System identification

About: System identification is a research topic. Over the lifetime, 21291 publications have been published within this topic receiving 439142 citations.

Model identification of signal transduction networks from data using a state regulator problem

Abstract: Advances in molecular biology provide an opportunity to develop detailed models of biological processes that can be used to obtain an integrated understanding of the system. However, development of useful models from the available knowledge of the system and experimental observations still remains a daunting task. In this work, a model identification strategy for complex biological networks is proposed. The approach includes a state regulator problem (SRP) that provides estimates of all the component concentrations and the reaction rates of the network using the available measurements. The full set of the estimates is utilised for model parameter identification for the network of known topology. An a priori model complexity test that indicates the feasibility of performance of the proposed algorithm is developed. Fisher information matrix (FIM) theory is used to address model identifiability issues. Two signalling pathway case studies, the caspase function in apoptosis and the MAP kinase cascade system, are considered. The MAP kinase cascade, with measurements restricted to protein complex concentrations, fails the a priori test and the SRP estimates are poor as expected. The apoptosis network structure used in this work has moderate complexity and is suitable for application of the proposed tools. Using a measurement set of seven protein concentrations, accurate estimates for all unknowns are obtained. Furthermore, the effects of measurement sampling frequency and quality of information in the measurement set on the performance of the identified model are described.

Model-free subspace-based LQG-design

Abstract: When only input/output data of an unknown system are available, the classical way to design a linear quadratic Gaussian controller for that system mainly consists of three separate parts. First a system identification step is performed to find the system parameters. With these parameters a Kalman filter is designed to find an estimate of the state of the system. Finally, this state is then used in an LQ-controller. In the literature these three steps are hardly ever considered as one joint problem. Based on techniques from the field of sub-space system identification the present paper gives a new, much more direct method to calculate a finite-horizon LQG-controller. The three steps of the LQG-controller design, i.e. system identification, Kalman filter and LQ-control design are replaced by a QR- and a SV-decomposition. The equivalence between the new subspace-based approach and the classical approach is proven.

Optimizing the echo state network with a binary particle swarm optimization algorithm

Abstract: The echo state network (ESN) is a novel and powerful method for the temporal processing of recurrent neural networks. It has tremendous potential for solving a variety of problems, especially real-valued, time-series modeling tasks. However, its complicated topologies and random reservoirs are difficult to implement in practice. For instance, the reservoir must be large enough to capture all data features given that the reservoir is generated randomly. To reduce network complexity and to improve generalization ability, we present a novel optimized ESN (O-ESN) based on binary particle swarm optimization (BPSO). Because the optimization of output weights connection structures is a feature selection problem and PSO has been used as a promising method for feature selection problems, BPSO is employed to determine the optimal connection structures for output weights in the O-ESN. First, we establish and train an ESN with sufficient internal units using training data. The connection structure of output weights, i.e., connection or disconnection, is then optimized through BPSO with validation data. Finally, the performance of the O-ESN is evaluated through test data. This performance is demonstrated in three different types of problems, namely, a system identification and two time-series benchmark tasks. Results show that the O-ESN outperforms the classical feature selection method, least angle regression (LAR) method in that its architecture is simpler than that of LAR.

Forecasting of ionospheric critical frequency using neural networks

Abstract: Multilayer perception type neural networks (NN) are employed for forecasting ionospheric critical frequency (foF2) one hour in advance. The nonlinear black-box modeling approach in system identification is used. The main contributions: 1. A flexible and easily accessible training database capable of handling extensive physical data is prepared, 2. Novel NN design and experimentation software is developed, 3. A training strategy is adopted in order to significantly enhance the generalization or extrapolation ability of NNs, 4. A method is developed for determining the relative significances (RS) of NN inputs in terms of mapping capability.