Research On Superheated Steam Temperature System Identification Based On Multi-model And Field Data

Superheated steam temperature control is an important step in thermal control system of the power plant, a better control effect needs an effective analysis to superheated steam temperature system. Analyze thermal objects need a precise mathematical model, asked the very high-precision model, however, superheated steam temperature is a big inertia, pure delay,non-linear and time variability of the complex system. In this paper, a multi-model modeling method for superheated steam temperature were established, experiments show that this method can effectively solve the problems which encountered in the establishment of superheated steam temperature system.For the case of the relative gap twist that might occur before and after the data standardization, selecting mean absolute deviation for data standardization, so that each of the data before and after the standardization of the relative gap is not distorted. Principal component analysis and multivariate process monitoring are used to get the auxiliary variables which have an influence on superheated steam temperature system, and as input variables in data model. Sliding window steady data extraction algorithm is used to extract quasi-steady state operation data from a large number of continuous operation data, static data is used to establish model.Spectral clustering algorithm which based on the optimization of sample is proposed and applied to cluster a large number of operation data. Through sample optimization, the number of samples is simplified, while the remaining samples also represent the operating conditions characteristic of the whole unit,spectral clustering algorithm can be used to cluster simplified samples, otherwise, the number of samples is too big which cause excessive spectral clustering operation. According to unit capacity and cluster results, selecting three cluster centers mean establishing three local models. Using the least squares support vector machine to build local superheated steam temperature data model, the data is correspond to each local model. By using the maximum absolute error, relative rms error, mean error, rms error and other evaluation indexs analyse train and test results we can find that the LSSVM data model established in this paper has a good learning ability and generalization ability.Considering the system instability caused by hard handoff, in this paper, using soft handoff coordination mechanism to coordinate the various local model. Corresponding weight of the local model is calculated by local model network algorithm. The output of multi-model is weight sum of local models' output and the weight of the local model. Gaussian function is selected as coordination function of local model network, local model network based on particle swarm optimization is proposed.In order to get the goal of optimum global forecast, particle swarm optimization is proposed to optimize the width of each gaussian function which calculated by the most neighborhood heuristic algorithm. It considers the degree of dispersion of the data in a cluster, at the same time, degree of dispersion of each cluster is also taken into account.Multi-model system built by this way has good global characteristics, for the static multi-model system has a problem of poor dynamics, using a multi-modeling method based on ARIMA establish the superheated steam temperature model. The experiments show that dynamic correction improves effect of superheated steam temperature model, it has some practical value.