【作者】 周滢露；

【导师】 王官洁；

【作者基本信息】 重庆大学 ， 电气工程， 2003， 硕士

【摘要】 在变电站电压无功的综合控制中，有载调压变压器和补偿电容器是重要的研究对象至于之一。论文主要研究变电站的电压/无功控制决策问题，为了确定一天24小时内合适的有载调压变压器分接头位置和并联电容器投切状态，提出一种基于人工神经网络的无功负荷预测和进化规划优化决策相结合的变电站电压和无功的综合控制决策。首先，建立径向基函数神经网络（RBFN）的短期负荷预测的模型，利用RBF神经网络的非线性逼近能力，预测出一天24小时的整点平均负荷值，为了得到合理的径向基函数中心参数，在本文中引入了减聚类算法，用来指导聚类学习，将具有一定相似度的样本归类为一组，此后通过一个聚类自动终止判据控制聚类的个数，这样即可确定比较合适的RBF径向基函数的参数，又可提高网络映射精度。因此提高了学习性能，具有较好的预测精度。然后，建立变电站电压无功控制的数学模型，考虑电压的调压要求和无功功率平衡，计及变压器变比和并联补偿电容的上下限约束，变压器分接头和电容器允许的日调节次数的限制，以电压偏差的平方和最小为目标函数。由于预先预测出无功负荷，提前了解了无功功率变化的趋势，可以有助于判定低压母线电压变化是由无功负荷变化引起还是由高压侧电压变化引起，从而适时决定是调有载调压变压器分接头还是投切电容器，以此避免了盲目和不充分的调节，实现在保证无功基本平衡和电压合格率的前提下，减少有载调压变压器分接头的调节和并联电容器组的投切次数。在对优化的具体实现过程中，由于进化规划着眼于整个整体的进化，对于所求解的优化问题无可微性要求，采用随机搜索技术，能以较大的概率求解全局最优解的特点，针对电压无功控制模型是一个多限制、多目标、非线性、离散的优化控制问题，因此应用进化规划算法进行模型的求解。更多还原

【Abstract】 This paper mainly discusses a control method of substation voltage and reactive power .In order to get suitable decision for one day 24 hours tap-transformer’s step switch and shunt capacitor switch, an approach of substation voltage and reactive power control on the basis of the combination of Artificial Neural Network (ANN) reactive power forecasting and evolutionary programming optimal decision-making is put forward. Firstly, Radial Basis Function Network(RBFN) is applied in short-term load forecasting ,it is abtained one-day 24 hours average load values based on nonlinear approximation capability of RBF neural network. In this paper ,subtractive clustering method is introduced for proper RBF centers, thus direct training the network, control the number of clustering by using automatical end-criterion, RBFN can obtain both the parameters of the neurons and the number of the hidden neurons, also can improve network inflection accuracy. The RBF network has a better performance, and better forecasting accuracy .Then mathematical model of substation Voltage/Var control is constructed, the squares minimization of Voltage differences as target, also considering requirement of Voltage and power balance, taking it into consideration that the magnitude constraint of transformer ratio and compensating capacitor, also that constraint of operation times of one day transformer tap and capacitor switch. Because of forecasting reactive load at first, it can detect the change of voltage at low-voltage bus from the change of reactive load or the change of voltage at high-voltage bus, then it can decide that adjusting transformer tap or capacitor switch, and avoid blindly and deficient adjusting. On the condition the reactive power is balanced and voltage qualified, it can realize the switching times of loaded taps and capacitors being efficiently decreased. Aimed at multiple-limit, multiple-object, non-linear, discrete of Voltage/Var optimization and control, on account of whole evolution of evolutionary programming, no demand for differentiability of optimal function, and random search, it can obtain global optimum with mayor probability, this paper solve optimal function with evolutionary programming.更多还原