Studies On Generalized Load Modeling Considering Wind Power Integration

Power system digital simulation is the main tool for designing, planning and operation. The accuracy of the simulation results has a significant impact on the safety, reliablility, economic operation of power system. As the basis of digital simulation, the accuracy of load models largely affects simulation results. However, due to the complexity, dispersion and randomness of the load, getting the accurate load models has always been a problem in the power research areas. At present, with the increasing scarcity of resources, the severity of environmental issues, renewable energy generations are considered to be an effective means to solve above issues, from this point, there is a rapid development of wind power. Load containing traditional load and small capacity generator is called generalized load. Wind power’s integration into the distribution network load nodes changes the case that load could simply consume power and makes load nodes possible to send power down to the grid. The neighboring wind farms connected to load nodes of multiple distribution networks locate in the same wind zone, and they have a strong correlation of wind speed, which results in relevant wind farm outputs. So, for the neighboring generalized load nodes with electrical connection, load modeling needs to take the relevance of wind farm outputs and the correlation of the electrical connection. The uncertainty and intermittence characteristic of wind power outputs along with the time-varying characteristics of load, make the generalized load nodes more varying. The large integration of wind power into the power system poses threat to the safety of power system operation and quantitative analysis on risk assessment index is required. If generalized load related factors are not taken into account and the risk analysis work is done with incomplete information, the results obtained from the analysis tend to be "optimistic" or "subjective" Inaccurate assessment misleads power system operation scheduling control. All those will bring new challenge to the load modeling.Large-scale wind power integration to the power system brings the problem that the power flow generalized load nodes turns to be uncertain, so whether the node is source or load is no longer clear. New demand of generalized load modeling is proposed. How to take full account of the volatility of random variable characteristics and the local relevance to get the accurate load models has become an urgent problem. For the uncertainty and relevance of generalized load, the resulting security risk assessment problem, some work have done from the static load characteristics of the generalized model with a long time scales, the work follows:(1) This paper presents a generalized steady state load model structure based on radial basis function (RBF) neural network interval model. The previous research choose Back Propagation (BP) neural network as interval model in generalized load modeling considered a long time scales, which has a local minimum problem and large structural parameters. Further more, it is difficult to determine the model structure. In order to improve the general applicability and robustness of the interval model, the paper proposed RBF neural network as interval model. RBF neural network output layer weight connection is just a simple structure and has a good value for the the global approach. Based on these advantages, RBF neural network is used to learn and extract each interval node segmentation features to build node characteristic unified model. Simulation results show that the proposed model structure can be accurately modeled and has less model parameters, good fitting effect, stable structure with universal applicability of interval model. Generalized steady state load unified model structure proposed in this paper, can analyse uncertainties by probability distribution scenario, which promotes structural model development of generalized load modeling.(2) This paper proposes a kind of generalized load united probability modeling method with probability information considering spatial correlation of power nodes. Due to related factors involved, the proposed method can create accurate model taking into account neighboring nodes power flow when the node characteristics are under description. First, according to power flow direction, the generalized load bus node characteristics are divided into source characteristics and load characteristics; Secondly, the measured active power is used to represent characteristics and segment, the range of the segments is determined adaptively and the probability distribution is got by probability statistics; Then, the RBF neural network is used to abstract the node characteristics considering relevant characteristic parameters between segments divided by different nodes based on the spatial correction method. Simulation results show that the united probability model fully takes into account the impact of nearby generalized nodes uncertainty. It has good fitting effect compared with uncorrelated modeling method and contains more comprehensive information in modeling process.(3) This paper presents a classification characteristic method of generalized load steady-state characteristics. This method can effectively identify generalized load node in the source characteristics and load characteristics, which is generally applicable for generalized load time-varying classification. The classification characteristic method is builded as follows:Firstly, the characteristics of generalized load bus active power is analyzed, the minimum time period of active power fluctuation sequence is determined by fluctuation intensity theory applied in dynamics area. Then, the daily time period characteristic vector is composed by fluctuation sequence and statistic within variable time periods. This classification characteristic represents the generalized load’s changes in amplitude and in flow direction. Making use of affinity propagation clustering method, the clustering number and center is adjusted adaptively. The simulation shows that the classification characteristic method builded by this paper distinguishes node day period power characteristics and load characteristics more clearly and applicabily. Accurate generalized load model with probability information is given after accurate clustering by the generalized characteristics.(4) This paper presents a risk analysis approach based generalized load models. Firstly, generalized load united probability model is used to actual power flow simulation system. Then, specific indicators based on risk analysis are used to risk accessment. The simulation shows that risk analysis can traverse the entire refining system space based on united probability model, which pointing out the high-risk scenarios set. The risk analysis results compared to single node modeling shows that the risk analysis results based on united probability model are larger than those based on single node modeling in the range of load characteristics, smaller in the range of source characteristics and similar in the range between load characteristics and source characteristics. So, it is helpful to recognize system operation risk status comprehensively, otherwise, cognitive biases arise, leading to economic and security risks. In connection with generalized load time-varying clustering and comprehensive, accurate generalized load model is given after accurate clustering by the generalized characteristics. It is applied to the risk simulation with wind power integration into load nodes. The risk results indicate the high-risk cluster category and the node power range scenes.