate, Recall, F1-measure and Kappa. In addition, six parameters that may affect the classification performance are also analyzed, including time-series length, embedding delay, embedding dimension, convolutional kernel size, kernel depth, and optimization function. The results indicate that the proposed RPs-CNN method can reach a 92.05% F1-measure and 87.19% Kappa which performs better than other classification methods. Meanwhile, the false alarm rate (FAR) of the RPs-CNN is testified by the FAC classification. Experimental results demonstrate that the proposed RPs-CNN significantly improves the detection probability over other classification detectors in low FAR cases.

Complexity of COVID-19 Dynamics.

With population explosion and globalization, the spread of infectious diseases has been a major concern. In 2019, a newly identified type of Coronavirus caused an outbreak of respiratory illness, popularly known as COVID-19, and became a pandemic. Although enormous efforts have been made to understand the spread of COVID-19, our knowledge of the COVID-19 dynamics still remains limited. The present study employs the concepts of chaos theory to examine the temporal dynamic complexity of COVID-19 around the world. The false nearest neighbor (FNN) method is applied to determine the dimensionality and, hence, the complexity of the COVID-19 dynamics. The methodology involves: (1) reconstruction of a single-variable COVID-19 time series in a multi-dimensional phase space to represent the underlying dynamics; and (2) identification of “false” neighbors in the reconstructed phase space and estimation of the dimension of the COVID-19 series. For implementation, COVID-19 data from 40 countries/regions around the world are studied. Two types of COVID-19 data are analyzed: (1) daily COVID-19 cases; and (2) daily COVID-19 deaths. The results for the 40 countries/regions indicate that: (1) the dynamics of COVID-19 cases exhibit low- to medium-level complexity, with dimensionality in the range 3 to 7; and (2) the dynamics of COVID-19 deaths exhibit complexity anywhere from low to high, with dimensionality ranging from 3 to 13. The results also suggest that the complexity of the dynamics of COVID-19 deaths is greater than or at least equal to that of the dynamics of COVID-19 cases for most (three-fourths) of the countries/regions. These results have important implications for modeling and predicting the spread of COVID-19 (and other infectious diseases), especially in the identification of the appropriate complexity of models.

Study on Multi-Model Soft Sensor Modeling Method and Its Model Optimization for the Fermentation Process of Pichia pastoris.

The problems that the key biomass variables in Pichia pastoris fermentation process are difficult measure in real time; this paper mainly proposes a multi-model soft sensor modeling method based on the piecewise affine (PWA) modeling method, which is optimized by particle swarm optimization (PSO) with an improved compression factor (ICF). Firstly, the false nearest neighbor method was used to determine the order of the PWA model. Secondly, the ICF-PSO algorithm was proposed to cooperatively optimize the number of PWA models and the parameters of each local model. Finally, a least squares support vector machine was adopted to determine the scope of action of each local model. Simulation results show that the proposed ICF-PSO-PWA multi-model soft sensor modeling method accurately approximated the nonlinear features of Pichia pastoris fermentation, and the model prediction accuracy is improved by 4.4884% compared with the weighted least squares vector regression model optimized by PSO.

A Study on Methods for Determining Phase Space Reconstruction Parameters

Abstract Several pairs of algorithms were used to determine the phase space reconstruction parameters to analyze the dynamic characteristics of chaotic time series. The reconstructed phase trajectories were compared with the original phase trajectories of the Lorenz attractor, Rössler attractor, and Chen's attractor to obtain the optimum method for determining the phase space reconstruction parameters with high precision and efficiency. The research results show that the false nearest neighbor method and the complex autocorrelation method provided the best results. The saturated embedding dimension method based on the saturated correlation dimension method is proposed to calculate the time delay. Different time delays are obtained by changing the embedding dimension parameters of the complex autocorrelation method. The optimum time delay occurs at the point where the time delay is stable. The validity of the method is verified by combing the application of the correlation dimension, showing that the proposed method is suitable for the effective determination of the phase space reconstruction parameters.

Deformation Prediction of Foundation Pit Based on Exponential Power Product Model of Improved Algorithm

In order to improve the prediction accuracy of foundation pit deformation, an improved optimization algorithm of supply and demand-exponential power product foundation pit deformation prediction model (ISDO-EPP model) is proposed. Through six standard test functions and three application examples, the optimization ability of the ISDO algorithm is verified, and the optimization results are compared with those of basic supply demand optimization algorithm (SDO), whale optimization algorithm (WOA), grey wolf optimization algorithm (GWO), moth swarm algorithm (MSA), and particle swarm optimization algorithm (PSO). Taking the settlement prediction of three foundation pits as an example, the delay time and embedding dimension of each case are determined by autocorrelation function method and false nearest neighbor method, and input and output vectors are constructed to train and predict each model. The results show that the search ability of the ISDO algorithm is better than that of SDO and other five algorithms, and the ISDO algorithm has better search accuracy, global search ability, and robustness. The absolute values of average relative errors of the ISDO-EPP model for three cases are 0.73%, 3.36%, and 1.33%, respectively, which are better than ISDO-SVM and ISDO-BP models. It shows that the ISDO algorithm can effectively optimize the parameters of the EPP model, and the ISDO-EPP model is feasible and effective for deformation prediction.

Integration of pairwise neighbor topologies and miRNA family and cluster attributes for miRNA-disease association prediction.

Identifying disease-related microRNAs (miRNAs) assists the understanding of disease pathogenesis. Existing research methods integrate multiple kinds of data related to miRNAs and diseases to infer candidate disease-related miRNAs. The attributes of miRNA nodes including their family and cluster belonging information, however, have not been deeply integrated. Besides, the learning of neighbor topology representation of a pair of miRNA and disease is a challenging issue. We present a disease-related miRNA prediction method by encoding and integrating multiple representations of miRNA and disease nodes learnt from the generative and adversarial perspective. We firstly construct a bilayer heterogeneous network of miRNA and disease nodes, and it contains multiple types of connections among these nodes, which reflect neighbor topology of miRNA-disease pairs, and the attributes of miRNA nodes, especially miRNA-related families and clusters. To learn enhanced pairwise neighbor topology, we propose a generative and adversarial model with a convolutional autoencoder-based generator to encode the low-dimensional topological representation of the miRNA-disease pair and multi-layer convolutional neural network-based discriminator to discriminate between the true and false neighbor topology embeddings. Besides, we design a novel feature category-level attention mechanism to learn the various importance of different features for final adaptive fusion and prediction. Comparison results with five miRNA-disease association methods demonstrated the superior performance of our model and technical contributions in terms of area under the receiver operating characteristic curve and area under the precision-recall curve. The results of recall rates confirmed that our model can find more actual miRNA-disease associations among top-ranked candidates. Case studies on three cancers further proved the ability to detect potential candidate miRNAs.

Learn to synchronize, synchronize to learn.

In recent years, the artificial intelligence community has seen a continuous interest in research aimed at investigating dynamical aspects of both training procedures and machine learning models. Of particular interest among recurrent neural networks, we have the Reservoir Computing (RC) paradigm characterized by conceptual simplicity and a fast training scheme. Yet, the guiding principles under which RC operates are only partially understood. In this work, we analyze the role played by Generalized Synchronization (GS) when training a RC to solve a generic task. In particular, we show how GS allows the reservoir to correctly encode the system generating the input signal into its dynamics. We also discuss necessary and sufficient conditions for the learning to be feasible in this approach. Moreover, we explore the role that ergodicity plays in this process, showing how its presence allows the learning outcome to apply to multiple input trajectories. Finally, we show that satisfaction of the GS can be measured by means of the mutual false nearest neighbors index, which makes effective to practitioners theoretical derivations.

A Sparse Dimensionality Reduction Approach Based on False Nearest Neighbors for Nonlinear Fault Detection

As a newly emerging multivariate statistical process monitoring method, non-negative matrix factorization (NMF) and its variants avoid the positive and negative cancellations between the extracted features because of the purely additive combination of non-negative components. Thus, they make a good match with this reality that the negative values of both observations and decomposed components are physically meaningless in many kinds of industrial processes. However, these methods are effective only for linearly separable problems and are not suitable for dealing with nonlinear process monitoring. In this article, the kernel-based method is integrated into the projective NMF (KPNMF) to improve the accuracy of fault detection, and the appropriate multiplicative update method is proven to be convergent. Furthermore, inspired by the false nearest neighbors (FNNs) method, a new dimensionality reduction approach (KPNMF-FNN) is presented to further reduce the original variables for determining the smallest dimension regression vector needed. Compared with the traditional methods, the proposed approach can greatly reduce the time and storage space required on the premise of maintaining a high fault detection rate and low false alarm rate. The experimental results on the Tennessee Eastman benchmark process and the pumping unit system show that the proposed algorithms have excellent performance and can effectively detect faults under the circumstances of retaining only the top 60% of the original variables.

Prediction of surface roughness of various machining processes by a hybrid algorithm including time series analysis, wavelet transform and multi view embedding

The momentary control of manufacturing processes is one of the ways to increase the productivity of production lines. The online measurement of the surface roughness by non-contact methods can be utilized in order to predict the future of surface texture and modify the machining parameters. In the technique proposed at this paper, the surface texture is extracted by combining the 2D surface photography and wavelet approach. Then, by extracting the time delay parameters, the embedding dimension and the false nearest neighbor of the produced surface texture, the future surface roughness is predicted. The results show that this technique can be used in lapping, grinding, turning, and milling processes. Although the maximum roughness error occurred in the surface roughness prediction is 24%, the prediction error is almost constant after Ra = 0.4 μm in different machining processes (about 7%). This study is in line with the development of the proposed method by Pour (2018).

Modeling of Robot's Low-Speed Motion Nonlinear Dynamics Based on Phase Space Reconstruction Neural Network

Abstract In order to improve the accuracy of the robot dynamics model, a low-speed motion nonlinear dynamics modeling method of industrial robot based on phase space reconstruction neural network is proposed. It is confirmed in advance by the largest Lyapunov exponent of joint motor torque data that the robot has chaotic characteristics at low-speed motion. Therefore, experimental data and chaos theory is used to analyze low-speed motion nonlinear dynamics, instead of considering each factor that may cause the robot's nonlinear dynamics separately. The phase space reconstruction parameters of each joint are determined by autocorrelation method and false nearest neighbor method. Through data preprocessing and analysis, some joint position derivatives related to the changing law of torque data are determined. The phase space reconstruction values of these derivatives are chosen as the inputs of neural network. Then the neural network and curve fitting method are combined to compensate for the nonlinear joint torque. Experimental results show that the proposed method can better describe the robot's low-speed motion nonlinear dynamics, and has smaller errors compared with ordinary back propagation (BP) neural network in the case of single joint rotation.

An Autoencoder-Based Deep Learning Approach for Load Identification in Structural Dynamics.

In civil engineering, different machine learning algorithms have been adopted to process the huge amount of data continuously acquired through sensor networks and solve inverse problems. Challenging issues linked to structural health monitoring or load identification are currently related to big data, consisting of structural vibration recordings shaped as a multivariate time series. Any algorithm should therefore allow an effective dimensionality reduction, retaining the informative content of data and inferring correlations within and across the time series. Within this framework, we propose a time series AutoEncoder (AE) employing inception modules and residual learning for the encoding and the decoding parts, and an extremely reduced latent representation specifically tailored to tackle load identification tasks. We discuss the choice of the dimensionality of this latent representation, considering the sources of variability in the recordings and the inverse-forward nature of the AE. To help setting the aforementioned dimensionality, the false nearest neighbor heuristics is also exploited. The reported numerical results, related to shear buildings excited by dynamic loadings, highlight the signal reconstruction capacity of the proposed AE, and the capability to accomplish the load identification task.

Structural classification of proteins based on the computationally efficient recurrence quantification analysis and horizontal visibility graphs.

Protein structural class prediction is one of the most significant problems in bioinformatics, as it has a prominent role in understanding the function and evolution of proteins. Designing a computationally efficient but at the same time accurate prediction method remains a pressing issue, especially for sequences that we cannot obtain a sufficient amount of homologous information from existing protein sequence databases. Several studies demonstrate the potential of utilizing chaos game representation along with time series analysis tools such as recurrence quantification analysis, complex networks, horizontal visibility graphs (HVG) and others. However, the majority of existing works involve a large amount of features and they require an exhaustive, time consuming search of the optimal parameters. To address the aforementioned problems, this work adopts the generalized multidimensional recurrence quantification analysis (GmdRQA) as an efficient tool that enables to process concurrently a multidimensional time series and reduce the number of features. In addition, two data-driven algorithms, namely average mutual information and false nearest neighbors, are utilized to define in a fast yet precise manner the optimal GmdRQA parameters. The classification accuracy is improved by the combination of GmdRQA with the HVG. Experimental evaluation on a real benchmark dataset demonstrates that our methods achieve similar performance with the state-of-the-art but with a smaller computational cost. The code to reproduce all the results is available at https://github.com/aretiz/protein_structure_classification/tree/main. Supplementary data are available at Bioinformatics online.

Effects of Zeros in Phase Space Reconstruction for Small and Large Solar Radiation Data Points during Wet and Dry Seasonal Modeling and Prediction

The effect of zeros in the behaviour of nature system has been a major global concern which have been reported to bias the output of the analysis. This study examines the effect of zeros on small and large solar radiation data points in Nsukka from a nonlinear dynamic perspective. The solar radiation data used were collected from National Research for Space and Development Agency (NARSDA) and covers the period of two years (January 2012–December 2013). The influence of zeros on average mutual information method for delay time (𝜏), False nearest neighbour (FNN) for embedding dimension (𝑚), and phase space reconstruction is investigated by considering two different cases (one hour and five minutes interval for small and large data points on monthly basis respectively). The results reveal that the phase space trajectories of the raw and non-zero small data points for dry and wet seasons show evidence of an attractorin a well-defined region while raw and non-zero large data points have no attractor like shape but regular patterns and well-defined shapes are visible in dry and wet seasons. These imply low-dimensional and deterministic chaotic nature of the underlying dynamics of raw and non-zero data for small and large data points during wet and dry seasons. It is observed that there is little or no significant difference in the phase space reconstruction of raw and non-zero data for both small and large data points due to the low percentage of zeros in the time series data.
 Keywords: mutual information method, Phase space reconstruction, False nearest neighbour, chaotic nature.

Complexity of Body Movements during Sleep in Children with Autism Spectrum Disorder.

Recently, measuring the complexity of body movements during sleep has been proven as an objective biomarker of various psychiatric disorders. Although sleep problems are common in children with autism spectrum disorder (ASD) and might exacerbate ASD symptoms, their objectivity as a biomarker remains to be established. Therefore, details of body movement complexity during sleep as estimated by actigraphy were investigated in typically developing (TD) children and in children with ASD. Several complexity analyses were applied to raw and thresholded data of actigraphy from 17 TD children and 17 children with ASD. Determinism, irregularity and unpredictability, and long-range temporal correlation were examined respectively using the false nearest neighbor (FNN) algorithm, information-theoretic analyses, and detrended fluctuation analysis (DFA). Although the FNN algorithm did not reveal determinism in body movements, surrogate analyses identified the influence of nonlinear processes on the irregularity and long-range temporal correlation of body movements. Additionally, the irregularity and unpredictability of body movements measured by expanded sample entropy were significantly lower in ASD than in TD children up to two hours after sleep onset and at approximately six hours after sleep onset. This difference was found especially for the high-irregularity period. Through this study, we characterized details of the complexity of body movements during sleep and demonstrated the group difference of body movement complexity across TD children and children with ASD. Complexity analyses of body movements during sleep have provided valuable insights into sleep profiles. Body movement complexity might be useful as a biomarker for ASD.

Icing Load and Risk Forecasting for Power Transmission Line Based on Multi-scale Time Series Phase-Space Reconstruction and Regression

Accurately forecasting the icing load on overhead power transmission lines is an important issue to ensure the security and reliability of the power grid. A multi-scale time series phase-space reconstruction and regression model for icing load prediction is proposed in this paper to treat the non-stationary, nonlinear, and intermittent volatility of power line icing load data. Those is motivated by the traditional icing load prediction models having many disadvantages in the forecasting accuracy, as well as the casualness of the parameters selected. Firstly, the icing load data are decomposed into a multi-scale time series of intrinsic model function (IMF) components with stability by using the ensemble empirical mode decomposition (EEMD), which can reduce the interactions between different types of feature information. Secondly, phase-space reconstruction (PSR) theory is applied using the mutual information and the false nearest neighbor to determine the optimal delay time and embedding dimension of each IMF component. Thirdly, considering the characteristics of each IMF component, different kernel functions and optimization parameters are selected to establish the prediction model based support vector regression (SVR). Finally, according to the load prediction results, fuzzy reasoning method was used to determine the risk status of transmission line towers in this paper. Upon experimentally evaluating the validity of the model using related transmission lines of the Yunnan Power Grid, it is shown that this method could predict the real-time icing load on overhead power lines, obtaining better regression performance. This model could be used on power transmission and distribution systems for deicing and maintenance decisions.