Short-term Load Combination Forecast Model Based On Similar Days And Intelligent Algorithms

Short-term load forecasting (STLF) is an important reference for electricity sectors to make power generation plans and maintenance plans, and it can help reduce the unnecessary rotating reserve capacity, lower cost, gain economic benefits and maintain grid security and stability. And it will also promote to build a more open electricity market. With the development of smart distribution grid, as a basis of grid loss analysis, network reconfiguration, state assessments, STLF plays an important role in enhancing the services of distribution sectors. Therefore, the research of seeking higher forecasting accuracy and more suitable distribution network STLF methods is of great value.By studying of the daily load characteristics of the distribution network, this paper emphasized the STLF models should take weather factors into account based on different day types. It described the pretreatment of historical load data, and it put forward a pretreatment method of weather factors to embed them into STLF model.Aiming at few load data sources, gathering omissions, and inadequate sample problems, and combining the different needs and requirements of STLF from lots of power sectors, this paper has proposed a hierarchical STLF model based on topology and big data. The STLF objects and hierarchies are low-voltage distribute-electricity transformer districts (low-voltage district) and lOkV distribution power lines (10kV line). After the analysis of low-voltage district's daily load characters, this paper presented a classification prediction model by the district's weather sensitivity. For weather-sensitive districts, the forecasting is implemented by choosing similar days first and then using least squares support vector machine (LSSVM), which can solve small sample, nonlinear, high dimension and local minimum value problems. And for weather-insensitive districts, extracting different frequency load series by empirical mode decomposition (EMD) and then combing LSSVM. This classification prediction approach has increased the flexibility and applicability, and thus improved the prediction accuracy.Taking the low-voltage district as the first or lower hierarchy, and the 10kV line as the second or higher hierarchy, through their topology, this paper illustrated the hierarchical STLF model. The 10kV line's forecasting result is obtained by composing its low-voltage districts'forecasting load, among which by setting coefficients for each district based on the topology of line and districts. And the key to determine the coefficients is combining particle swarm optimization (PSO) and a training process similar to neural network. Further, it has established a peak load forecasting optimization model for lOkV line. The hierarchical STLF model can apply variety data samples as possible, and weaken adversely affect caused by a single source data. What's more, it can reflect the lower layer's load characters. The practical example has shown that the hierarchy model offers a higher average prediction accuracy for 10kV line and its peak load forecasting accuracy is significantly improved based on the optimization model.Finally, this paper developed a corresponding STLF software, and explained the graphical user interface (GUI), and its functions and operations in detail.