Time Series Prediction Algorithm Research Articles

Optimization of quantum-inspired neural network using memetic algorithm for function approximation and chaotic time series prediction

Abstract Heuristic and deterministic optimization methods are extensively applied for the training of artificial neural networks. Both of these methods have their own advantages and disadvantages. Heuristic stochastic optimization methods like genetic algorithm perform global search, but they suffer from the problem of slow convergence rate near global optimum. On the other hand deterministic methods like gradient descent exhibit a fast convergence rate around global optimum but may get stuck in a local optimum. Motivated by these problems, a hybrid learning algorithm combining genetic algorithm (GA) with gradient descent (GD), called HGAGD, is proposed in this paper. The new algorithm combines the global exploration ability of GA with the accurate local exploitation ability of GD to achieve a faster convergence and also a better accuracy of final solution. The HGAGD is then employed as a new training method to optimize the parameters of a quantum-inspired neural network (QINN) for two different applications. Firstly, two benchmark functions are chosen to demonstrate the potential of the proposed QINN with the HGAGD algorithm in dealing with function approximation problems. Next, the performance of the proposed method in forecasting Mackey–Glass time series and Lorenz attractor is studied. The results of these studies show the superiority of the introduced approach over other published approaches.

A Quantum based Evolutionary Algorithm for Stock Index and Bitcoin Price Forecasting

Quantum computing has emerged as a new dimension with various applications in different fields like robotic, cryptography, uncertainty modeling etc. On the other hand, nature inspired techniques are playing vital role in solving complex problems through evolutionary approach. While evolutionary approaches are good to solve stochastic problems in unbounded search space, predicting uncertain and ambiguous problems in real life is of immense importance. With improved forecasting accuracy many unforeseen events can be managed well. In this paper a novel algorithm for Fuzzy Time Series (FTS) prediction by using Quantum concepts is proposed in this paper. Quantum Evolutionary Algorithm (QEA) is used along with fuzzy logic for prediction of time series data. QEA is applied on interval lengths for finding out optimized lengths of intervals producing best forecasting accuracy. The algorithm is applied for forecasting Taiwan Futures Exchange (TIAFEX) index as well as for Bitcoin crypto currency time series data as a new approach. Model results were compared with many preceding algorithms.

Communication Network Time Series Prediction Algorithm Based on Big Data Method

In order to solve the low prediction accuracy and prediction rate of traditional time series prediction algorithm in the analysis of massive data, according to the characteristics of communication network performance index, time series prediction algorithm based on the big data method is proposed. In addition, based on the traditional three components of time series feature extraction, the concept of emergency component is introduced, and the outlier detection and processing as well as prediction analysis are carried out on the basis of the extraction results. The results show that the algorithm based on big data, through the analysis, fitting, modelling, and prediction of massive data, smaller granularity decomposition of the time series value is conducted, which significantly improved the credibility and accuracy of prediction. Based on the above findings, it is summarized that the time series prediction algorithm has good performance.

Nonlinear Time Series Prediction Performance Using Constrained Motion Particle Swarm Optimization

Constrained Motion Particle Swarm Optimization (CMPSO) is a general framework for optimizing Support Vector Regression (SVR) free parameters for nonlinear time series regression and prediction. CMPSO uses Particle Swarm Optimization (PSO) to determine the SVR free parameters. However, CMPSO attempts to fuse the PSO and SVR algorithms by constraining the SVR Lagrange multipliers such that every PSO epoch yields a candidate solution that meets the SVR constraint criteria. The fusion of these two algorithms provides a numerical efficiency advantage since an SVR Quadratic Program (QP) solver is not necessary for every particle at every epoch. This reduces the search space of the overall optimization problem. It has been demonstrated that CMPSO provides similar (and in some cases superior) performance to other contemporary time series prediction algorithms for nonlinear time series benchmarks such as Mackey-Glass data. This paper details the CMPSO algorithm framework and tests its performance against other SVR time series prediction algorithms and data including the European Network on Intelligent Technologies for Smart Adaptive Systems (EUNITE) competition data and the Competition on Artificial Time Series (CATS) competition data.

Based on SARIMA-BP hybrid model and SSVM model of international crude oil price prediction research

We propose two hybrid prediction models for the international crude oil price: SARIMA-BP hybrid model; and SSVM model. The SARIMA-BP hybrid model combines seasonality analysis and autoregressive integrated moving average with back propagation neural network model. The SSVM model combines seasonality analysis with support vector machines. New York Mercantile Exchange (NYMEX) crude oil's monthly closing price, which ranges from January 2002 to April 2016, is selected as the experimental data sets. Experimental results are compared among the SARIMA-BP hybrid model, SSVM model and single SARIMA model. Empirical analysis shows that the SSVM model has highest prediction accuracy, and the single SARIMA model has lowest prediction accuracy. Thus, the SSVM model displays a better performance in oil price prediction. Further, the SSVM model predicts NYMEX crude oil's closing price will approach 50 dollars per barrel in May 2016. References Z. B. Zhou, X. C. Dong, Analysis about the seasonality of china's crude oil import based on x-12-arima, Energy 42 (42) (2012) 281–288. P. F. Pai, C. S. Lin, A hybrid arima and support vector machines model in stock price forecasting, Omega 33 (6) (2005) 497–505. T. Kriechbaumer, A. Angus, D. Parsons, M. R. Casado, An improved waveletcarima approach for forecasting metal prices, Resources Policy 39 (1) (2014) 32–41. T. Koutroumanidis, K. Ioannou, G. Arabatzis, Predicting fuelwood prices in greece with the use of arima models, artificial neural networks and a hybrid arimacann model, Energy Policy 37 (9) (2009) 3627–3634. H. Mohammadi, L. Su, International evidence on crude oil price dynamics: Applications of arima-garch models, Energy Economics 32 (5) (2010) 1001–1008. H. Chiroma, S. Abdulkareem, T. Herawan, Evolutionary neural network model for west texas intermediate crude oil price prediction, Applied Energy 142 (2015) 266–273. L. A. Laboissiere, R. A. S. Fernandes, G. G. Lage, Maximum and minimum stock price forecasting of brazilian power distribution companies based on artificial neural networks, Applied Soft Computing 35 (2015) 66–74. H. Al-Askar, A. J. Hussain, D. Al-Jumeily, N. Radi, Regularized dynamic self organized neural network inspired by the immune algorithm for financial time series prediction, Lecture Notes in Computer Science 8590 (2014) 56–62. V. Sampathkumar, M. H. Santhi, J. Vanjinathan, Forecasting the land price using statistical and neural network software, Procedia Computer Science 57 (2015) 112–121. J. P. Maran, B. Priya, Modeling of ultrasound assisted intensification of biodiesel production from neem ( azadirachta indica ) oil using response surface methodology and artificial neural network, Fuel 143 (2015) 262–267. D. Singhal, K. S. Swarup, Electricity price forecasting using artificial neural networks, International Journal of Electrical Power and Energy Systems 33 (44) (2008) 111–118. X. Y. Zeng, L. Shu, G. M. Huang, J. Jiang, Triangular fuzzy series forecasting based on grey model and neural network, Applied Mathematical Modelling 40 (3) (2015) 1717–1727. L. Fan, S. Pan, Z. Li, H. Li, An ica-based support vector regression scheme for forecasting crude oil prices, Technological Forecasting and Social Change. F. Wen, J. Xiao, H. E. Zhifang, X. Gong, Stock price prediction based on ssa and svm , Procedia Computer Science 31 (2014) 625–631. X. Guo, D. C. Li, A. Zhang, Improved support vector machine oil price forecast model based on genetic algorithm optimization parameters, Aasri Procedia 1 (4) (2012) 525–530. J. L. Zhang, Y. J. Zhang, L. Zhang, A novel hybrid method for crude oil price forecasting, Energy Economics 49 (2015) 649–659. C. C. Chang, C. J. Lin, Libsvm: A library for support vector machines, Acm Transactions on Intelligent Systems \(and\) Technology 2 (3, article 27) (2007) 389–396. S. E. Yang, L. Huang, Financial crisis warning model based on bp neural network, Systems Engineering-theory and Practice 25 (1) (2005) 12–19.

An autonomic prediction suite for cloud resource provisioning

One of the challenges of cloud computing is effective resource management due to its auto-scaling feature. Prediction techniques have been proposed for cloud computing to improve cloud resource management. This paper proposes an autonomic prediction suite to improve the prediction accuracy of the auto-scaling system in the cloud computing environment. Towards this end, this paper proposes that the prediction accuracy of the predictive auto-scaling systems will increase if an appropriate time-series prediction algorithm based on the incoming workload pattern is selected. To test the proposition, a comprehensive theoretical investigation is provided on different risk minimization principles and their effects on the accuracy of the time-series prediction techniques in the cloud environment. In addition, experiments are conducted to empirically validate the theoretical assessment of the hypothesis. Based on the theoretical and the experimental results, this paper designs a self-adaptive prediction suite. The proposed suite can automatically choose the most suitable prediction algorithm based on the incoming workload pattern.

The improved grey model based on particle swarm optimization algorithm for time series prediction

Grey theory is one of the most common methods for solving uncertain problems using limited data and poor information, due to its high performance in time series prediction. However, the inappropriate background value and initial value are the main factors affecting prediction accuracy of the Grey Model GM(1,1). An improved grey model based on particle swarm optimization algorithm named PGM(1,1) is proposed for time series prediction in this paper. The development coefficient of the grey model is calculated by PGM(1,1) based on particle swarm optimization, targeting at minimizing the average relative errors between the restored value and real value of the model to avoid the problem caused by background value optimization. In addition, the initial value of the Grey Model GM(1,1) is optimized and a sliding window is introduced to improve both precision and adaptability. Finally, three data sets, featuring increasing trend, decreasing trend, and wide fluctuations, are used in the experiments, showing that the proposed method achieves better prediction accuracy.

Wavelet neural network with improved genetic algorithm for traffic flow time series prediction

In the traditional wavelet neural network (WNN) prediction model, the parameter optimization is performed using a unidirectional gradient descent algorithm, which has the problems of slow convergence and local optimum. To improve the predication accuracy of short-term traffic flow, a predication model based on clustering search strategy improved genetic algorithm (IGA) and WNN (IGA-WNN) is proposed. The IGA is used to optimize the initial connection weights, translation factor and scaling factor of WNN. The algorithm is applied to the short-term traffic flow of empirical research. The experimental results show that IGA-WNN model has a higher predication accuracy and a better nonlinear fitting ability compared with the traditional WNN and GA-WNN prediction models.

Multiple measures-based chaotic time series for traffic flow prediction based on Bayesian theory

Considering the chaotic characteristics of traffic flow, this study proposes a Bayesian theory-based multiple measures chaotic time series prediction algorithm. In particular, a time series of three traffic measures, i.e., speed, occupancy, and flow, obtained from different sources is used to reconstruct the phase space using the phase space reconstruction theory. Then, data from the multiple sources are combined using Bayesian estimation theory to identify the chaotic characteristics of traffic flow. In addition, a radial basis function (RBF) neural network is designed to predict the traffic flow. Compared to the consideration of a single source, results from numerical experiments demonstrate the improved effectiveness of the proposed multi-measure method in terms of accuracy and timeliness for the short-term traffic flow prediction.

Online ARIMA Algorithms for Time Series Prediction

Autoregressive integrated moving average (ARIMA) is one of the most popular linear models for time series forecasting due to its nice statistical properties and great flexibility. However, its parameters are estimated in a batch manner and its noise terms are often assumed to be strictly bounded, which restricts its applications and makes it inefficient for handling large-scale real data. In this paper, we propose online learning algorithms for estimating ARIMA models under relaxed assumptions on the noise terms, which is suitable to a wider range of applications and enjoys high computational efficiency. The idea of our ARIMA method is to reformulate the ARIMA model into a task of full information online optimization (without random noise terms). As a consequence, we can online estimation of the parameters in an efficient and scalable way. Furthermore, we analyze regret bounds of the proposed algorithms, which guarantee that our online ARIMA model is provably as good as the best ARIMA model in hindsight. Finally, our encouraging experimental results further validate the effectiveness and robustness of our method.

Selected an Stacking ELMs for Time Series Prediction

Extreme learning machine (ELM) has several interesting and significant features. In this paper, a novel pruned Stacking ELMs (PS-ELMs) algorithm for time series prediction (TSP) is proposed. It employs ELM as the level-0 algorithm to train several models for Stacking. And our previously proposed reduce-error pruning for TSP (ReTSP)-Trend pruning technique is used to solve the problem that the level-0 learners might make many correlated error predictions. ReTSP-Trend refers to an evaluation measure for reduce-error pruning for TSP (ReTSP), which takes into account the time series trend and the forecasting error direction. What's more, ELM and simple averaging are used to generate the level-1 model. With the development of PS-ELMs, firstly, those essential advantages of ELM will be naturally inherited. Secondly, those specific defects of ELM are ameliorated to some extent, with the help of ensemble pruning paradigm. Thirdly, ensemble pruning is employed to raise the robustness and accuracy of time series forecasting, making up for the shortages of the existing research. Fourthly, our previously proposed pruning measure ReTSP-Trend is employed in PS-ELMs, which indeed guarantees that the remaining predictor which supplements the subensemble the most will be selected. And finally, the development of PS-ELMs will promote our investigation to the popular ensemble technique of Stacked Generalization. The experimental results on four benchmark financial time series datasets verified the validity of the proposed PS-ELMs algorithm.

Regularized dynamic self-organized neural network inspired by the immune algorithm for financial time series prediction

This paper presents a novel type of recurrent neural network, the regularized dynamic self-organized neural network inspired by the immune algorithm. The regularization technique is used with the dynamic self-organized multilayer perceptrons network that is inspired by the immune algorithm. The regularization has been addressed to improve the generalization and to solve the over-fitting problem. In this work, the average values of 30 simulations generated from 10 financial time series are examined. The results of the proposed network were compared with the standard dynamic self-organized multilayer perceptrons network inspired by the immune algorithm, the regularized multilayer neural networks and the regularized self-organized neural network inspired by the immune algorithm. The simulation results indicated that the proposed network showed average improvement using the annualized return for all signals of 0.491, 8.1899 and 1.0072 in comparison to the benchmarked networks, respectively.

A novel single multiplicative neuron model trained by an improved glowworm swarm optimization algorithm for time series prediction

To better predict time series, in this paper the single multiplicative recurrent neuron (SMRN) is constructed by adding feedforward and feedback links at the nodes of the original single multiplicative neuron (SMN). Glowworm swarm optimization (GSO) algorithm as a method for training the parameters of various kinds of neural network models gets easily into locally optimal traps during optimization process and its movement stability is also poor because of no memory about search history. To overcome the aforementioned disadvantages, firstly the linearly declining inertia weight is incorporated into the location update formula of standard GSO (LWGSO). After that in order to further enhance the robustness capability, differential evolution (DE) algorithm is introduced into LWGSO forming LWGSODE. Standard unimodal and multi-modal static test functions in high dimensions have been used to test its properties. The statistically experimental results show that the proposed LWGSODE approach performs much better than basic GSO whatever in terms of solutions precision, robustness or convergence speed. Moreover, the function optimization results are also competitive when compared with other state-of-the-art methods in the literature. Finally, the LWGSODE algorithm is used to train SMRN for time series prediction, and approximation results have improved significantly. All the results obtained reveal the novel SMRN model combined with the proposed LWGSODE algorithm provides a promising means to approximate nonlinear series in the future.

A big data inspired chaotic solution for fuzzy feedback linearization model in cyber-physical systems

Cyber-Physical System (CPS) that integrates computational and physical capabilities has emerged as a promising topic. It interacts with physical world and humans through ad hoc communications. In contemporary, with the fast development of CPS theory and applications, CPS generates a large volume of data, which may lead its control into a chaotic status. Hence, there is a pressing demand to solve the chaotic status CPS control. To this end, the chaotic time series prediction algorithm is employed to resolve the chaotic status featured by a fuzzy feedback linearization model. Modeling the CPS under big data without taking chaotic features into account may lead to unexpected results. This is because chaotic CPS is dramatically sensitive to small disturbances or minor changes of initiators. This paper developed a CPS model in light of fuzzy feedback linearization. Further, the chaotic time prediction algorithm is applied to solve the chaotic control problem in CPS. The developed algorithm takes both tracking control problem and synchronization control problem into account. The numerical results suggest that the developed method is feasible and efficient in tracking control and synchronization of two chaotic CPS.

Fast Prediction for Sparse Time Series: Demand Forecast of EV Charging Stations for Cell Phone Applications

This paper proposes a new cellphone application algorithm which has been implemented for the prediction of energy consumption at electric vehicle (EV) charging stations at the University of California, Los Angeles (UCLA). For this interactive user application, the total time for accessing the database, processing the data, and making the prediction needs to be within a few seconds. We first analyze three relatively fast machine learning-based time series prediction algorithms and find that the nearest neighbor (NN) algorithm (k NN with k = 1) shows better accuracy. Considering the sparseness of the time series of the charging records, we then discuss the new algorithm based on the new proposed time-weighted dot product (TWDP) dissimilarity measure to improve the accuracy and processing time. Two applications have been designed on top of the proposed prediction algorithm: one predicts the expected available energy at the outlet and the other one predicts the expected charging finishing time. The total time, including accessing the database, data processing, and prediction is approximately 1 s for both applications. The granularity of the prediction is 1 h and the horizon is 24 h; data have been collected from 20 EV charging outlets.

Sequential behavior prediction based on hybrid similarity and cross-user activity transfer

The proliferation of smart phones has opened up new kinds of data to model human behavior and predict future activity but this prediction can be tempered by the relative sparsity of data. In this paper, we integrate a time-dependent instance transfer mechanism, driven by a hybrid similarity measure, into learning and predicting human behavior. In particular, transfer component analysis (TCA) is utilized for domain adaptation from different data types to overcome data sparsity. The hybrid user similarity measure is developed based on three different characteristics: eigen-behavior, longest common behavior (LCB), and daily common behavior (DCB). Extensive comparisons are made against state-of-the-art time series prediction algorithms using the Nokia Mobile Data Challenge (MDC) dataset and the MIT Reality Mining dataset. We compare the prediction performance given (i) no additional data, (ii) only data from identical behavior from other users, and (iii) data from any type of behavior from other users. Experimental results show that our proposed algorithm significantly improves the performance of behavior prediction.

Predicting financial time series data using artificial immune system-inspired neural networks

This paper investigates a set of approaches for the prediction of noisy time series data; specifically, the prediction of financial signals. A novel dynamic self-organised multilayer neural network based on the immune algorithm for financial time series prediction is presented, combining the properties of both recurrent and self-organised neural networks. In an attempt to overcome inherent stability and convergence problems, the network is derived to ensure that it reaches a unique equilibrium state. The accuracy of the comparative evaluation is enhanced in terms of profit earning; empirical testing used in this work includes normalised mean square error NMSE to evaluate forecast fitness and also evaluates predictions against financial metrics to assess profit generation. Extensive simulations for multi-step prediction in stationary and non-stationary time series were performed. The resulting forecast made by the proposed network shows substantial profits on financial historical signals when compared to various solely neural network approaches. These simulations suggest that dynamic immunology-based self-organised neural networks have a better ability to capture the chaotic movement in financial signals.

Hermite orthogonal basis neural network based on improved teaching-learning-based optimization algorithm for chaotic time series prediction

Chaos phenomenon which exists widely in nature and society affects people's production and life. It has great important significance to find out the regularity of chaotic time series from a chaotic system. Since chaotic system has extremely complex dynamic characteristics and unpredictability, and chaotic time series prediction through traditional methods has low prediction precision, slow convergence speed and complex model structure, a prediction model about Hermite orthogonal basis neural network based on improved teaching-learning-based optimization algorithm is proposed. Firstly, according to the chaotic time series, autocorrelation method and Cao method are used to determine the best delay time and the minimum embedding dimension respectively, then a phase space is reconstructed to obtain the refactoring delay time vector. Secondly, on the basis of phase space reconstruction and best square approximation theory, combined with the neural network topology, a prediction model about Hermite orthogonal basis neural network with excitation functions based on the Hermite orthogonal basis functions is put forward. Thirdly, in order to optimize the parameters of the prediction model, an improved teaching-learning-based optimization algorithm is proposed, where a feedback stage is introduced at the end of the learning stage based on the teaching-learning-based optimization algorithm. Finally, the parameter optimization problem is transformed into a function optimization problem in the multidimensional space, then the improved teaching-learning-based optimization algorithm is used for parameter optimization of the prediction model so as to establish it and analyze it. Lorenz and Liu chaotic systems are taken as models respectively, then the chaotic time series which will be used as simulation object is produced by the fourth order Runge-Kutta method. The comparison experiments with other prediction models are conducted on single-step and multi-step prediction for the chaotic time series. The simulation results and numerical analysis show that compared with radial basis function neural network, echo state network, least square support vector machine prediction model and Hermite orthogonal basis neural network based on teaching-learning-based optimization algorithm, the proposed prediction model has the mean absolute error and root mean square error reduced significantly, has a decision coefficient close to 1, meanwhile, has a mean modeling time shortened greatly. So the proposed prediction model can improve the prediction precision, accelerate the convergence speed and simplify the model structure, thus the prediction model is effective and feasible, which makes it promoted and applied easily.

Non-classical Algorithm for Time Series Prediction of the Range of Economic Phenomena With Regard to the Interaction of Financial Market Indicators

The aim of the article is to present non-clasical copyrighted algorithm for prediction of time series, presenting macroeconomic indicators and stock market indices. The algorithm is based on artificial neural networks and multi-resolution analysis (the algorithm is based on Daubechies wavelet). However, the main feature of the algorithm, which gives a good quality of the forecasts, is all included in the series analysis division into, a few partial under-series and prediction dependence on a number of other economic series. The algorithm used for the prediction, is copyrighted algorithm, labeled M.H-D in this article. Application of the algorithm was performed on a series presenting WIG 20. The forecast of WIG 20 was conditional on trading the Dow Jones, DAX, Nikkei, Hang Seng, taking into account the sliding time window. As an example application of copyrighted model, the forecast of WIG 20 for a period of two years, one year, six month was appointed. An empirical example is described. It shows that the proposed model can predict index with the scale of two years, one year, a half year and other intervals. Precision of prediction is satisfactory. An average absolute percentage error of each forecast was: 0.0099%—for two-year forecasts WIG 20; 0.0552%—for the annual forecast WIG 20; and 0.1788%—for the six-month forecasts WIG 20.

Adaptive step-size modified fractional least mean square algorithm for chaotic time series prediction

This paper presents an adaptive step-size modified fractional least mean square (AMFLMS) algorithm to deal with a nonlinear time series prediction. Here we incorporate adaptive gain parameters in the weight adaptation equation of the original MFLMS algorithm and also introduce a mechanism to adjust the order of the fractional derivative adaptively through a gradient-based approach. This approach permits an interesting achievement towards the performance of the filter in terms of handling nonlinear problems and it achieves less computational burden by avoiding the manual selection of adjustable parameters. We call this new algorithm the AMFLMS algorithm. The predictive performance for the nonlinear chaotic Mackey Glass and Lorenz time series was observed and evaluated using the classical LMS, Kernel LMS, MFLMS, and the AMFLMS filters. The simulation results for the Mackey glass time series, both without and with noise, confirm an improvement in terms of mean square error for the proposed algorithm. Its performance is also validated through the prediction of complex Lorenz series.