Neural network usage for solving the problem of short-term local forecast of outdoor temperature

To increase the efficiency of control automation of the heating system an adaptive model of short-term local forecast of outdoor temperature is designed and optimized. At the operation of the given model and long-term forecasting there is a high risk of occurrence and accumulation of an error.To avoid this negative effect a special adaptive mechanism of the model made by neural network is constructed in accordance with Rosenblatt perceptron scheme. As a learning method the back propagation algorithm is used as the best method for continual improvement and network learning capacity in length of time. Based on the constructed model many numerical experiments are carried out to predict the temperature during the year. The analysis and comparison of the results are performed, the influence of various characteristics of neural network on the quality of the received forecast is determined.