No abstract

Primary fan system plays an important role in the operation of a power plant. However, due to the complicated working conditions of the primary fan and the strong coupling characteristics of multi-state variables, it is necessary to carry out feature engineering before using multivariate state estimation technique (MSET). In addition, no-linear operator should be designed to make sure matrix being invertible. This paper proposes an Auto-encoder based model to automatically construct a normal state memory matrix through unsupervised learning of neural networks. It reduces human intervention as well as the difficulty in giving a suitable non-linear operator design. This model is applied to the early warning of primary fan failure in a power plant in eastern Shandong Province.

The randomness and volatility of wind power generation are the main reasons restricting the capacity of power grid to absorb wind power. Thermal power units are the main force of power grid frequency modulation. The speed of load adjustment rate determines their response ability to power grid load. Based on the analysis of the characteristics of wind power generation, the nonlinear multiscale decomposition of the automatic power generation control (AGC) load command is carried out. Combined with the different load regulation rates of different types of units, the rational allocation of unit combinations can effectively ensure the power grid load regulation capacity and compensate the random disturbance of wind power.

Due to the increasing scale of new energy and the increasingly close interconnection of regional power grids, higher requirements are put forward for the adjustment of active power under power grid faults. Automatic generation control(AGC) is an important function of the energy management system(EMS), it controls the output of the frequency modulation units, and keeps the system in a safe and stable operating state. By analyzing the basic principle and assessment index of AGC control, combined with AGC control process, the emergency active power control strategy of thermal power unit is proposed. By cooperating with AGC of superior dispatching, the effective suppression of power grid fluctuation is realized, which provides effective protection for grid security control.

A multi-layer LSTM (Long short-term memory) model is proposed for condenser vacuum degree prediction of power plants. Firstly, Min-max normalization is used to pre-process the input data. Then, the model proposes the two-layer LSTM architecture to identify the time series pattern effectively. ADAM（Adaptive moment）optimizer is selected to find the optimum parameters for the model during training. Under the proposed forecasting framework, experiments illustrates that the two-layer LSTM model can give a more accurate forecast to the condenser vacuum degree compared with other simple RNN (Recurrent Neural Network) and one-layer LSTM model.