This paper presents a fault early warning approach of coal mills based on the Thermodynamic Law and data mining. The Thermodynamic Law is used to describe the working characteristics of coal mills and to determine the multi-parameter vector that characterize the operating state of the coal mill. Data mining technology is applied to analysis the interrelationships among elements of the multi-parameter vector. Then the abnormal boundaries of parameters are calculated based on the distribution of parameters under different working conditions according to the Pauta criterion. Finally, the fault early warning model is implemented combining the abnormal boundaries and the confidence algorithm that can detect the working status of coal mills. Two actual numerical examples are used to illustrate the proposed method is capable of estimating the abnormality of coal mills before the fault happens.

SummaryThe superheated steam temperature system of the thermal power plant has the characteristics of large inertia, nonlinearity, and strong time variation, which make it difficult to be controlled. To address these problems, this paper proposes a generalized predictive control algorithm with an adaptive forgetting factor. First, based on a fuzzy algorithm and a recursive least squares algorithm, the controlled object's model can be quickly and accurately obtained with the adaptive forgetting factor in real time. It overcomes the nonlinear and time‐varying problems of the controlled object in the control progress. Meanwhile, it also solves the problem of data saturation and the weight assignment of the “new and old” data during online identification. Second, an adaptive generalized predictive controller algorithm has been developed with the controlled object. It solves the large inertia problem of the controlled object. Finally, through establishing simulation model of the superheated steam temperature system and simulating, the results show that the proposed method has better control performance, antidisturbance ability, adaptability, and robustness. Moreover, it has a certain reference significance for the design of a practical control system.

The continuous development of the power Internet of Things (IOT) has enabled power market participants to obtain a large amount of data. Simultaneously, the power IOT has an increasing demand for power load and electricity price forecasting; Since the forecasting of electricity load and electricity price is a single task, and the model calculation accuracy is not high, this brings great challenges to the accurate forecasting of electricity load and electricity price. In this paper, two power load and electricity price forecasting models via multi-task deep learning are established perform high-precision joint forecasting of power load and electricity price Experimental results demonstrate that the prediction results of the proposed deep learning models are superior to the other compared approaches in terms of the main task and the auxiliary task, and show superior prediction performance, verifying the practicability and superiority of the power load and electricity price multi-task forecasting model.