Abstract:The theories of system identification have been highly elaborated so as to achieve the true system. This paper much discuses regarding the stochastic processes along with the divergent of whether or not the system has zero-mean under scenario of either white and coloured noise. The mathematical foundations, including mean, variance, covariance, optimal parameters, along with some modified scenarios among them, are presented in detail from various system along with some basic idea behind them. The families of auto-regressive (1) and (2) are compared both mathematical and simulation in order to obtain the best design approaching the true system, Moreover, the least-square algorithm is used to examine the effectiveness of some number of iteration along with "batch" theorem
What problem does this paper attempt to address?
The paper primarily explores the autoregressive model in system identification and its performance under different noise conditions, validated through theoretical analysis and numerical simulations.
Specifically, the objectives of the paper include:
1. **Mathematical Foundation**: Detailed explanation of basic concepts such as stochastic processes, mean, variance, covariance, and discussion on the selection of optimal parameters.
2. **Autoregressive Model**: Comparison of the mathematical representation and simulation results of autoregressive models (AR(1) and AR(2)) to determine the best design method.
3. **Least Squares Algorithm**: Using the least squares algorithm to evaluate the impact of iteration times on model performance and introducing the "batch processing theorem" to further optimize the model.
4. **System Stability**: Discussion of the stability characteristics of the system using the Lyapunov method.
5. **Impact of Noise Types**: Study of the model's performance under conditions of zero mean and non-zero mean, as well as white noise and colored noise.
The paper defines the system model and analyzes three scenarios in detail:
- **Independent and Zero Mean**: In this scenario, the input process is independent and zero mean. The paper provides the expected value, variance, and covariance function of the model, and based on these analyses, finds the optimal parameters.
- **Independent and Non-Zero Mean**: When the input process becomes non-zero mean, although the system's variance remains unchanged, the expected value is affected. The paper also provides the calculation formula for the optimal parameters.
- **Parameter Assumption Changes**: Changing the model parameters while keeping the input process zero mean to observe the changes in model performance.
In addition, the paper uses the least squares algorithm for numerical simulations, evaluating the convergence and stability of the model by adjusting the number of iterations \(N\) and the batch size \(\kappa\). The simulation results show that as \(N\) and \(\kappa\) increase, the model's performance in approximating the real system improves.
In summary, this paper systematically studies the performance of autoregressive models under different noise conditions through theoretical analysis and numerical simulations, and proposes an effective least squares estimation method to optimize the model.