| Supervisory Information System (SIS) has a rapid application in the process of informationalization of electric power enterprises and stimulates the interest of researchers at same time. The online performance assessment and operation optimization of power unit are the key issuses. The thermo-economic state equations-based system analysis method is developed in this thesis. The reliability of online performance model is investigated basing on the operation data and two artificial intelligence methods are imported to cope with the new problems from the established applications.The relationship between matrix equation and system structure is deduced in the paper. Under the same assumption with theses traditional methods, the state space expression of linear time invariant systems is imported. The steam-water distribution equation is revealed as the stable form of the state equation. The concept of thermal disturbance vector and its construction rules are proposed by regarding the auxiliary steam/water as the disturbances imposed on the main system. Then, the uniform formula is obtained for the calculation of system output, which simplifies the system analysis greatly. State space expression abstracts the thermal system to a general system. The study discovered the nature of matrix based analysis, and the essential of the local quantitative analysis of thermal system is its structure analysis.The reliability of the online performance model is investigated basing on the abundant operation data. Statistic analysis proceeds on the available data satisfying the performance test approximately. The uncertainty is 2% for a continuous working condition and 6% for the conditions under similar boundary parameters (load and inlet temperature of circle water). Thus, the possible scope for energy saving through optimal operation is 4%, which is encouraging for further research. However, the unexpected uncertainty of a continuous working condition brings the new challenge. The important inputs of online model are worked out by global sensitivity analysis, which can be referenced for system maintenance.The CBR is used for the operation advice. Those key issues are settled, such as, the selection of characteristic process and parameters. A decision factor is proposed for ranking operation priority basing on three working conditions (current condition, similar condition and optimal condition). The unsteady working condition will go to the similar condition without adjustment, so those parameters whose value in similar condition lapse from the value in the optimal condition. The new character of optimal operation value by CBR is its integrity and feasibility compared with current methods.Due to the complicated structure, strong coupling among sub-systems and changeable properties of matter, the performance of power unit maybe still takes on a sub-health condition with fault-free devices. This needs the deep diagnosis and decision. A probabilistic casual reasoning method is proposed for the diagnosis of optimal operation parameters. A causal model (Bayesian networks) is constructed using system-level properties of power unit. The model parameterized by operational data can be used for parameters deviation diagnosis on the actual power unit directly. A relative entropy based operation support method is proposed. Its essential is to search an optimal combination of parameters under the current constraints by probabilistic reasoning.At last, a machine learning-based intelligence decision support system prototype is design for the operation optimization. The multi-model fusion techniques are discussed. Based on the characteristic of the two AI methods used, an idea of decision support alarm is proposed.
|
PDF Full Text Request