Chia-Nan Ko

Short-term load forecasting using lifting scheme and ARIMA models

Research highlights? Short-term load forecasting is achieved using a lifting scheme and autoregressive integrated moving average (ARIMA) models. ? Lifting scheme is embedded into the ARIMA models to enhance forecasting accuracy. ? The Coeflet 12 wavelet is factored into lifting scheme steps. ? Apply the proposed algorithm to different practical load data types. ? Forecasting performance of the proposed approach is superior to that of the back-propagation network (BPN) algorithm and traditional ARIMA models. Short-term load forecasting is achieved using a lifting scheme and autoregressive integrated moving average (ARIMA) models. The lifting scheme is a general and flexible approach for constructing bi-orthogonal wavelets that are usually in the spatial domain. The lifting scheme is embedded into the ARIMA models to enhance forecasting accuracy. Based on wavelet multi-revolution analysis (MRA) results, the lifting scheme decomposes the original load series into different sub-series at different revolution levels, which display the different frequency characteristic of a load. The sub-series are then forecast using properly fitted ARIMA models. Finally, forecasting results at different levels are reconstructed to generate an original load prediction by the inverse lifting scheme. In this study, the Coeflet 12 wavelet is factored into lifting scheme steps. The proposed algorithm was tested by applying it to different practical load data types from the Taipower Company in 2007 for one-day-ahead load forecasting. Simulation results indicate that the forecasting performance of the proposed approach is superior to that of the back-propagation network (BPN) algorithm and traditional ARIMA models.

A PSO method with nonlinear time-varying evolution based on neural network for design of optimal harmonic filters

A particle swarm optimization method with nonlinear time-varying evolution based on neural network (PSO-NTVENN) is proposed to design large-scale passive harmonic filters (PHF) under abundant harmonic current sources. The goal is to minimize the cost of the filters, the filters loss, and the total harmonic distortion of currents and voltages at each bus, simultaneously. In the PSO-NTVENN method, parameters are determined by using a sequential neural network approximation. Meanwhile, based on the concept of multi-objective optimization, how to define the fitness function of the PSO to include different performance criteria is also discussed. To show the feasibility of the proposed method, illustrative examples of designing optimal passive harmonic filters for a chemical plant are presented.

A PSO Method With Nonlinear Time-Varying Evolution for Optimal Design of Harmonic Filters

A particle swarm optimization method with nonlinear time-varying evolution (PSO-NTVE) is employed in the planning of large-scale passive harmonic filters for a multibus system under abundant harmonic current sources. The objective is to minimize the cost of the filter, the filters loss, the total harmonic distortion of currents and voltages at each bus simultaneously. In the PSO-NTVE method, parameters are determined by using matrix experiments with an orthogonal array, in which a minimal number of experiments would have an effect that approximates the full factorial experiments. To compare the performance of the proposed PSO-NTVE method with existing ones, five well-known benchmarks are used and the results reveal its superiority over the others. This PSO-NTVE method is then applied to design optimal harmonic filters in a steel plant, where both ac and dc arc furnaces are used and a static var compensator (SVC) is installed. From the results of the illustrative examples, the feasibility of the PSO-NTVE to design an optimal passive harmonic filter of a multibus system is verified.

Identification of MIMO systems using radial basis function networks with hybrid learning algorithm

When a radial basis function network (RBFN) is used for identification of a nonlinear multi-input multi-output (MIMO) system, the number of hidden layer nodes, the initial parameters of the kernel, and the initial weights of the network must be determined first. For this purpose, a systematic way that integrates the support vector regression (SVR) and the least squares regression (LSR) is proposed to construct the initial structure of the RBFN. The first step of the proposed method is to determine the number of hidden layer nodes and the initial parameters of the kernel by the SVR method. Then the weights of the RBFN are determined by solving a simple minimization problem based on the concept of LSR. After initialization, an annealing robust learning algorithm (ARLA) is then applied to train the RBFN. With the proposed initialization approach, one can find that the designed RBFN has few hidden layer nodes while maintaining good performance. To show the feasibility and superiority of the annealing robust radial basis function networks (ARRBFNs) for identification of MIMO systems, several illustrative examples are included.

ARFNNs with SVR for prediction of chaotic time series with outliers

This paper demonstrates an approach to predict the chaotic time series with outliers using annealing robust fuzzy neural networks (ARFNNs). A combination model that merges support vector regression (SVR), radial basis function networks (RBFNs) and simplified fuzzy inference system is used. The SVR has the good performances to determine the number of rules in the simplified fuzzy inference system and initial weights for the fuzzy neural networks (FNNs). Based on these initial structures, and then annealing robust learning algorithm (ARLA) can be used effectively to overcome outliers and adjust the parameters of structures. Simulation results show the superiority of the proposed method with different SVR for training and prediction of chaotic time series with outliers.

Genetic Auto-Tuning and Rule Reduction of Fuzzy PID Controllers

This paper presents a novel method for parameter auto-tuning of a fuzzy proportional-integral-derivative (PID) controller. Different from PID controllers with fixed gains, the fuzzy PID controller is expressed in terms of fuzzy rules, in which the input variables are the error signals and their derivatives, while the output variables are the PID gains. In this manner, the PID gains are adaptive and the fuzzy PID controller has more flexibility and capability than the traditional ones. When tuning the fuzzy PID gains, a genetic-algorithm-based method is proposed, in which the centers and the widths of the Gaussian membership functions, the fuzzy control rules corresponding to every possible combination of input linguistic variables, and the PID gains are chosen as parameters to be determined. In encoding the parameters into corresponding chromosomes, a mixed-coding technique is adopted. To expedite the convergence speed of the evolutionary process, the concept of enlarged sampling space and ranking mechanism are used. To show the effectiveness and validity of the designed fuzzy PID controller, a multivariable seesaw system is used for illustration.

Radial basis function networks with hybrid learning for system identification with outliers

This paper demonstrates that radial basis function networks (RBFNs) with support vector regression (SVR) and annealing robust learning algorithm (ARLA) can be used effectively for the identification of the nonlinear dynamic systems with outliers. When the RBFNs are used for the identification of the nonlinear dynamic system, the number of hidden nodes, the initial parameters of the kernel, and the initial weights of the network must be determined first, a SVR approach is proposed to solve the initial problem of RBFNs. That is, the SVR uses the quadratic programming optimization to determine the initial structure of the RBFNs. Besides, the new cost function for the system identification with outliers is also proposed. That is, the proposed annealing robust radial basis function networks (ARRBFNs) are trained by the ARLA, which uses the annealing concept in the cost function of the robust back-propagation learning algorithm, can overcome the error measurement caused by the outliers. Simulation results show the superiority of the proposed method with different SVR.

ARRBFNs with SVR for prediction of chaotic time series with outliers

In this article, annealing robust radial basis function networks (ARRBFNs), which consist of a radial basis function network and a support vector regression (SVR), and an annealing robust learning algorithm (ARLA) are proposed for the prediction of chaotic time series with outliers. In order to overcome the initial structural problems of the proposed neural networks, the SVR is utilized to determine the number of hidden nodes, the initial parameters of the kernel, and the initial weights for the proposed ARRBFNs. Then the ARLA that can conquer the outliers is applied to tune the parameters of the kernel and the weights in the proposed ARRBFNs under the initial structure with SVR. The simulation results of Mackey-Glass time series show that the proposed approach with different SVRs can cope with outliers and give a fast learning speed. The results of the simulation are also given to demonstrate the validity of proposed method for chaotic time series with outliers.

A Nonlinear programming method for time-optimal control of an omni-directional mobile robot

The time-optimal control problem of a three-wheeled omni-directional mobile robot is studied in this paper. In contrast to traditional methods, in which the Pontryagins Minimum Principle is usually used, an iterative procedure is proposed to transform the time-optimal problem into a nonlinear programming (NLP) one. In the formulated NLP problem, the number of control steps is fixed initially and the sampling period is treated as a variable to be minimized in the optimization process. An upper limit on the sampling period is set in advance considering the accuracy of discretization. If the value of the sampling period is larger than the upper limit, then an update of the number of control steps is needed. To generate initial feasible solutions of the NLP problem, a genetic algorithm is adopted. Since different initial feasible solutions can be generated, the optimization process can be started from different points to find the optimal solution. In this manner, one can find a time-optimal movement of the omni-directional mobile robot between two configurations. To show the feasibility of the proposed method, simulation results are included for illustration.

Integration of support vector regression and annealing dynamical learning algorithm for MIMO system identification

This paper presents a robust approach to identify multi-input multi-output (MIMO) systems. Integrating support vector regression (SVR) and annealing dynamical learning algorithm (ADLA), the proposed method is adopted to optimize a radial basis function network (RBFN) for identification of MIMO systems. In the system identification, first, SVR is adopted to determine the number of hidden layer nodes, the initial structure of the RBFN. After initialization, ADLA with nonlinear time-varying learning rate is then applied to train the RBFN. In the ADLA, the determination of the learning rate would be an important work for the trade-off between stability and speed of convergence. A computationally efficient optimization method, particle swarm optimization (PSO) method, is adopted to simultaneously find optimal learning rates. Due to the advantages of SVR and ADLA (SVR-ADLA), the proposed RBFN (SVR-ADLA-RBFN) has good performance for MIMO system identification. Two examples are illustrated to show the feasibility and superiority of the proposed SVR-ADLA-RBFNs for identification of MIMO systems. Simulation results are provided to demonstrate the effectiveness of the proposed algorithm.