Pseudo-Worst-Case Forecast With Neural Networks In Low Voltage Grids

In low-voltage smart grids, data are received from advanced measurement systems (smart meters). These data play a fundamental role in monitoring and control. However, as these data are being received in wide snapshot sample times, real-time monitoring is a huge challenge. The next time-step data forecasting seems to be a promising solution to improve the monitoring, observability, and controllability of low voltage smart grids. The existing fluctuations in current and voltage profiles due to the stochastic behavior of each household lead to imprecise short-term forecasting results. However, for the preventive control method, precise forecasting is not necessary. In this paper, the neural network method is used to forecast the next time-step data. Input and output data for the neural network are based on three different strategies. According to the simulation results, the accuracy of forecasting such volatile data is low. Thus, the neural network method is combined with the pseudo worst-case forecast method. In our previous publication, this method was introduced to forecast worst-case scenarios instead of exact values. The pseudo-worst-case forecast method is a heuristic algorithm that can be modified based on operator knowledge and grid behavior. To obtain an almost stand-alone method able to adjust itself without operator support, the pseudo-worst-case forecast method is improved by the neural network method. The paper presents this combined heuristic method and validates it through simulations done in MATLAB software using data from real low-voltage grids located in Germany. The proposed method's success rate in forecasting the probable worst-case scenario is more than 98%.