Probabilistic Power Flow Calculation Based On Stacked Denoising Auto-Encoders

In recent years,due to the expansion of the scale of the power grid and the integration of intermittent renewable energy system containing photovoltaic energy and wind power,the complexity and uncertainty in power system is increasingly obvious.In order to guarantee the economic operation and security and stability analysis of power system better,higher demands are put forward for the accuracy,speed and cost of probabilistic load flow calculation,which can fully consider the stochastic factors of power system.In the existing probabilistic power flow method,analytical method and approximation method have small calculation,but the approximate treatment must be carried out.The Monte-Carlo simulation(MCS)method can handle a variety of complex conditions without simplification,but the amount of calculation is large.Generally,improvements for MCS method can be divided into improvement of sampling and acceleration of power flow calculations.Both of them have many research results,but it is still difficult to give consideration to the accuracy and speed of calculation.Another way of thinking is to improve hardware,such as using supercomputers or parallel clusters.Although it can greatly improve computing speed under the condition of guaranteeing accuracy,the economic cost and technical cost of implementation are relatively high.As a contrast,regarding accuracy,deep neural network can represent an approximation of any function to an arbitrary degree of accuracy.In terms of speed,deep neural network does not need iterative calculation of non-linear equations,and is usually high-parallelizable.On the cost of computing,it can be implemented in the normal PC.Aiming at the shortcomings of the existing MCS method,this paper introduces the idea and model of deep neural network into the field of power flow/ probabilistic power flow,and improves the MCS method from the point of improving the power flow calculation method.This article mainly completes the following two tasks:(1)Stacked Denoising AutoEncoders(SDAE)-based power flow solving model with corresponding improved training methods is proposed.Based on the approximation and solvability judgment capability of SDAE in nonlinear power flow equations,SDAE-based power flow solving model is constructed in this paper.Then,cross-entropy loss function,mini-batch gradient descent algorithm and momentum learning rate are implemented in unsupervised pre-training and supervised fine-tuning to efficiently optimize the parameters of proposed model.After properly trained,SDAE-based power flow solving model is able to judge the solvability of power system and non-iteratively calculate power flow.Simulations are implemented on modified IEEE 39-bus system and modified IEEE 118-bus system integrated with renewable energy,which verifies the effectiveness,accuracy and robustnessof proposed SDAE-based power flow solving model.(2)A calculation method of probabilistic power flow based on SDAE and combined with MCS is proposed.First get the training sample of SDAE-based power flow solving model,normalizing the training samples and determining the hyper-parameters of SDAE.Then optimize the parameters of SDAE using two stage training method.Finally extract working samples using MCS method and stored as a matrix form in a given format.Input the matrix into trained SDAE-based power flow solving model,and the power flow solvability and the calculation results of power flow values in the corresponding matrix form can be obtained.Simulations are implemented on modified IEEE 118-bus system integrated with renewable energy,which verifies the accuracy,effectiveness and adaptability of proposed SDAE-based probabilistic power flow calculation method.